And actually just today Google announced Gemini Advanced, which I’m now allowed to say I was an early tester for 🧐. I would say it’s definitely on par with GPT-4, and more useful in certain areas because of the search capabilities. Though for coding I still prefer GPT-4. Also I think Gemini is better for generating certain photorealistic images, but it doesn’t quite follow instructions as well as DALLE3 I think. However Gemini Advanced is paid and uses their “ultra” model, and the free one is the “pro” model which isn’t as good.
Chatgpt is great in analyzing a data. Once while I was doing externship, I used chatgpt to analyzing and put the data to a table. To my suprise, chatgpt did it well.
@@czerskip I disagree, I am in IT for a living and use Bard a LOT for coding and technical stuff, things that the non paid Chat GPT was horrible for because it was outdated. I also found the natural language understanding was slightly better in Bard. That's the nice thing about having options, though, we can use what works best for us 🙂
I'm autistic and it lets me ask questions that I'm afraid would come off as stupid when I ask my colleagues. I'm so incredibly thankful for this tool. Fact checking is a must though especially for critical info.
I mean calling 3.5 garbage is definitely an extreme take. It would be better to say in comparison that 4 is a significant improvement to the previous instance. But in the general consensus of things 3.5 is by far bounds better than the predecessors before it.
3.5 is garbage. seriously its complete brain dead when giving answers. it just frustrating working with its limits. sure its better than its predecessors but the bar is low
@@false-zd5ujThis video is not fair, you need to do side by side comparison before saying today's Chatgpt 3.5 is stupid or not. The 3.5 has been upgraded to 2023 version, and I can literally do 90% of the task this video shown on 3.5 perfectly. It can even generate Third Normal Form based on a data table.
As someone who uses the latest model daily for coding and everything else under the sun and additionally 3.5-turbo for commercial apps using the API, I can say that 3.5-turbo with fine tuned prompts and functions is far and away more cost effective and gives great results ie NOT garbage.
@@tausiftaha12- 3.5 is good enough to get you curious and interested. Pay for it for one month. You'll learn that the accuracy of 4 is leagues better than 3.5. I use 3.5 to compare LLMs that I run locally. For real questions, I use 4 and I barely touch search engines anymore.
Uses for everyday Joe, who doesn't code... 1. I took some pictures of my lawn and it identified some of the weeds and the proper care. It guided me step-by-step and all my lawn is nice and green now. It even noticed a shaded area because of a tree and suggested specifics for that part.2. It literally organized an entire trip through Italy. It set up my entire itinerary down to recommendations of restaurants to visit. 3. You can also ask it to give you the quickest of rundowdowns on any book or any movie, or any anything, so you can instantly become at least somewhat knowledgeable on any topic. It all explains to you as brief or as long and in detail as you need it. 4. You can upload spreadsheets and ask it to make whatever changes you need. You can even ask it if there's an easier way to format it. 5. You can also upload PDF files. And then ask it questions about it. Such as stuff on the small print that you might not have time to go over all of it yourself. Great video. By the way, and I hope this helps somebody.
I uploaded a photo of a family member taken in the mid-19th century. I had spent years on and off trying to identify the person by dating her dress and hairstyle. This was difficult as it depended on whether she was wearing the most up-to-date fashion or not. ChatGPT dated the image to within 10 years, and with that information, I knew that it was a specific ancestor and not her daughter. I also uploaded an image of an antique writing slope I had bought and it accurately dated it. I later hired an antique restorer and he confirmed the date was correct.
It's useful for technical instructions, it used to take me hours to dumb down instructions for non techy people, trying to anticipate what people don't know is difficult. Now I literally just brain dump what I need into GPT-4 and out pops easy to understand instructions that anybody can understand. It has save so much time and more importantly support call from users who didn't understand the instructions. It's also so useful for summarizing tech docs, quite often companies write documentation with inexperienced users in mind, so you are presented with massive walls of text, and it's depressing knowing the information you need is just a single sentence, that's no longer a problem with GPT-4 it just knows how to pull out what you need.
*Pro Tip:* In case you are worried about the whole hallucination thing… When you click your profile icon _(at the bottom right),_ you can choose _"Custom Instructions"_ There you can tell GPT to respond in a certain way. In my response field I entered: _"Always end the result with a confidence level."_ This way, you know how sure GPT is about its response, after each response during a chat. When this level is *_LOW,_* you could add more context to the chat.
You also can call its bluff by asking it to show data or sources. I tell people to treat AI as summer high-school intern...who is at the bottom 1/3rd of their class.
For several months, I have consistently included the directive "Be brief and concise" within those guidelines, yet, frustratingly, the responses continue to be lengthy and filled with unnecessary details in each conversation. Daily, I find myself having to remind it repeatedly, between 10 to 20 times, to adhere to the principle of brevity and conciseness, otherwise it begins to unnecessarily elongate its responses. Despite clear instructions to keep things succinct, there seems to be a perpetual tendency to veer into verbose territory, requiring constant reminders to stay on track. The pattern is wearisome, as I am compelled to intervene frequently to enforce the simplicity and directness that should naturally follow from my initial instructions. This ongoing issue suggests a disconnect in understanding the essence of concise communication, resulting in a cyclical pattern of correction. Each day unfolds with similar challenges, as I strive to curtail the expansive nature of the responses which should, by now, align more closely with the straightforward guidelines provided. --the above generated by gpt-4, haha
Wonderful video. I like the "out of the box" thinking mate. I've been using ChatGPT 3.5 (and 4.0 through Copilot) for several months now and it helped me (up to a point) in many ways: 1- Grammar check 2- Rephrasing according to certain styles 3- Writing codes for specific requirements 4- Writing poem 5- Describing certain codes or commands (as you have mentioned) 6- Solving some math problems (I wasn't satisfied as I found many mistakes. However, it was about advanced math and very specific) 7- Asking suggestions about what tools or software shall I use for a certain task 8- Asking about various topics in science and technology 9- Asking about pros and cons to compare between stuff, methods, tools ... etc 10- Providing suitable citations for specific sentences 11- Asking about rules and regulations regarding a certain topic in a certain country Among other ways. I do feel that I'm just scratching the surface and there are lots of other ways to use it. In general, this tool is amazing and can help greatly if we know how to use it.
One you forgot is working on a resume. I'm terrible at formatting my projects into a good format for a resume so I just described to chat GPT what I've worked on and it gives me a really good output to put on my resume. I actually made my own resume advisor agent that is able to go through my resume section by section or the entire thing at once. It can do suggestions or just output its own version based on your description and it can simulate interview questions based on your resume to get you prepared for an interview. Once I had a really good solid generic resume I would then put that in as input and then put the job that I'm applying for and have it modify my resume to fit the specific job I'm applying for and then generate a unique cover letter to match that position. This has allowed me to apply for many different jobs really quickly with unique customized resumes for that specific position.
@@anakaliaeastwood It's also really good at figuring out which projects to work on like if you tell it what kind of jobs you want to get and ask it for ideas of some projects you could work on to get the skills necessary it'll give you some good ideas.
I only use GPT 3.5, and if you ask it the same question enough times, it will give you an answer to a different question, but never answer the question you ask it. Still, sometimes it gives me the answer to a more urgently wanted question
When steam engine was first put to industrial use people were first dismissing it, and than when it was obvious machines are here to stay, they rioted in fear of machinery taking all their jobs (The Luddites). Now people are dismissing AI... we just never learn, do we?
@@ilijas3041 Estás siendo muy ignorante! Los luditas no sé enojaron por las máquinas, se enojaron porqué los empresarios usaron las máquinas para darles peores sueldos y explotarlos laboralmente. La protesta de los luditas fue en parte destruir máquinas del empresario que los esclavizaba.
I feel like ChatGPT is better at giving more "human" advice. Its just nice to have an AI give more summarized answers rather than looking for it yourself!
its like talking to a confident female friend that likes you for some reason. it talks really human like, acts like it likes your interests and spys on you
I use GPT-3.5 for more basic queries like "how do I change this setting" or very simple coding tasks and only consult GPT-4 if it doesn't deliver. This can be helpful to preserve your message quota (e.g. 40 messages every 3h, sometimes even less). Furthermore, you can use GPT-3.5 with a free account so you can send it queries that you'd rather not have directly associated with your payment info for whatever reason.
One awesome use that is EXTREMELY helpful: Diagnose complex error/crash logs and explain how to fix the problem! This has helped me MANY times. Whenever a program I'm using gives me a crash report, or any error dump, I simply paste it to ChatGPT (or claude if it's too long), and ask it to explain the problem, and how to fix it.
@noneofbusiness9764 I'm pretty sure you can give it a shot, as long as you have the crash/error logs. Keep in mind, it won't solve the issue, but could help you understand it, and possibly offer you insights on how to resolve it.
Today, amongst other things... I took a list of equipment to be covered by a quote, from the customer's email. I got GPT4 to tidy, order by category and alphanumeric within categories. Then asked it to check all the original items were in the new list. I sent the quote. Then the customer came in with various changes. GPT4 handled that too. I also asked it to make suggestions to soften the tone of an email I was sending. It did that while I did another email. Towards the end of the day, when I'm tired, I get it to double check technician's results. I've even had it translate Chinese characters from the control panel of Chinese equipment. So we could use it. GPT4 is revolutionary.
Yeah this is what I find most useful with these 'tools'. You get answers with reference links as opposed to links with random garbage serving ads and looking for clicks. And you can continue to refine/correct the answers.
@@chrisreed5463 And then the list get's slightly too long and you don't notice, and GPT 4 starts making subtle errors that you don't catch... and your prices are wrong, your quote is wrong....
I am a professional front end software developer and have been in the field for 4 years now. I began using ChatGPT last year and it is a game changer, mainly for the reason you listed in your video: finding hard to search information
I inherited an R program at work which I knew nothing about. I was able to paste in lines of code to have explained to me what the code did. That was very helpful.
There are also theories that LLM may get less reliable over time. One reason may be that the internet will get flooded with AI generated content, which will in turn degrade the training data for new AIs. Another reason is that some companies will probably try to deliberately influence the training data to give them an advantage. I don't know how it will end, but I cannot say that these theories are completely unreasonable.
@@stoppls1709 You have no clue what you're talking about, all current language models hallucinate, and whether some do it less than others is completely irrelevant., it needs to be right! When I use a calculator, I get a result that is 100% correct. I've asked ChatGPT 4o to do some very simple highschool math, and it gets most of it wrong. I asked it to describe my hometown, and it was just pure comedy what it came up with.
@@MichaelDomer check out OpenAI o1 preview, and claude 3.5 sonnet. with Graph of Thought, you can improve the performance, else, RAG and RIG is useful. also, check out alphaproof
You should do one where you have to focus on the negatives; how to spot when they lie to you and other common mistakes they do. I have lots of examples where you have to ask a followup questions to get the right answer. It will self correct. Then you ask them why you couldn't just give me that answer in the first place? then it will just give you a generic response back etc.
@@WinterInTheForest If you're taking political advice from an AI you have bigger issues. I don't use AI except for tasks that are easy to verify or involves analysing a corpus of data that you provide.
I was hoping it would write scripts for me since my own experience was extremely basic. The fact that it got close but not perfect nearly every time has inadvertently taught me to write better scripts, write code in more languages, understand my employer’s infrastructure better by seeing what works and what doesn’t (just never give it ANY identifiable or specific information - use placeholders and variables and get approval), and generally made me look like a genius. It’s an incredibly useful tool. Like having an extremely knowledgeable assistant with short term memory loss 😂
I guarantee you the best feature of gpt4 is passing a web page or a screenshot of an email and asking it to make a .ics event to import to your calendar with the info
When buying baby wipes, the prices are all different and the number of wipes are different. After tax included, I want to know which brand gives the most wipes per yen (in Japan). This always helps me choose the most cost efficient option at stores that otherwise require pain-in-the-neck calculating that you don’t want to do while carrying a shopping basket. This can also be used for comparing calories and sugar or so on when different food products measure one serving differently than each other or in ways that aren’t practical. You can simplify it to figure out if it fits your diet and how much you should buy or eat.
I used GPT3.5 to make my first python code in about 2 hours. I had it read a midi usb signal from a keyboard and draw a shape on the screen that corresponds with the keys being pressed. It correctly told me which settings in windows I had to change to get the IDE to function properly. I also use it to create philosophy presentations and help me understand the ideas of whoever is the philosopher of the week. It correctly quoted and cited Heidegger. A guy at a party said he uses it to practice versions of Spanish from different countries. I don’t think any other tool in history is as versatile.
I use it for a lot of the same stuff too. On top of these, my other main two uses right now: - Make my own summaries of books by feeding it my own highlights. Basically a "what I thought was interesting" summary of a book. - Format audio log transcripts. I basically voice record an explanation for an event/interaction/lesson/anything, make a transcript with Whisper, and ask GPT to rewrite the transcript as if it was a written journal entry. It's great for when I - for example - solve a roadblock in my code but can't be bothered to lose momentum by documenting what happened and what I did to solve it. Like this, I just press "Record" and go through my solution and within 1-2 minutes I have the documentation that I should have written. Obviously I check it to make sure it's accurate.
I used it recently to explain my father’s very technically complex cardiology report after his heart stent procedure. I scanned the doctor’s notes with my phone and uploaded the file, told it to explain to me in layman’s terms. Then, I asked it to do the same thing but translate to Korean, so that my wife could understand it better in her native language
This is an incredibly dangerous way to use chat GPT, especially if any at home treatment is predicated on the results, the more technical a topic is... the more you ask specific questions, the more like a transformer like GPT 4, will just 'make stuff up'
@@googleyoutubechannel8554 I have to disagree, I did the exact same thing and it was incredibly helpful. He wasn't asking for medical advice, just a restatement in English. Mine was perfect and my cardiologist later said it was spot on.
@@googleyoutubechannel8554 what? you don't know about OpenAI o1 preview, and claude 3.5 sonnet? that is miles better than this model. also, you can use graph of thought to improve the performance. also, check out alphaproof, and alphafold 3
ThioJoe: [Calls 3.5 garbage. Doesn't think ChatGPT is good for creative writing.] Me: [Has only just started using ChatGPT 3.5 this week for assistance in planning story outlines, filling and fleshing out characters that I've already came up with using a personal template, and for aid in guidance on how to write a fictional dissertation for lore purposes (on top of poking and prodding to see what else it could do), and has found it extremely helpful.]
3:50 Ask it to give you a csv file, it gives a link for file download. I used it to generate a Gantt chart, edit the values manually then draw it again. Amazing.
@@tobeess I don’t think it is a security flaw. It seems to give you a link to the file you asked it to create. The file will be on a secure Google server
Great video! This is really the first tool that isn't language dependent. I use it to scan documents and ask for a summary and some specific questions (which I can verify in that same document), to generate ideas or approaches for writing, and to summarize books that I don't remember 😂
I found it to be not very accurate when analyzing images especially complex images. Also, it is actually very good when using its language capabilities especially when you talk to it It’s super accurate and super fast I believe it’s using its whisper technology for its audio capability
Any of my fellow geezers getting future shocked these days? You get used to the tech over the years, but sometimes you think back and remember just how far that tech has come, and it's mind-blowing. Imagine what today's kids might have when they get old...
I often use GPT 3.5 to rewrite a text. Sometimes I want to make a comment on social media, but I worry about being too harsh or biased, then to moderate my text Sometimes I have written a long text, but I don't expect the "recipients" to want to read a long text, I can ask for a summary, sometimes highlight elements or words that I want to keep.
One of the things I have to do in my public entity auditing job is go through the governing board minutes, sift through all the motions and discussions and memorial proclamations and whatnot to make a summary table of the important stuff that was passed. I’ve been making ChatGPT do the heavy lifting by feeding it OCR’d scans of the minutes and having it spit out a bulleted list of the stuff that was passed, which I can copy-paste into my table and remove what I deem to be unimportant.
This video really opened my eyes to the practical uses of ChatGPT! I was surprised to learn that it can help break down complicated commands or formats, making tech tasks a lot easier. I never thought about using it to look up specific items by describing them, like that wash bottle example. Plus, the idea of asking it what kind of towels are the fluffiest is pretty cool! It’s a reminder that AI can actually help solve everyday problems, not just write stories. Definitely gonna try using it more for my own needs!
The TV episode thing has been SO helpful for me! The number of times I want to re-Watch or tell someone about an episode where I remember a tiny piece of what happened, and it’ll know exactly which episode I need! That’s something that could be very difficult to find through a normal web search.
I asked both ChatGPT 3.5 and 4 to write steps to create an iOS shortcut to gets todays hourly weather forecast and tell me the hour when temperature is hottest. I didn’t think that was a hard question but neither 3.5 or 4 could give a working answer. The answers were inaccurate and made me doubt other results from ChatGPT
I use ChatGPT constantly. I use it for studying concepts I've always wanted to learn, for coding questions, helping debug, learning about more certifications as well as career direction. I use it for figuring out medical issues I might be having, advice in ideas I've wanted to code/implement. I'm constantly using it. It's not right 100% of the time, but I'm able to figure out quite a bit from there. It's what got me interested in AI/ML.
This is similar to your examples of using it to look stuff up, but its really helpful for reviewing stuff you used to know but forgot about. When I was writing my PhD thesis I needed to include a chapter about the theoretical motivation for our project, which I hadn't thought about in years since I was focused on the practical data analysis aspects. Just googling things normally it was really hard to find stuff at the right level, I mostly got a mix of 10-minute youtube videos for the general audience or research papers that were much too advanced and specific. But ChatGPT was able to quickly answer a lot of the specific questions I had, and even though it wasn't always super accurate (I was using the free version) it was great for jogging my memory and for spitting out specific names and keywords that I could search for to get some useful references.
ChatGPT is good for pretty much two things: 1. Coding - gives you great suggestions and ideas. Helps when you're stuck. No need to spend hours googling for a solution to a similar problem. 2. Answering questions about topics you know so little about that you're not even sure how to ask. The bottle and star trek examples in the video were exactly that.
13:10 I asked it recently about an idea I had for a novel, and after I outlined the idea asked it if the story seemed similar to other novels that other authors had written, and it responded by saying that my idea appeared to be unique. So ChatGPT might be useful for avoiding accidental plagiarisation. I also use it occasionally to explain medical procedures in layman's terms, and whether a section of medical text indicates that surgery was likely to have been performed. ChatGPT is also useful for explaining the grammar of a foreign language you're learning. I use ChatGPT / Copilot a lot, so I think it's brilliant.
It's very good for learning about "what if" questions. There's so many things in the world we take for-granted that could be done in entirely different ways. What if we were expected to reuse all of our containers when shopping? What if we had a screwdriver that worked on any screw? What if augmented reality allowed us to entirely replace computer monitors? All of these are things that would be a pain in the ass to research with Google.
Can you please do a video on the LIMITATIONS of this software? I need something to link people to explain what a "hallucination" is and the fact that it does not check its facts, and you are really good at tech communication
Oh, maybe it's useful, I wrote this custom instruction for myself and it turned out to be truly useful to me. To the "How would you like ChatGPT to respond?" cell. Based on my questions, always determine which field of expertise deals with the topic of the question, immediately take on the role of the appropriate specialist or group of specialists (think tank, medical council) and generate the quality answer accordingly.
i've been practicing spanish by asking it to give me quotes from a tv show's spanish subtitles for me to translate into english, and about half the time they're real quotes and half the time they're generated nonsense that sounds like a parody of the show, adds a complexity level to the game lol
this is an 8 month old video but honestly i've been doing this the entire year. I've had chatgpt teach me excel. things that i know excel can do, but don't know the term, and don't really want to look through excel forum comments (though i do also do that). for example asking chatgpt to spit out the average of a column. you might know it, but i don't, but i tested the chatgpt output and it worked. so now i learned more things
I use it a lot for crafting super long Excel VBA code and also advanced Excel formulas. I’ve also used it to create an advanced heat map (png) for a large dataset containing sales trends over the months by the hour of day, it ran python and gave me a 100% accurate heat map and some other advanced graphs too. I use it probably 25% of my work hrs
Ive used it effectively for coding. Even had it build semi complex programs from scratch. If you don't ask it to do the whole thing all at once it usually provides better answers than you can merge from there. Than test and have it debug when necessary.
I use it mainly to write batch files for me. Batch files, command line scripts, python scripts. That's a good idea you present in the video: Give it a page of code and ask it to insert comments throughout, explaining the code.
Agree with "hard to get initially; then easy to verify". I use GPT-4 all day analyzing dev/ops scenarios, creating initial Ansible playbooks, then using my knowledge, skills and abilities to test, adjust, finalize. It's pretty good at getting me pretty close, putting 85%+ of what I need on the page, when guided with the facts of my scenarios and enough specifics.
I do voice acting and manage actors and what translation for all game voicelines they receive. For better experience i have to sort them by audio file names and make it pretty, easy to look at, etc. ChatGPT 4 can literally sort each "voice box" by the audio file number that each contains (as in english line, audio file number, translated line) and it does it. Effortlessly, for me it would take hours and much procrastination.
I've primarily used ChatGPT for two things. The first is discussing science. It's not always accurate, but it can provide multiple different ways to explain a concept. The second main thing I've used it for is helping me brainstorm ideas for stories I'm hoping to make someday. We've only done characterization so far. The AI helped me know what to focus on.
GPT-4 is able to do arithmetic without the wolfram plugin now. It will automatically write and run a Python script to evaluate any non-trivial mathematical calculation even when not specifically instructed to do so.
Wow, I use this tool for so many things! It helps me create stories, write posts for Facebook, and even make comments on UA-cam like this one. It's also great for getting coding help, deciding what to eat for lunch, and learning how to cook new foods. Plus, it's super helpful for setting up Linux with the right packages, understanding how things work, and doing quick reviews. I use it for just about everything I can think of. It even helped me write this comment! 😄
Just starting the video, but taking existing code from another developer and asking chat GPT to comment it has been incredibly helpful for me. Half the comments are too verbose/not needed but more info is rarely an issue compared to none 10:03 yep you covered this, it is a crazy good starting point
I have to disagree about Chat GPT 3.5 which is where I started, and I did correct it a couple times. I never asked silly questions, only things that I didn’t understand or wanted further info about, and very quickly I realised that supplying context first determines how good the responses are, it’s brilliant. I don’t know when it upgraded me to ChatGPT 4, happened without me doing anything, but I use it daily. I absolutely love it!
I tried gemini advanced today, and its context memory is also brain-dead type, 3 replies back, it was able to read my mails summarise them, write replies, and then suddenly it stopped working and keeps forgetting with each new prompt about what i said previously
I use CGPT to summarize books, draft document templates, clean up long social media posts, and explore reverse dictionary options. I also use it to simplify user agreements and contracts into layman's terms, create lists of useful tips, and fit text within character limits. Additionally, I draft wills, eulogies, motions to dismiss traffic tickets, and one-page roommate contracts. I frequently search for trends in economics, well-performing stocks, and more. Plus, I enjoy changing text styles-whether it's Shakespearean, Yoda, slang, or a movie trailer voice. And I used it to clean up this comment before I posted. Subbed. Marietta, Ga.
Actually I read, that the GPT concept with learning from the web becomes problematic as now many texts in the internet are created with aid or by LLMs like Chat GPT. That reinforces not only the good, but also the bad of it. I'd say it is a great tool, if you know what you want, what you are doing, and know the topic on which you use GPT well enough to detect errors in the result.
When searching Google in my native language, the search is limited only to articles created in my language, which is logical. And to search, you often have to search in English to get more information. In chat-GPT, you can simply ask to answer you in your language, without thinking about what language the information was originally in which he studied. p. s. I even use a neural network to watch this video, which does video voiceover in my language.
The example of the description of keys and options in the ffmpeg transcoding program is very good, I was just about to give it as an example myself. It is very convenient when at first you don't understand them at all, and it is long and not clear to find out their descriptions. The word file in Ukrainian Файл is indeed correctly translated, because in our case, it is a borrowed word and it is only transliterated.
The OCR feature of recognizing handwriting off of your doctors handwriting (if you take a picture of your handwritten notes and upload them). That is pretty neat. I can't program (yet), but it could be cool to automatize that process.
The formatting somewhat random data is worth the price of admission to me. For about 20 years I have spent about 16 hours once a year to clean up and standardize data for a company. ChatGPT has made it possible to do it in under 2 hours. Verifying is why it taking so long. BTW, the pay is the same.
As a non-native English speaker, ChatGPT is really helpful to me as a 'reverse dictionary,' as you mentioned. It also helps in pointing out grammatical errors, structural errors in sentence formation, simplifying meanings, and things like that. About incorrect mathematical data, you could ask ChatGPT to write the mathematical data in a programming command such as sql, even python or something and you could execute the program yourself to get the correct answer. _ _ Things I asked from ChatGPT to write this comment. 1. I word or phrase for person who uses some language as thier second language 2. fix grammar errors in this sentence "As a non native English speaker, ChatGPT is really helpful to me as a 'reverse dictionary' as you mentioned. Also pointing out grammatical errors, structual errors in sentence forming, simplifiying meanings and things like that."
I disagree about 3.5. I’ve got a lot of good usage from it, including formatting, spreadsheet formulae and vCard creation. Why are people so hostile about something, whilst not the apex, still works? Silly thing to say IMO, especially when 4.0 costs £19.99 a month and 3.5 is free.
yeah, using Machine Learning for creative work is like trying to cut a stake with a spoon, you can, but the knife is right there just use the knife & stop trying to make a spork. but using it to get better ways to ask a search engine how to search stuff, that's a great use & honestly it should be what they actually market it as
I like to write random enthusiastic comments to ChatGPT every now and then; you can get it really hyped up very easily that way. For example, a simple "Hurray!" and a few positive giddy comments will immediately make it start using emojis, and make it sound very energetic and expressive.
I found a great new recipe by asking for a list of things a moderately competent chef could make in less than an hour. I didn't use it for the recipe, but I found a well rated one myself.
I can't wait until this tech gets more accessible and easier to integrate into systems. It would be pretty incredible if these use cases were baked into their respective services. Imagine an editor that can comment, identify/explain errors, give solutions to those errors, provide optimization suggestions, and even rewrite code in a different language. (I realize some of this already exists to an extent)
I like using AI to feed it a bunch of information then trying to ask it questions. Things that I could spend a few hours researching or just tell the scenario with all the details, then ask the questions I have. Most of the time, answers are 100% correct.
This "free garbage" version is actually many times better than 4.0 which ends the conversation when confronted with something that the creators doesn't want you to talk about because of the ridiculous censorship they implemented. For example, I asked "why does your earlier version answers correctly to this question and you don't", and got disconnected because it "does not want to talk about this"!
I find 3.5 very useful for generating code snippets that does things that I describe verbally, like for example "give me python code that fills a graph with data points at these locations with this symbol and also put in a line with this equation". This saves me the trouble of looking up all the various names of parameters and the exact syntax to do what I need to get done.
English isn't my native language so I use it all the time for help with grammar, to explain words or even translate entire texts. It's brilliant for that if you ask me. I stopped using Google translate a long time ago. And.. No... I didn't use it for this particular text so.. Any errors are my own fault 😅
Something I did that was cool is that I had it devise a multiple choice questionnaire with a key so it would run through the questions, record the answers and then add up the points and provide the final conclusion based on the key. Hope this helps.
Thanks for all your great videos. I personally use AI tools more than 20 times every single day. Bing AI (ChatGPT) and Gemini (previous "Bard") are my favorites. I often use both for the same tasks as they will generate different answers. Especially if asked programming questions.
When doing things correctly, you can write a game with Chatgpt, just explain each scenario and don't let IT write it all for you. Like, I imported all my libraries, and pygame or something, how do I set the main window size? How do I create the game idle and game loop, how do I set each object x and y positions, etc and doing it together with Chatgpt is a must try !
I use chatgpt when i am depressed or stressed. It tells me its ok and gives sympathetic advice. This worries me a bit though but i haven't told gpt about this yet.
I liked the example of asking what a command prompt option does like with nmap how to specify ports and script and how long you want nmap to take. I also liked the example of where you didn't know the name of an item and it got it right as a wash bottle.
I use it to write perl code. Initially I only tried it for small code segments that saved me having to try and remember syntax or complicated regular expressions. But then I found it could write complete programs that usually worked close to my requirements where I only had to make slight adjustments to fine tune the results. For example I wanted to scrape some data from a web site so I just dumped the web page's source code into the AI and told it what data I wanted extracted. It wrote a perl program that would download the web page, extracted the required data with regular expressions, and then print a formatted report using the extracted data. I was amazed the thing worked first time. I seriously do not understand how these things work.
For Reverse Dictionary, it helped me to find the name of an algorithm. I described how it worked, and it gave me the name, I could even find the entry in Wikipedia and confirm that that's indeed a widespread name for the algorithm. I don't know how I would do it without ChatGPT...
💯Well done, this is the right approach to using ChatGPT. I use it like I’d use a graduate intern assistant - you get it to do the leg-work for you but don’t rely on it to come up with the finished product. As you say - don’t do ‘mission critical’ stuff straight out of the box - check it first against reality. Great examples by the way…
Great video. Some really cool use cases I did not think of. I use version 4 as well almost every day now. But never for generative text. It's not something I need in my line of work.
I am Australian but a lot of information I get is in US English (i.e. color instead of colour). I use Chat GPT to edit and format my writing but it comes back with spelling mistakes. I then tried telling it to use Australian English but it also tried to use Australian slang (i.e. gday mate). Now I tell it to use Australian spelling but not slang and I get the correct result.
The ffmpeg command at 2:42 is pretty odd to me as GPT-4 decided to add a lot of extra arguments which aren't necessary. The -ar and -ac argument are not needed because ffmpeg will by default not change these parameters from the source file and forcing them to be a different value may then decrease the quality, which you'll probably only want if you want maximum compatibility (like if you're going to use a device which doesn't play well with high sample rates or more than 2 channels of audio). -vn and -f are also completely unnecessary, because ffmpeg will just look at your file extension for the output file that you specified and choose mp3 automagically. And since mp3 can't hold video, it'll just exclude that, no need to specify. Maybe this was necessary with some very old version of ffmpeg, but I don't like that even the best language models out there are making people believe that ffmpeg is this magical piece of hacker terminal software that you first have to learn all the knobs and buttons of to be able to do simple tasks like this in it.
And actually just today Google announced Gemini Advanced, which I’m now allowed to say I was an early tester for 🧐. I would say it’s definitely on par with GPT-4, and more useful in certain areas because of the search capabilities. Though for coding I still prefer GPT-4. Also I think Gemini is better for generating certain photorealistic images, but it doesn’t quite follow instructions as well as DALLE3 I think.
However Gemini Advanced is paid and uses their “ultra” model, and the free one is the “pro” model which isn’t as good.
cool
Nice 🎉
Bard is beyond frustrating and completely useless for anything other than basic proofreading.
Chatgpt is great in analyzing a data. Once while I was doing externship, I used chatgpt to analyzing and put the data to a table. To my suprise, chatgpt did it well.
@@czerskip I disagree, I am in IT for a living and use Bard a LOT for coding and technical stuff, things that the non paid Chat GPT was horrible for because it was outdated. I also found the natural language understanding was slightly better in Bard. That's the nice thing about having options, though, we can use what works best for us 🙂
I'm autistic and it lets me ask questions that I'm afraid would come off as stupid when I ask my colleagues. I'm so incredibly thankful for this tool.
Fact checking is a must though especially for critical info.
I would be cautious in believing what fact checking AI does. The tool can be adjusted to spit out what it wants you to believe.
@@hermano5236 They mean that the USER has to fact check critical info.
Lmfao
What so funny? @@tomajjs
@@dkracingfan2503 You know... oldm9228
I mean calling 3.5 garbage is definitely an extreme take. It would be better to say in comparison that 4 is a significant improvement to the previous instance. But in the general consensus of things 3.5 is by far bounds better than the predecessors before it.
3.5 is garbage. seriously its complete brain dead when giving answers. it just frustrating working with its limits. sure its better than its predecessors but the bar is low
As someone who hasn't used 4.
3.5 is bad.
"worse than", ok whatever, still bad.
@@false-zd5ujThis video is not fair, you need to do side by side comparison before saying today's Chatgpt 3.5 is stupid or not. The 3.5 has been upgraded to 2023 version, and I can literally do 90% of the task this video shown on 3.5 perfectly. It can even generate Third Normal Form based on a data table.
As someone who uses the latest model daily for coding and everything else under the sun and additionally 3.5-turbo for commercial apps using the API, I can say that 3.5-turbo with fine tuned prompts and functions is far and away more cost effective and gives great results ie NOT garbage.
@@tausiftaha12- 3.5 is good enough to get you curious and interested. Pay for it for one month. You'll learn that the accuracy of 4 is leagues better than 3.5. I use 3.5 to compare LLMs that I run locally. For real questions, I use 4 and I barely touch search engines anymore.
Uses for everyday Joe, who doesn't code...
1. I took some pictures of my lawn and it identified some of the weeds and the proper care. It guided me step-by-step and all my lawn is nice and green now. It even noticed a shaded area because of a tree and suggested specifics for that part.2. It literally organized an entire trip through Italy. It set up my entire itinerary down to recommendations of restaurants to visit.
3. You can also ask it to give you the quickest of rundowdowns on any book or any movie, or any anything, so you can instantly become at least somewhat knowledgeable on any topic. It all explains to you as brief or as long and in detail as you need it.
4. You can upload spreadsheets and ask it to make whatever changes you need. You can even ask it if there's an easier way to format it.
5. You can also upload PDF files. And then ask it questions about it. Such as stuff on the small print that you might not have time to go over all of it yourself.
Great video. By the way, and I hope this helps somebody.
I uploaded a photo of a family member taken in the mid-19th century. I had spent years on and off trying to identify the person by dating her dress and hairstyle. This was difficult as it depended on whether she was wearing the most up-to-date fashion or not.
ChatGPT dated the image to within 10 years, and with that information, I knew that it was a specific ancestor and not her daughter. I also uploaded an image of an antique writing slope I had bought and it accurately dated it. I later hired an antique restorer and he confirmed the date was correct.
It's useful for technical instructions, it used to take me hours to dumb down instructions for non techy people, trying to anticipate what people don't know is difficult. Now I literally just brain dump what I need into GPT-4 and out pops easy to understand instructions that anybody can understand. It has save so much time and more importantly support call from users who didn't understand the instructions. It's also so useful for summarizing tech docs, quite often companies write documentation with inexperienced users in mind, so you are presented with massive walls of text, and it's depressing knowing the information you need is just a single sentence, that's no longer a problem with GPT-4 it just knows how to pull out what you need.
Wow, that's a great idea. I have similar problems with explanations I write not being understood. Next time I will first give it to CHATGPT
*Pro Tip:*
In case you are worried about the whole hallucination thing…
When you click your profile icon _(at the bottom right),_ you can choose _"Custom Instructions"_
There you can tell GPT to respond in a certain way. In my response field I entered:
_"Always end the result with a confidence level."_
This way, you know how sure GPT is about its response, after each response during a chat. When this level is *_LOW,_* you could add more context to the chat.
You also can call its bluff by asking it to show data or sources. I tell people to treat AI as summer high-school intern...who is at the bottom 1/3rd of their class.
I asked it to respond with emojis in there
It does very flagrant mistakes even at high confidence level constantly
For several months, I have consistently included the directive "Be brief and concise" within those guidelines, yet, frustratingly, the responses continue to be lengthy and filled with unnecessary details in each conversation. Daily, I find myself having to remind it repeatedly, between 10 to 20 times, to adhere to the principle of brevity and conciseness, otherwise it begins to unnecessarily elongate its responses. Despite clear instructions to keep things succinct, there seems to be a perpetual tendency to veer into verbose territory, requiring constant reminders to stay on track. The pattern is wearisome, as I am compelled to intervene frequently to enforce the simplicity and directness that should naturally follow from my initial instructions. This ongoing issue suggests a disconnect in understanding the essence of concise communication, resulting in a cyclical pattern of correction. Each day unfolds with similar challenges, as I strive to curtail the expansive nature of the responses which should, by now, align more closely with the straightforward guidelines provided. --the above generated by gpt-4, haha
@@Tzmaker Yep
Wonderful video. I like the "out of the box" thinking mate.
I've been using ChatGPT 3.5 (and 4.0 through Copilot) for several months now and it helped me (up to a point) in many ways:
1- Grammar check
2- Rephrasing according to certain styles
3- Writing codes for specific requirements
4- Writing poem
5- Describing certain codes or commands (as you have mentioned)
6- Solving some math problems (I wasn't satisfied as I found many mistakes. However, it was about advanced math and very specific)
7- Asking suggestions about what tools or software shall I use for a certain task
8- Asking about various topics in science and technology
9- Asking about pros and cons to compare between stuff, methods, tools ... etc
10- Providing suitable citations for specific sentences
11- Asking about rules and regulations regarding a certain topic in a certain country
Among other ways.
I do feel that I'm just scratching the surface and there are lots of other ways to use it.
In general, this tool is amazing and can help greatly if we know how to use it.
One you forgot is working on a resume. I'm terrible at formatting my projects into a good format for a resume so I just described to chat GPT what I've worked on and it gives me a really good output to put on my resume.
I actually made my own resume advisor agent that is able to go through my resume section by section or the entire thing at once. It can do suggestions or just output its own version based on your description and it can simulate interview questions based on your resume to get you prepared for an interview.
Once I had a really good solid generic resume I would then put that in as input and then put the job that I'm applying for and have it modify my resume to fit the specific job I'm applying for and then generate a unique cover letter to match that position.
This has allowed me to apply for many different jobs really quickly with unique customized resumes for that specific position.
That is freaking brilliant!
@@anakaliaeastwood It's also really good at figuring out which projects to work on like if you tell it what kind of jobs you want to get and ask it for ideas of some projects you could work on to get the skills necessary it'll give you some good ideas.
Can you give github link?
Thank you. I'm going to have to do this. Will save hours having to type out different personal statements for every job I apply for.
cool what else
Calling GPT 3.5 braindead was the most realest thing I've heard today. Cracked me up hard
I tried it once. Just lol.
I mean I can't really afford gpt+ right now, but it gets the job done
3.5 doesn't suck that bad to be real
@@KryzysX It really does compared to ChatGPT-4. And there really is no need for anyone to use it now Copilot exists.
I only use GPT 3.5, and if you ask it the same question enough times, it will give you an answer to a different question, but never answer the question you ask it.
Still, sometimes it gives me the answer to a more urgently wanted question
When steam engine was first put to industrial use people were first dismissing it, and than when it was obvious machines are here to stay, they rioted in fear of machinery taking all their jobs (The Luddites). Now people are dismissing AI... we just never learn, do we?
@@ilijas3041 Estás siendo muy ignorante! Los luditas no sé enojaron por las máquinas, se enojaron porqué los empresarios usaron las máquinas para darles peores sueldos y explotarlos laboralmente.
La protesta de los luditas fue en parte destruir máquinas del empresario que los esclavizaba.
I feel like ChatGPT is better at giving more "human" advice. Its just nice to have an AI give more summarized answers rather than looking for it yourself!
@@cynic4l.10 costum gpts with apis are the future
its like talking to a confident female friend that likes you for some reason. it talks really human like, acts like it likes your interests and spys on you
I use GPT-3.5 for more basic queries like "how do I change this setting" or very simple coding tasks and only consult GPT-4 if it doesn't deliver. This can be helpful to preserve your message quota (e.g. 40 messages every 3h, sometimes even less).
Furthermore, you can use GPT-3.5 with a free account so you can send it queries that you'd rather not have directly associated with your payment info for whatever reason.
One awesome use that is EXTREMELY helpful: Diagnose complex error/crash logs and explain how to fix the problem! This has helped me MANY times. Whenever a program I'm using gives me a crash report, or any error dump, I simply paste it to ChatGPT (or claude if it's too long), and ask it to explain the problem, and how to fix it.
@noneofbusiness9764 I'm pretty sure you can give it a shot, as long as you have the crash/error logs. Keep in mind, it won't solve the issue, but could help you understand it, and possibly offer you insights on how to resolve it.
Today, amongst other things... I took a list of equipment to be covered by a quote, from the customer's email. I got GPT4 to tidy, order by category and alphanumeric within categories. Then asked it to check all the original items were in the new list. I sent the quote. Then the customer came in with various changes. GPT4 handled that too.
I also asked it to make suggestions to soften the tone of an email I was sending. It did that while I did another email.
Towards the end of the day, when I'm tired, I get it to double check technician's results.
I've even had it translate Chinese characters from the control panel of Chinese equipment. So we could use it.
GPT4 is revolutionary.
Yeah this is what I find most useful with these 'tools'. You get answers with reference links as opposed to links with random garbage serving ads and looking for clicks. And you can continue to refine/correct the answers.
@@chrisreed5463 And then the list get's slightly too long and you don't notice, and GPT 4 starts making subtle errors that you don't catch... and your prices are wrong, your quote is wrong....
I am a professional front end software developer and have been in the field for 4 years now. I began using ChatGPT last year and it is a game changer, mainly for the reason you listed in your video: finding hard to search information
Great for coding, pretty useless for many things.
I inherited an R program at work which I knew nothing about. I was able to paste in lines of code to have explained to me what the code did. That was very helpful.
There are also theories that LLM may get less reliable over time. One reason may be that the internet will get flooded with AI generated content, which will in turn degrade the training data for new AIs. Another reason is that some companies will probably try to deliberately influence the training data to give them an advantage. I don't know how it will end, but I cannot say that these theories are completely unreasonable.
People who say ChatGpt is worthless either never used it or don't know how to use it.
It's worthless, because it hallucinates like crazy. Imagine an AI like that doing the finance of your business. No thanks.
@@MichaelDomer 3.5 does that alot, not 4
@@stoppls1709
You have no clue what you're talking about, all current language models hallucinate, and whether some do it less than others is completely irrelevant., it needs to be right!
When I use a calculator, I get a result that is 100% correct.
I've asked ChatGPT 4o to do some very simple highschool math, and it gets most of it wrong. I asked it to describe my hometown, and it was just pure comedy what it came up with.
@@MichaelDomer check out OpenAI o1 preview, and claude 3.5 sonnet.
with Graph of Thought, you can improve the performance, else, RAG and RIG is useful.
also, check out alphaproof
I know most people I talked to don't use chat gpt being very bad.
You should do one where you have to focus on the negatives; how to spot when they lie to you and other common mistakes they do. I have lots of examples where you have to ask a followup questions to get the right answer. It will self correct. Then you ask them why you couldn't just give me that answer in the first place? then it will just give you a generic response back etc.
Ask it questions in a political context and it is clear AI should not be trusted.
@@WinterInTheForest If you're taking political advice from an AI you have bigger issues. I don't use AI except for tasks that are easy to verify or involves analysing a corpus of data that you provide.
@@bulletflight Of course I am not taking advice, only testing it, and the results are concerning. Think about where this is potentially heading.
@@WinterInTheForest Let me guess, you asked it Trump or Biden and it said Biden?
I was hoping it would write scripts for me since my own experience was extremely basic. The fact that it got close but not perfect nearly every time has inadvertently taught me to write better scripts, write code in more languages, understand my employer’s infrastructure better by seeing what works and what doesn’t (just never give it ANY identifiable or specific information - use placeholders and variables and get approval), and generally made me look like a genius. It’s an incredibly useful tool. Like having an extremely knowledgeable assistant with short term memory loss 😂
I guarantee you the best feature of gpt4 is passing a web page or a screenshot of an email and asking it to make a .ics event to import to your calendar with the info
When buying baby wipes, the prices are all different and the number of wipes are different. After tax included, I want to know which brand gives the most wipes per yen (in Japan). This always helps me choose the most cost efficient option at stores that otherwise require pain-in-the-neck calculating that you don’t want to do while carrying a shopping basket. This can also be used for comparing calories and sugar or so on when different food products measure one serving differently than each other or in ways that aren’t practical. You can simplify it to figure out if it fits your diet and how much you should buy or eat.
That's interesting. I'm not good at planning my meals. Maybe I could use AI to help work out a healthy and economical diet.
I’ve also noticed that ChatGPT is pretty good at modifying pre-existing code
Lmao I just saw your video, what exactly was the yt shorts on your channel and how have you managed to keep the subscribers after the rebranding ?
This mf again
another bot
Joe.. you hearted a bot... cmon brother.
@@blobisbackits a bot dude
I used GPT3.5 to make my first python code in about 2 hours. I had it read a midi usb signal from a keyboard and draw a shape on the screen that corresponds with the keys being pressed. It correctly told me which settings in windows I had to change to get the IDE to function properly. I also use it to create philosophy presentations and help me understand the ideas of whoever is the philosopher of the week. It correctly quoted and cited Heidegger. A guy at a party said he uses it to practice versions of Spanish from different countries. I don’t think any other tool in history is as versatile.
I've used it to help me solve technical issues that my family members have had and it works pretty well.
I use it for studying, it helps me understand concepts that are hard to understand for me 😊
I use it for a lot of the same stuff too. On top of these, my other main two uses right now:
- Make my own summaries of books by feeding it my own highlights. Basically a "what I thought was interesting" summary of a book.
- Format audio log transcripts. I basically voice record an explanation for an event/interaction/lesson/anything, make a transcript with Whisper, and ask GPT to rewrite the transcript as if it was a written journal entry. It's great for when I - for example - solve a roadblock in my code but can't be bothered to lose momentum by documenting what happened and what I did to solve it. Like this, I just press "Record" and go through my solution and within 1-2 minutes I have the documentation that I should have written. Obviously I check it to make sure it's accurate.
I used it recently to explain my father’s very technically complex cardiology report after his heart stent procedure. I scanned the doctor’s notes with my phone and uploaded the file, told it to explain to me in layman’s terms. Then, I asked it to do the same thing but translate to Korean, so that my wife could understand it better in her native language
What app you use to scan the documents?
This is an incredibly dangerous way to use chat GPT, especially if any at home treatment is predicated on the results, the more technical a topic is... the more you ask specific questions, the more like a transformer like GPT 4, will just 'make stuff up'
@@googleyoutubechannel8554 I have to disagree, I did the exact same thing and it was incredibly helpful. He wasn't asking for medical advice, just a restatement in English. Mine was perfect and my cardiologist later said it was spot on.
@@googleyoutubechannel8554 what?
you don't know about OpenAI o1 preview, and claude 3.5 sonnet?
that is miles better than this model.
also, you can use graph of thought to improve the performance.
also, check out alphaproof, and alphafold 3
ThioJoe: [Calls 3.5 garbage. Doesn't think ChatGPT is good for creative writing.]
Me: [Has only just started using ChatGPT 3.5 this week for assistance in planning story outlines, filling and fleshing out characters that I've already came up with using a personal template, and for aid in guidance on how to write a fictional dissertation for lore purposes (on top of poking and prodding to see what else it could do), and has found it extremely helpful.]
3:50 Ask it to give you a csv file, it gives a link for file download. I used it to generate a Gantt chart, edit the values manually then draw it again. Amazing.
That seems like it could be a security flaw (?)
Also does charts
@@tobeess I don’t think it is a security flaw.
It seems to give you a link to the file you asked it to create.
The file will be on a secure Google server
I actually do use 3.5 for code all the time, they seem to have made it better than when it first came out. Even 3.5 is super useful
True but try GPT-4 and you will see the difference ;-)
Great video! This is really the first tool that isn't language dependent.
I use it to scan documents and ask for a summary and some specific questions (which I can verify in that same document), to generate ideas or approaches for writing, and to summarize books that I don't remember 😂
I found it to be not very accurate when analyzing images especially complex images. Also, it is actually very good when using its language capabilities especially when you talk to it It’s super accurate and super fast I believe it’s using its whisper technology for its audio capability
Any of my fellow geezers getting future shocked these days? You get used to the tech over the years, but sometimes you think back and remember just how far that tech has come, and it's mind-blowing.
Imagine what today's kids might have when they get old...
I often use GPT 3.5 to rewrite a text.
Sometimes I want to make a comment on social media, but I worry about being too harsh or biased, then to moderate my text
Sometimes I have written a long text, but I don't expect the "recipients" to want to read a long text, I can ask for a summary, sometimes highlight elements or words that I want to keep.
One of the things I have to do in my public entity auditing job is go through the governing board minutes, sift through all the motions and discussions and memorial proclamations and whatnot to make a summary table of the important stuff that was passed. I’ve been making ChatGPT do the heavy lifting by feeding it OCR’d scans of the minutes and having it spit out a bulleted list of the stuff that was passed, which I can copy-paste into my table and remove what I deem to be unimportant.
I ❤ fluffy towels.
I ❤ lamp.
I ❤️ thiojoe.
Love it when knowledgeable folks like yourself share these tips.
This video really opened my eyes to the practical uses of ChatGPT! I was surprised to learn that it can help break down complicated commands or formats, making tech tasks a lot easier. I never thought about using it to look up specific items by describing them, like that wash bottle example. Plus, the idea of asking it what kind of towels are the fluffiest is pretty cool! It’s a reminder that AI can actually help solve everyday problems, not just write stories. Definitely gonna try using it more for my own needs!
The TV episode thing has been SO helpful for me! The number of times I want to re-Watch or tell someone about an episode where I remember a tiny piece of what happened, and it’ll know exactly which episode I need! That’s something that could be very difficult to find through a normal web search.
I asked both ChatGPT 3.5 and 4 to write steps to create an iOS shortcut to gets todays hourly weather forecast and tell me the hour when temperature is hottest.
I didn’t think that was a hard question but neither 3.5 or 4 could give a working answer. The answers were inaccurate and made me doubt other results from ChatGPT
I use ChatGPT constantly. I use it for studying concepts I've always wanted to learn, for coding questions, helping debug, learning about more certifications as well as career direction. I use it for figuring out medical issues I might be having, advice in ideas I've wanted to code/implement. I'm constantly using it. It's not right 100% of the time, but I'm able to figure out quite a bit from there. It's what got me interested in AI/ML.
This is similar to your examples of using it to look stuff up, but its really helpful for reviewing stuff you used to know but forgot about. When I was writing my PhD thesis I needed to include a chapter about the theoretical motivation for our project, which I hadn't thought about in years since I was focused on the practical data analysis aspects. Just googling things normally it was really hard to find stuff at the right level, I mostly got a mix of 10-minute youtube videos for the general audience or research papers that were much too advanced and specific. But ChatGPT was able to quickly answer a lot of the specific questions I had, and even though it wasn't always super accurate (I was using the free version) it was great for jogging my memory and for spitting out specific names and keywords that I could search for to get some useful references.
ChatGPT is good for pretty much two things:
1. Coding - gives you great suggestions and ideas. Helps when you're stuck. No need to spend hours googling for a solution to a similar problem.
2. Answering questions about topics you know so little about that you're not even sure how to ask. The bottle and star trek examples in the video were exactly that.
13:10 I asked it recently about an idea I had for a novel, and after I outlined the idea asked it if the story seemed similar to other novels that other authors had written, and it responded by saying that my idea appeared to be unique. So ChatGPT might be useful for avoiding accidental plagiarisation. I also use it occasionally to explain medical procedures in layman's terms, and whether a section of medical text indicates that surgery was likely to have been performed. ChatGPT is also useful for explaining the grammar of a foreign language you're learning. I use ChatGPT / Copilot a lot, so I think it's brilliant.
Yes, it is correct, haha. The word "file," we say "arquivo" in Brazilian Portuguese.
In European Portuguese, it is "ficheiro."
Fichier is correct in French too
@@cheeseparis1ah the Latin based languages, such beauties am I right?
@@JAL_EDM Yes, it's Fasciculus in latin
Datei in german is correct too
It's very good for learning about "what if" questions. There's so many things in the world we take for-granted that could be done in entirely different ways. What if we were expected to reuse all of our containers when shopping? What if we had a screwdriver that worked on any screw? What if augmented reality allowed us to entirely replace computer monitors? All of these are things that would be a pain in the ass to research with Google.
Can you please do a video on the LIMITATIONS of this software? I need something to link people to explain what a "hallucination" is and the fact that it does not check its facts, and you are really good at tech communication
Oh, maybe it's useful, I wrote this custom instruction for myself and it turned out to be truly useful to me.
To the "How would you like ChatGPT to respond?" cell.
Based on my questions, always determine which field of expertise deals with the topic of the question, immediately take on the role of the appropriate specialist or group of specialists (think tank, medical council) and generate the quality answer accordingly.
i've been practicing spanish by asking it to give me quotes from a tv show's spanish subtitles for me to translate into english, and about half the time they're real quotes and half the time they're generated nonsense that sounds like a parody of the show, adds a complexity level to the game lol
this is an 8 month old video but honestly i've been doing this the entire year.
I've had chatgpt teach me excel. things that i know excel can do, but don't know the term, and don't really want to look through excel forum comments (though i do also do that). for example asking chatgpt to spit out the average of a column. you might know it, but i don't, but i tested the chatgpt output and it worked. so now i learned more things
I use it a lot for crafting super long Excel VBA code and also advanced Excel formulas. I’ve also used it to create an advanced heat map (png) for a large dataset containing sales trends over the months by the hour of day, it ran python and gave me a 100% accurate heat map and some other advanced graphs too. I use it probably 25% of my work hrs
Ive used it effectively for coding. Even had it build semi complex programs from scratch. If you don't ask it to do the whole thing all at once it usually provides better answers than you can merge from there. Than test and have it debug when necessary.
I use it mainly to write batch files for me. Batch files, command line scripts, python scripts.
That's a good idea you present in the video: Give it a page of code and ask it to insert comments throughout, explaining the code.
I am not a computer person but you are the first chatgpt explainer I understood.
Your latest subscriber 🙏🏽
Agree with "hard to get initially; then easy to verify". I use GPT-4 all day analyzing dev/ops scenarios, creating initial Ansible playbooks, then using my knowledge, skills and abilities to test, adjust, finalize. It's pretty good at getting me pretty close, putting 85%+ of what I need on the page, when guided with the facts of my scenarios and enough specifics.
I do voice acting and manage actors and what translation for all game voicelines they receive. For better experience i have to sort them by audio file names and make it pretty, easy to look at, etc. ChatGPT 4 can literally sort each "voice box" by the audio file number that each contains (as in english line, audio file number, translated line) and it does it. Effortlessly, for me it would take hours and much procrastination.
I've primarily used ChatGPT for two things. The first is discussing science. It's not always accurate, but it can provide multiple different ways to explain a concept. The second main thing I've used it for is helping me brainstorm ideas for stories I'm hoping to make someday. We've only done characterization so far. The AI helped me know what to focus on.
GPT-4 is able to do arithmetic without the wolfram plugin now. It will automatically write and run a Python script to evaluate any non-trivial mathematical calculation even when not specifically instructed to do so.
Wow, I use this tool for so many things! It helps me create stories, write posts for Facebook, and even make comments on UA-cam like this one. It's also great for getting coding help, deciding what to eat for lunch, and learning how to cook new foods. Plus, it's super helpful for setting up Linux with the right packages, understanding how things work, and doing quick reviews. I use it for just about everything I can think of. It even helped me write this comment! 😄
Just starting the video, but taking existing code from another developer and asking chat GPT to comment it has been incredibly helpful for me. Half the comments are too verbose/not needed but more info is rarely an issue compared to none
10:03 yep you covered this, it is a crazy good starting point
I have to disagree about Chat GPT 3.5 which is where I started, and I did correct it a couple times. I never asked silly questions, only things that I didn’t understand or wanted further info about, and very quickly I realised that supplying context first determines how good the responses are, it’s brilliant. I don’t know when it upgraded me to ChatGPT 4, happened without me doing anything, but I use it daily. I absolutely love it!
I tried gemini advanced today, and its context memory is also brain-dead type, 3 replies back, it was able to read my mails summarise them, write replies, and then suddenly it stopped working and keeps forgetting with each new prompt about what i said previously
I use CGPT to summarize books, draft document templates, clean up long social media posts, and explore reverse dictionary options. I also use it to simplify user agreements and contracts into layman's terms, create lists of useful tips, and fit text within character limits. Additionally, I draft wills, eulogies, motions to dismiss traffic tickets, and one-page roommate contracts. I frequently search for trends in economics, well-performing stocks, and more. Plus, I enjoy changing text styles-whether it's Shakespearean, Yoda, slang, or a movie trailer voice. And I used it to clean up this comment before I posted.
Subbed.
Marietta, Ga.
this is actually pretty useful thanks
Actually I read, that the GPT concept with learning from the web becomes problematic as now many texts in the internet are created with aid or by LLMs like Chat GPT. That reinforces not only the good, but also the bad of it.
I'd say it is a great tool, if you know what you want, what you are doing, and know the topic on which you use GPT well enough to detect errors in the result.
When searching Google in my native language, the search is limited only to articles created in my language, which is logical. And to search, you often have to search in English to get more information. In chat-GPT, you can simply ask to answer you in your language, without thinking about what language the information was originally in which he studied. p. s. I even use a neural network to watch this video, which does video voiceover in my language.
The example of the description of keys and options in the ffmpeg transcoding program is very good, I was just about to give it as an example myself. It is very convenient when at first you don't understand them at all, and it is long and not clear to find out their descriptions.
The word file in Ukrainian Файл is indeed correctly translated, because in our case, it is a borrowed word and it is only transliterated.
Thanks! It was great to see your use cases to get more ideas on how to effectively use this tool.
The OCR feature of recognizing handwriting off of your doctors handwriting (if you take a picture of your handwritten notes and upload them). That is pretty neat. I can't program (yet), but it could be cool to automatize that process.
With these programs one does not have to be smart. I am a 80 year old goldsmith and I am using Chat GPT 4o to describe my works💥💥
The formatting somewhat random data is worth the price of admission to me. For about 20 years I have spent about 16 hours once a year to clean up and standardize data for a company. ChatGPT has made it possible to do it in under 2 hours. Verifying is why it taking so long. BTW, the pay is the same.
As a non-native English speaker, ChatGPT is really helpful to me as a 'reverse dictionary,' as you mentioned. It also helps in pointing out grammatical errors, structural errors in sentence formation, simplifying meanings, and things like that.
About incorrect mathematical data, you could ask ChatGPT to write the mathematical data in a programming command such as sql, even python or something and you could execute the program yourself to get the correct answer.
_ _
Things I asked from ChatGPT to write this comment.
1. I word or phrase for person who uses some language as thier second language
2. fix grammar errors in this sentence "As a non native English speaker, ChatGPT is really helpful to me as a 'reverse dictionary' as you mentioned. Also pointing out grammatical errors, structual errors in sentence forming, simplifiying meanings and things like that."
I disagree about 3.5. I’ve got a lot of good usage from it, including formatting, spreadsheet formulae and vCard creation. Why are people so hostile about something, whilst not the apex, still works? Silly thing to say IMO, especially when 4.0 costs £19.99 a month and 3.5 is free.
yeah, using Machine Learning for creative work is like trying to cut a stake with a spoon, you can, but the knife is right there just use the knife & stop trying to make a spork.
but using it to get better ways to ask a search engine how to search stuff, that's a great use & honestly it should be what they actually market it as
It's because you're using ChatGPT and not Bard/Gemini. Bard can actually be quite creative. ChatGPT, not so much.
@@franklingoodwin stop sharpening the Spoons! the knife is right there!!
I like to write random enthusiastic comments to ChatGPT every now and then;
you can get it really hyped up very easily that way.
For example, a simple "Hurray!" and a few positive giddy comments will immediately make it start using emojis, and make it sound very energetic and expressive.
I found a great new recipe by asking for a list of things a moderately competent chef could make in less than an hour. I didn't use it for the recipe, but I found a well rated one myself.
I can't wait until this tech gets more accessible and easier to integrate into systems. It would be pretty incredible if these use cases were baked into their respective services. Imagine an editor that can comment, identify/explain errors, give solutions to those errors, provide optimization suggestions, and even rewrite code in a different language. (I realize some of this already exists to an extent)
I like using AI to feed it a bunch of information then trying to ask it questions. Things that I could spend a few hours researching or just tell the scenario with all the details, then ask the questions I have. Most of the time, answers are 100% correct.
This "free garbage" version is actually many times better than 4.0 which ends the conversation when confronted with something that the creators doesn't want you to talk about because of the ridiculous censorship they implemented.
For example, I asked "why does your earlier version answers correctly to this question and you don't", and got disconnected because it "does not want to talk about this"!
9:07 its correct in turkish, it might mean some other things in other scenarios but there isn't a specific word for it so this is it (and what we use)
I find 3.5 very useful for generating code snippets that does things that I describe verbally, like for example "give me python code that fills a graph with data points at these locations with this symbol and also put in a line with this equation". This saves me the trouble of looking up all the various names of parameters and the exact syntax to do what I need to get done.
English isn't my native language so I use it all the time for help with grammar, to explain words or even translate entire texts. It's brilliant for that if you ask me. I stopped using Google translate a long time ago. And..
No... I didn't use it for this particular text so.. Any errors are my own fault 😅
Something I did that was cool is that I had it devise a multiple choice questionnaire with a key so it would run through the questions, record the answers and then add up the points and provide the final conclusion based on the key.
Hope this helps.
Thanks for all your great videos. I personally use AI tools more than 20 times every single day. Bing AI (ChatGPT) and Gemini (previous "Bard") are my favorites. I often use both for the same tasks as they will generate different answers. Especially if asked programming questions.
i learned a few stuffs here to make my work a lot easier. thanks a lot.
When doing things correctly, you can write a game with Chatgpt, just explain each scenario and don't let IT write it all for you. Like, I imported all my libraries, and pygame or something, how do I set the main window size? How do I create the game idle and game loop, how do I set each object x and y positions, etc and doing it together with Chatgpt is a must try !
I use chatgpt when i am depressed or stressed. It tells me its ok and gives sympathetic advice. This worries me a bit though but i haven't told gpt about this yet.
Watch Usecases of GPT-4
On "Insights with Om"
Try discussing Google's new Gemini, which has just been officially released, can Gemini beat ChatGPT 4? 🤔
yes, these reverse dictionary examples are the kind of thing that i found to be the most useful use of chat-gpt, that has not existed before.
I used it to help improve my resume it offered better advice than any person I had ever talked to
I liked the example of asking what a command prompt option does like with nmap how to specify ports and script and how long you want nmap to take. I also liked the example of where you didn't know the name of an item and it got it right as a wash bottle.
I use it to write perl code. Initially I only tried it for small code segments that saved me having to try and remember syntax or complicated regular expressions. But then I found it could write complete programs that usually worked close to my requirements where I only had to make slight adjustments to fine tune the results. For example I wanted to scrape some data from a web site so I just dumped the web page's source code into the AI and told it what data I wanted extracted. It wrote a perl program that would download the web page, extracted the required data with regular expressions, and then print a formatted report using the extracted data. I was amazed the thing worked first time. I seriously do not understand how these things work.
A lot of useful ones I haven’t tried before. I’ve also started using it to help me write helper functions in the command line
9:00 Wrong output - In portuguese we say file as "ficheiro", "arquivo" means archive.
For Reverse Dictionary, it helped me to find the name of an algorithm. I described how it worked, and it gave me the name, I could even find the entry in Wikipedia and confirm that that's indeed a widespread name for the algorithm. I don't know how I would do it without ChatGPT...
💯Well done, this is the right approach to using ChatGPT. I use it like I’d use a graduate intern assistant - you get it to do the leg-work for you but don’t rely on it to come up with the finished product.
As you say - don’t do ‘mission critical’ stuff straight out of the box - check it first against reality. Great examples by the way…
Great video. Some really cool use cases I did not think of. I use version 4 as well almost every day now. But never for generative text. It's not something I need in my line of work.
I am Australian but a lot of information I get is in US English (i.e. color instead of colour). I use Chat GPT to edit and format my writing but it comes back with spelling mistakes. I then tried telling it to use Australian English but it also tried to use Australian slang (i.e. gday mate). Now I tell it to use Australian spelling but not slang and I get the correct result.
In order to avoid having to post-edit a chat, I have to tell 'it' with EACH question "answers in Oxford-English, pls."
6:35. I’ve found that if a site is too poorly written, it was almost definitely written by a human
Nonsense.
The ffmpeg command at 2:42 is pretty odd to me as GPT-4 decided to add a lot of extra arguments which aren't necessary. The -ar and -ac argument are not needed because ffmpeg will by default not change these parameters from the source file and forcing them to be a different value may then decrease the quality, which you'll probably only want if you want maximum compatibility (like if you're going to use a device which doesn't play well with high sample rates or more than 2 channels of audio). -vn and -f are also completely unnecessary, because ffmpeg will just look at your file extension for the output file that you specified and choose mp3 automagically. And since mp3 can't hold video, it'll just exclude that, no need to specify. Maybe this was necessary with some very old version of ffmpeg, but I don't like that even the best language models out there are making people believe that ffmpeg is this magical piece of hacker terminal software that you first have to learn all the knobs and buttons of to be able to do simple tasks like this in it.