actually, Elsevier even allowed the use of AI in our journal writing. They only demand that we are the one that must be responsible for anything written by the AI, thus making us to thoroughly verify the data given by the AI. That’s it.
And.. how they will judge? how appropriate was our verification? Science was a state of trust, maybe AI will generate more trustful science than humans made historically?🚮
Here's how I use GPT for research so I always have the essence of a paper for quick reference: 1. I export my PDF to a .txt file, then run it through a "RecursiveSummarizer" with my API key. This is available on youtube - use it. 2. I do this once or twice depending on how concise I want the summary to be. 3. I then run the summary through ChatGPT, asking for concise bullet points. 4. I use a custom recursive quoting modification of the recursive summarizer script to pull out what GPT thinks are the most important quotes in a text. This also helps me read more quickly - I can quickly skim through whole articles with their most important quotes in minutes.
@@dukedex5043 that's what you think! It's tested it on a pdf with underlined text. The quotes it picked out corresponded to the underlines more than half of the time!
So ChatGPT extracted this key points of your video: 1. The speaker is impressed with what Chat GPT4 can do and thinks it's perfect for research. 2. Chat GPT4 can provide real references and links to papers in various fields, though not all links may work. 3. Chat GPT4 can analyze a large amount of text and create scaffolds for writing abstracts, presentations, and other research. 4. By analyzing conference presentations, Chat GPT4 can extract important information and turn it into useful outcomes.
I find it helpful to tell it what I want and ask it to rephrase it as a more effective prompt. Then use the prompt to generate the desired response. I get much better results that way.
I love seeing meaningful youtube content on ChatGPT prompting. I have three tips (and a comparison) to offer you @Andy that will help with the ideas you've expressed in this video: First, If you want to guardrail hallucinations, add the instruction "If you do not know the answer, do not speculate." Hallucinations are powerful for creative outcomes but dangerous for research outcomes. This instruction can be layered with many other instructions and appears non ambiguous in multiple context. This is not full proof, I'll admit, because the AI does not know the difference between an answer it tells itself and an answer you tell it, or an answer contained in its training data. But this will guardrail hallucination to a noticeable degree. Second, An extremely simple and concise reverse engineering prompt is: "Rewrite this as a prompt for an AI language model. Write it as an instruction to the model using the template: 'You are [role]. Your goal is [goal]. To achieve this goal, these are the rules: [rules]'" Third, There are multiple critical layers (prompts) you can pass the resulting prompt through to ensure that the prompt retains all critical information while simplifying and standardizing the prompt style so that it is clearly understandable by the AI language model. I'll let you iterate through that. It's a fun journey. Fourth, for comparison, I use "Read and acknowledge the following {text} then await further instructions. text = [text to read]". There are lots of tips around how to improve productivity through the use of punctuation and formatting of the input because of the way that the text is tokenized behind the scenes. It's fascinating stuff. Best wishes in the future of your channel.
The ability to summarise is very attractive to my short attention span and the "use as many words as possible to make sure you know its important" style many like to use. I think most people like to learn but often just skip it all together because the process is often un enjoyable eg. reading a whole book for a week just to learn a couple of important things. Your average person is about to become a whole lot more knowledgeable about a whole lot more things.
I got such a kick out of you telling ChatGPT to read something and to tell you when it's done. 😂 It's instant. You could just say, "Create a scaffold abstract using the below abstracts as templates" and then copy and paste all of the abstracts in the same chat entry.
idk whether anyone already commented this, but I jsut wanted to leave the tip, that it's not actually necessary to tell gpt to "read this and say Read" 🙂 It is very capable to distinguish your instructions from the text you're pasting, as long as you paste it with quotation marks. So you could start your prompt saying "I will be pasting X bellow, I want you to do Y with it." make a new paragraph and paste all the text at once with Quotation marks. Gpt has no problems distinguishing your isntructions from its materials, so that extra step is unecessary and you might save time like that 🙏 * should also say it also doesn't struggle to find additional isntructions BETWEEN pasted texts, as long as you narrate it properly like "pasted text" And here is info on X: "more pasted text" The bellow text I would like you to remember in order to do XYZ: "test" Using all of the texts above, I want you to X, V, Y and ABC. I haven't run into any issues with complex requests on many snippets, as long as they are properly quoted and the isntructions are clear
I had the same problem, even after having pasted 4 emails, chat started to hallucinate. Did you manage to fix it by changing from chatgpt 3.5 to 4, or by providing him small bits of information, one by one?
@@crypto__. I’m using gpt4, and I combined this method with a prompt-refining prompt, getting the model to generate a NEW prompt that would reliably repeat what the video describes
Hi Andy, very interesting video and good job. I'm a resident-physician and have very limited time to do research due to clinical duties. I'm wondering if you would be able to make a video on how to best use GPT-4 or other AI tools to perform data analysis (ex: asking ChatGPT or AI to interpret table/diagram/figure/chart).
I would love to know as well! I haven't tried it yet but I will now that I have seen your post: maybe copy and paste the paper's methods and results section and have it analyze the data?
I haven't had the opportunity to use GPT-4 yet, but I have tried to achieve what you're interested in using GPT-3. Unfortunately, it seems that a natural language model is not able to accurately interpret data. However, it can help with scripting automation (I use R, but any language can be used) to interpret particular types of data. In this regard, ChatGPT has been super helpful to me.
In gpt3.5 it depends on the topic you research for some topics every link is perfect for some its about half (working links but different studies.) And for some topics no links work.
I think ultimately it has to be up to the person writing the prompts to make sure it's accurate information displayed. Just simply doing a few minutes of searching against gpt does wonders on getting the ai on the right track if it is wrong. Also making sure the stuff you are prompting it is within the times of it latest data. You can ask it when it's data cutoff is and referring to anything after that it struggles alot
Just tried yesterday, you can write 2048 letter at once, so you can not upload whole paper at once. And even if you subacribe, you can only write 25 messages in every 3 hours. Just count how many day it takes to upload hundreds of papers and extract specific datas. It’s not available with chatgpt even with version gpt4. And probably it will be much more expensive than $20, if there would be a plug-in. Plug-in would not work with subscription, it will work by token for every usage. And I still do not trust precision of chatgpt, it seems like becoming lazier after every repetitive command, answer’s quality declining
Can't wait for this. Having recently completed a review article, I know the pains of sifting through all that information myself. I learned a lot though
Hi Andy, great video. I’m a researcher myself, and I’m interested in incorporating AI into my daily research tasks. However, my comment is more of an open discussion on how you feel using AI in the ways that you’ve explored in your video. I’ve tried using AI in these ways but I just have this feeling of guilt. Guilt, for using AI knowing that researchers who have come before me didn’t and that I’m somewhat ‘less’ of a legit researcher. I would be interested to hear your thoughts. Thanks!
Not a PhD, but work in pharma, more specifically in validation and compliance. I'm very excited to explore how I can apply ChatGPT and other AI tools to my work. Most of my time is spent drafting risk assessments and protocols for very specific things. I still think it would need a lot of input, but trained properly on ICH guidelines, all the pharmacopoeia and other guidelines. Would still need real time info on system condition, materials used, lab stuff and other limitations. But this could be prompted by the AI given the right instructions. 😊
This is brilliant and so well timed. I'm starting a duel masters in Public Health and Health Administration in two weeks. Slightly shitting myself as I work full time and have two young kids! Chatgpt will hopefully be my saviour 😂
It was a few weeks ago, but I asked it a genetics question, and good a decent answer. I asked for citations and they were VERY realistic looking but the ones I got weren't real. It even had correct journal names and volume/year, and then made up the rest. This may have been 3.5.
there's an easier way to get the text from a UA-cam video without the need for a python program. You just copy the transcript that UA-cam provides directly (which will have lots of time stamps in) and then paste that into chatgpt (even 3.5 will do this) and then ask it to remove all the time stamps and turn it into a continuous piece of text
Imagine when those models gain Theorem Prover capabilities, like COQ, they will be able to actually analyze the logical propositions of those papers by cross examining them and given the laws of nature and logic that we now verify if they are logically valid or not. Engineering will be a walk in the park after this for instance.
I'm thinking this could act very similarly to CAQDAS in generating text from semi-structured interviews, focus group transcriptions, etc. and if it can analyze graphical data, it could easily be used for qualitative research. I imagine you could even feed it information allowing it to adopt a theoretical framework, methodology, etc.
Still amazed at the amount of people who's jobs revolve, in large parts, around typing and don't make any effort in improving their typing. At age 40, that was me. Now I can touch type and it's a massive boost in productivity and well being. Learn to touch type and thank me later
The chatgpt4 window of context is around 8000 words. Which relates to: Academic articles: GPT-4 can process and comprehend most academic articles, including research papers and review articles, which typically range from 3,000 to 10,000 words. Magazine features: GPT-4 can handle in-depth magazine feature articles, such as those found in publications like National Geographic or The New Yorker, which often range from 3,000 to 7,000 words.
You can summarize the conclusions of articles and create a database of information for whatever you want. The possibilities, even with GPT-3.5, are unlimited.
For the lecture notes, once you extract them I would put them in Obsidian and make it all searchable. Be sure to have ChatGTP convert the notes to Markup first.
I'm quite new to this stuff, what exactly do you mean by converting the notes to Markup first? With Obsidian, for Obsidian, at which point, etc? Hope you can clarify :)
Chat GPT told me that if you convert an MP3 audio file into a text file, that it can extract a transcript from the data. I haven't tested it yet, but that's what it told me.
Chat GPT is good for rehashing existing material, but it assumes the information that has been fed to it is correct. If you try it on a more difficult problem, then it isn't so clever. A good one to try are the Apollo missions. Ask it to calculate the weight, trajectory and fuel requirements. Then ask it if the calculated results agree with the particulars given by NASA. Have fun!
@@hasanmiah5181 i disagree I think its way better than the openai one because here it has Access to the internet. So I use BingAI instead of ChatGPT open Ai one
Hi Dr. ANDY I really benefited from ur content. I'm master degree in materials science and my research interests is about incorporating silver nanoparticles in membranes fabrication but I couldn't find a problem or a gap research to bridge, so l'd to get ur help here or anyone with the right experience.
Ask chatgpt about the current research gaps in your area. You may need to do that more than once and ask the questions in different ways till you get what you want.
Thank yoy for posting this video. This is extremely controversial in academia and writing world. Pro and cons are an ongoing discussion. Your video shows an ethical way to use this IA tool. It shows the commands and inputs make the difference. Simple commands may be more effective. For example, You showed great examples of commands a bit wordier, so I wonder if shorter commands would be more efficient. Again I appreciate the way you suggested to use it which contains the ethical components that researchers fear being jeopardize at the use of this new technology
Regarding wordier prompts, non generic prompts help it to make more interesting non generic answers. Hence wordier prompts that detail what the user wants can be beneficial when they are not satisfied with the generic answer.
As a Masters student, I have thoroughly enjoyed your content. I'm looking for a time tracking app that also includes options for goals setting and managing projects. Any recommendations? Maybe a good topic for a video - apps for students?
For gpt 4's analysis of images (not yet out) it was able to solve a university physics problem with a figure as shown in the openAI website. It also understands why memes are funny. It's clear that it can understand an images from a scientific article. I would expect that the probably very expensive 32000 token limit version (maybe even ones with a smaller limit) can explain an article in it's entirety if you feed some definition of terms and maybe even other highly cited articles it references
Be careful with the references, my Bing beta AI (runs on GPT4) gives me mostly real and relevant references, but it mixes them in with "fake" ones when it can't find a reference for some predetermined opinion. It will reference a real paper, but the paper will be irrelevent if you read it, it will not support the data that the AI is claiming it is supporting, it usually won't even mention the piece of data in question. Very subtle things that can mislead you. The bing AI / edge copilot can read your tabs, so if you have the papers open in different tabs, you can just ask it to do things with it. In one instance, I got Bing AI to extract data from the papers across the tabs, standardise the units from the data, and then package the standardised data into a table for easy comparison. It did it perfectly, the math, the data, and the table.
first of all "pip install openai-whisper", an here the code: import whisper from pytube import UA-cam import os def create_and_open_txt(text, filename): # create and write the text to a .txt file with open(filename, "w") as file: file.write(text) os.startfile(filename)
url = input("Enter the UA-cam video URL: ") yt = UA-cam(url) audio_stream = yt.streams.filter(only_audio=True).first() output_path = "UA-camaudios" filename = "audio.mp3" audio_stream.download(output_path=output_path, filename=filename) print(f"Audio downloaded to {output_path}/{filename}") model = whisper.load_model("base") result = model.transcribe("UA-camaudios/audio.mp3") print(result["text"]) create_and_open_txt(result["text"], 'output.txt') For launch after adding the Pythonxx and Pythonxx/Scripts path; go inside the directory of your script and type inside a "cmd" windows: py mon_script.py
Can I let chat gpt wirte about a topic, then change some sentences etc. if necessary. And then search for fitting sources for the parts written. I have problems with writing but i am good at searching. Do you think this would work for a thesis? At the end I would definitely run the paper through a plagiarism scanner.
It's awesome for an already published work. Although, I wonder how safe it is to upload some of my unpublished work and create an abstract or rough draft of the manuscript with this?
It will give you an initial boost thats all. You won't have to start from the scratch but rather you have something to start with. Based on your input it will give you a professional response or writing but in the end you gotta read it and modify it based on your expertise of the field.
Anything you upload to it is now public domain. Anything unpublished or confidential should never be fed into it because you relinquish sole ownership of the information in doing so.
This is very impressive indeed, but there is a difference between doing actual research and producing lots of useless papers just to boost your h-index
Dear Dr based on you explanation I found that chatgpt is like a child that we can train it with correct feed and prompt give it. Please correct or confirm my feedback
I've tried to do a literature review with the free version and I got a lot of fake papers, titles and even authors (dangerous and impressive at the same time haha)...Did anyone else get an improved experience with the GPT-4?
I don't think so, no. I naturally write like an AI apparently because my own write-ups have been detected for 60% AI content 🙂 It hasn't been a problem though (yet).
i tried AI to help me write an article, but for deep level version, it makes me spent more time to check and change the wrong points than originally wrote a new article.
Hi Andy Stapleton, great video on research! Could you kindly share the real links for the mentioned ChatGPT-4 in the video description or comment section? It would be greatly appreciated for easy access to the discussed resources. Thank you!
I am making discoveries in my field which profoundly challenge status quo ideas. I found that ChatGPT4 is like an amalgam of all the worst skeptics in academia. I have to dance around with it for 2-3 hours to bypass its defense of the status quo position. I also found that at the free tier, its citations were entirely fictional. Given this experience with it, I'm not inclined to shell out for the paid tier. It might provide real citations, but I doubt the bias problem is any better. Has anyone else had this problem with it, that it defends the mainstream position on a controversial subject like a bulldog?
Actually it's not really a fictional citation but it uses author/ editor of that article. Meaning that you have to open the website of the article , then you can see the author's/ editor's name
@@farishakim6759 No, it provides entirely made-up citations, which I have followed through on and found to be non-existing. This is being called "hallucinating."
ChatGPT is terrible at math, I tried both 3.5 and 4 and they suck at basic math problems. Other than that, they’re pretty mediocre at things but it tries to get the job done regardless. You have to continually correct it and input additional prompts because it does a lot of things wrong. It’s still pretty cool, and I’m sure in the future it will improve more, seeing that it has developed this quickly already!
actually, Elsevier even allowed the use of AI in our journal writing. They only demand that we are the one that must be responsible for anything written by the AI, thus making us to thoroughly verify the data given by the AI. That’s it.
And.. how they will judge? how appropriate was our verification? Science was a state of trust, maybe AI will generate more trustful science than humans made historically?🚮
Would you send a hyperlink or sth?
Elsevier already took that decision back. Isn't it?
Link of your info pls!
Wouldn’t this inevitably result in a certain percentage of students choosing to not even check their work and skate by on a free C- degree?
Here's how I use GPT for research so I always have the essence of a paper for quick reference:
1. I export my PDF to a .txt file, then run it through a "RecursiveSummarizer" with my API key. This is available on youtube - use it.
2. I do this once or twice depending on how concise I want the summary to be.
3. I then run the summary through ChatGPT, asking for concise bullet points.
4. I use a custom recursive quoting modification of the recursive summarizer script to pull out what GPT thinks are the most important quotes in a text.
This also helps me read more quickly - I can quickly skim through whole articles with their most important quotes in minutes.
Chatgpt doesn't know the importance of something unless it's reading commentaries, number of citations etc.
@@dukedex5043 that's what you think! It's tested it on a pdf with underlined text. The quotes it picked out corresponded to the underlines more than half of the time!
Just did that for this video.
which makes me think the whole practice of publishing academic journal is pointless because no one, real people at least, don't read our papers!
@@freedd.5942 there's just not enough time. Reading consumes hours of my time, and I still feel as if I'm behind in the literature.
Thank you again, Andy. Your videos allowed me to finish my thesis a year in advance
So ChatGPT extracted this key points of your video:
1. The speaker is impressed with what Chat GPT4 can do and thinks it's perfect for research.
2. Chat GPT4 can provide real references and links to papers in various fields, though not all links may work.
3. Chat GPT4 can analyze a large amount of text and create scaffolds for writing abstracts, presentations, and other research.
4. By analyzing conference presentations, Chat GPT4 can extract important information and turn it into useful outcomes.
This all made up. Only GPT-5 will provide real, useful links.
And to think I was about to watch the whole thing
The singularity is near!
Bro can you please tell me the steps of how you achieved this ?
@@elevatewithmettavijnana6170 The end, or singularity, of UA-cam's advertising profit model.
I find it helpful to tell it what I want and ask it to rephrase it as a more effective prompt. Then use the prompt to generate the desired response. I get much better results that way.
I love seeing meaningful youtube content on ChatGPT prompting. I have three tips (and a comparison) to offer you @Andy that will help with the ideas you've expressed in this video:
First, If you want to guardrail hallucinations, add the instruction "If you do not know the answer, do not speculate." Hallucinations are powerful for creative outcomes but dangerous for research outcomes. This instruction can be layered with many other instructions and appears non ambiguous in multiple context. This is not full proof, I'll admit, because the AI does not know the difference between an answer it tells itself and an answer you tell it, or an answer contained in its training data. But this will guardrail hallucination to a noticeable degree.
Second, An extremely simple and concise reverse engineering prompt is:
"Rewrite this as a prompt for an AI language model. Write it as an instruction to the model using the template: 'You are [role]. Your goal is [goal]. To achieve this goal, these are the rules: [rules]'"
Third, There are multiple critical layers (prompts) you can pass the resulting prompt through to ensure that the prompt retains all critical information while simplifying and standardizing the prompt style so that it is clearly understandable by the AI language model. I'll let you iterate through that. It's a fun journey.
Fourth, for comparison, I use "Read and acknowledge the following {text} then await further instructions. text = [text to read]". There are lots of tips around how to improve productivity through the use of punctuation and formatting of the input because of the way that the text is tokenized behind the scenes. It's fascinating stuff.
Best wishes in the future of your channel.
Just started using ChatGPT a few days ago and this Sounds like very solid advice, thank you.
The ability to summarise is very attractive to my short attention span and the "use as many words as possible to make sure you know its important" style many like to use. I think most people like to learn but often just skip it all together because the process is often un enjoyable eg. reading a whole book for a week just to learn a couple of important things. Your average person is about to become a whole lot more knowledgeable about a whole lot more things.
I got such a kick out of you telling ChatGPT to read something and to tell you when it's done. 😂 It's instant. You could just say, "Create a scaffold abstract using the below abstracts as templates" and then copy and paste all of the abstracts in the same chat entry.
idk whether anyone already commented this, but I jsut wanted to leave the tip, that it's not actually necessary to tell gpt to "read this and say Read" 🙂 It is very capable to distinguish your instructions from the text you're pasting, as long as you paste it with quotation marks. So you could start your prompt saying "I will be pasting X bellow, I want you to do Y with it." make a new paragraph and paste all the text at once with Quotation marks. Gpt has no problems distinguishing your isntructions from its materials, so that extra step is unecessary and you might save time like that 🙏
* should also say it also doesn't struggle to find additional isntructions BETWEEN pasted texts, as long as you narrate it properly like
"pasted text"
And here is info on X:
"more pasted text"
The bellow text I would like you to remember in order to do XYZ:
"test"
Using all of the texts above, I want you to X, V, Y and ABC.
I haven't run into any issues with complex requests on many snippets, as long as they are properly quoted and the isntructions are clear
Hi Andy, Thank you for your videos and online material. MS Word is capable of extracting text from audio files and from dictation
You just helped me unlock a great way to keep GPT-4 from hallucinating while I cram a 100,000 word novel I've written into it. Subbed, sir.
I had the same problem, even after having pasted 4 emails, chat started to hallucinate. Did you manage to fix it by changing from chatgpt 3.5 to 4, or by providing him small bits of information, one by one?
@@crypto__. I’m using gpt4, and I combined this method with a prompt-refining prompt, getting the model to generate a NEW prompt that would reliably repeat what the video describes
I bought plus for gpt4 and I can also say it's waaay better overall. It actually feels smart unlike 3 which felt like a robot.
hey I can't seem to find a way to sign-up or subscribe for ChatGPT4, their website does not allow me too.
You can also ask the chat interface of the new Bing (which has gpt 4 ) or Elicit for references from a natural language prompt.
Hi Andy, very interesting video and good job. I'm a resident-physician and have very limited time to do research due to clinical duties. I'm wondering if you would be able to make a video on how to best use GPT-4 or other AI tools to perform data analysis (ex: asking ChatGPT or AI to interpret table/diagram/figure/chart).
I would love to know this too!
Me too
I would love to know as well! I haven't tried it yet but I will now that I have seen your post: maybe copy and paste the paper's methods and results section and have it analyze the data?
Fellow resident-physician here, this is the exact thing I've been waiting for.
I haven't had the opportunity to use GPT-4 yet, but I have tried to achieve what you're interested in using GPT-3. Unfortunately, it seems that a natural language model is not able to accurately interpret data. However, it can help with scripting automation (I use R, but any language can be used) to interpret particular types of data. In this regard, ChatGPT has been super helpful to me.
This is crazy useful...but will definitely need more people to verify the accuracy of the information
In gpt3.5 it depends on the topic you research for some topics every link is perfect for some its about half (working links but different studies.) And for some topics no links work.
I think ultimately it has to be up to the person writing the prompts to make sure it's accurate information displayed. Just simply doing a few minutes of searching against gpt does wonders on getting the ai on the right track if it is wrong. Also making sure the stuff you are prompting it is within the times of it latest data. You can ask it when it's data cutoff is and referring to anything after that it struggles alot
Another suggestion: how AI/chatGPT can help in a systematic review/meta-analysis
Just wait for plugins, it will change the research field
@@NotabilityJa-mh4db why plugins?
Just tried yesterday, you can write 2048 letter at once, so you can not upload whole paper at once.
And even if you subacribe, you can only write 25 messages in every 3 hours.
Just count how many day it takes to upload hundreds of papers and extract specific datas.
It’s not available with chatgpt even with version gpt4. And probably it will be much more expensive than $20, if there would be a plug-in. Plug-in would not work with subscription, it will work by token for every usage.
And I still do not trust precision of chatgpt, it seems like becoming lazier after every repetitive command, answer’s quality declining
@@lucias1276 It connects chatgpt to a specific AI (designed to do certain jobs).😉👍
Can't wait for this. Having recently completed a review article, I know the pains of sifting through all that information myself. I learned a lot though
I'm binge-watching a lot your videos! They are very informative ! Keep doing the excelent work, Andy! Hugs from Brazil!!
Awesome! Thank you!
ANDY, I AM SO EXCITED ABOUT THE IMAGE ANALYSIS!!! I hope it comes super soon!
you have helped it change my life.... ps it does decent recipes when given ingredients and a intention too
i love how you still correct your typos 😍
The million dollar question is if it can synthesize multiple articles into one paragraph
Such a good new!. Thanks for all the amazing content 🤓. Cheers from Bolivia.
Thanks for watching!
I loved your video. Thank you! I am new to ChtGPT and want to play around with it. You gave me a good push!
Im going to use GPT4 to help combine papers about FTL travel to build a framework for warp field propulsion and navigation.
Gpt4 gave me a reading list. It’s links were suspect but the books were spot on
Hi Andy, great video. I’m a researcher myself, and I’m interested in incorporating AI into my daily research tasks. However, my comment is more of an open discussion on how you feel using AI in the ways that you’ve explored in your video. I’ve tried using AI in these ways but I just have this feeling of guilt. Guilt, for using AI knowing that researchers who have come before me didn’t and that I’m somewhat ‘less’ of a legit researcher. I would be interested to hear your thoughts. Thanks!
Not a PhD, but work in pharma, more specifically in validation and compliance. I'm very excited to explore how I can apply ChatGPT and other AI tools to my work.
Most of my time is spent drafting risk assessments and protocols for very specific things.
I still think it would need a lot of input, but trained properly on ICH guidelines, all the pharmacopoeia and other guidelines.
Would still need real time info on system condition, materials used, lab stuff and other limitations. But this could be prompted by the AI given the right instructions.
😊
Thanks for your nice take on this
This is brilliant and so well timed. I'm starting a duel masters in Public Health and Health Administration in two weeks. Slightly shitting myself as I work full time and have two young kids! Chatgpt will hopefully be my saviour 😂
Nice to see you being positive about stuff.
Greetings from Brazil!
I think these are all great ideas, we just need to be wary about what text you're exposing to the OpenAI and the wider internet.
bro this tips are going to increase the productivity of your viewers by several times.
It was a few weeks ago, but I asked it a genetics question, and good a decent answer. I asked for citations and they were VERY realistic looking but the ones I got weren't real. It even had correct journal names and volume/year, and then made up the rest. This may have been 3.5.
I Like this guys „provide..me..with..“ voice.
there's an easier way to get the text from a UA-cam video without the need for a python program. You just copy the transcript that UA-cam provides directly (which will have lots of time stamps in) and then paste that into chatgpt (even 3.5 will do this) and then ask it to remove all the time stamps and turn it into a continuous piece of text
Jane Smith seems to be its favourite researcher. When I gave it inputs from a paper I’d written Jane Smith got the credit there too!
That must be John Doe's wife.
Incredibly, it seems the thinking part is being outsourced and you are an editor.
Can you share the code you used to transcribe the UA-cam video?
I was waiting this video!!
Excellent video on Chat GPT usage and benefits ,Thanks
Imagine when those models gain Theorem Prover capabilities, like COQ, they will be able to actually analyze the logical propositions of those papers by cross examining them and given the laws of nature and logic that we now verify if they are logically valid or not.
Engineering will be a walk in the park after this for instance.
I'm thinking this could act very similarly to CAQDAS in generating text from semi-structured interviews, focus group transcriptions, etc. and if it can analyze graphical data, it could easily be used for qualitative research. I imagine you could even feed it information allowing it to adopt a theoretical framework, methodology, etc.
Or a rubric. That's what I'm doing atm for my graduate capstone.
Thank you for sharing. I was asked to share with my colleagues how to use AI apps for research, I must be honest, I will use more from your videos😂
I like that you are so friendly to the AI and say please with every question. But is it actually necessary?
Still amazed at the amount of people who's jobs revolve, in large parts, around typing and don't make any effort in improving their typing. At age 40, that was me. Now I can touch type and it's a massive boost in productivity and well being. Learn to touch type and thank me later
Thanks Andy !
The chatgpt4 window of context is around 8000 words. Which relates to:
Academic articles: GPT-4 can process and comprehend most academic articles, including research papers and review articles, which typically range from 3,000 to 10,000 words.
Magazine features: GPT-4 can handle in-depth magazine feature articles, such as those found in publications like National Geographic or The New Yorker, which often range from 3,000 to 7,000 words.
Thank you for sharing valuable experience . It really helped me in my Phd journey ❤
You can summarize the conclusions of articles and create a database of information for whatever you want. The possibilities, even with GPT-3.5, are unlimited.
Hi Andy! Really thanks a lot for the tips! I am learning a lot from your videos. Greetings from Brazil! 👏👏👏👏👏
please can you make a guideline video on chatgpt 4.0 plus in scientific rerearch?
For the lecture notes, once you extract them I would put them in Obsidian and make it all searchable. Be sure to have ChatGTP convert the notes to Markup first.
I'm quite new to this stuff, what exactly do you mean by converting the notes to Markup first? With Obsidian, for Obsidian, at which point, etc? Hope you can clarify :)
@@thurlynox Markup probably means markdown, the formatting syntax used by Obsidian
Super interesting! GPT just makes things way more efficient...
Found it funny that you actually say "please"
This app is very very impresive thankyou for this vedio
so envy for the recent phd students :/ shoulda do another one after seeing the recent developments in ai :D coulda so much easier and faster to finish
I LOVE your videos. Thank you so much!
Chat GPT told me that if you convert an MP3 audio file into a text file, that it can extract a transcript from the data. I haven't tested it yet, but that's what it told me.
Great content, subscribed!!
Chat GPT is good for rehashing existing material, but it assumes the information that has been fed to it is correct. If you try it on a more difficult problem, then it isn't so clever. A good one to try are the Apollo missions. Ask it to calculate the weight, trajectory and fuel requirements. Then ask it if the calculated results agree with the particulars given by NASA. Have fun!
Making a huge assumption that the out put is accurate, correct and is actually something new and of value.
This is a game changer. Thanks for sharing! Is it possible to use gpt 4 already without paying?
Not yet. I think it costs $20 US a month...
Microsoft has integrated this chatgpt4 model with their bing search, but it's not very responsive and smooth if u use it normally through openai
@@hasanmiah5181 i disagree I think its way better than the openai one because here it has Access to the internet. So I use BingAI instead of ChatGPT open Ai one
Thank you 😎
Looks like the UA-cam algorithm knows me better than my barber! Is Andy Stapleton the beard or the man behind it? Either way, I'm intrigued.
Hi Dr. ANDY
I really benefited from ur content. I'm master degree in materials science and my research interests is about incorporating silver nanoparticles in membranes fabrication but I couldn't find a problem or a gap research to bridge, so l'd to get ur help here or anyone with the right experience.
Ask chatgpt about the current research gaps in your area. You may need to do that more than once and ask the questions in different ways till you get what you want.
Thank yoy for posting this video. This is extremely controversial in academia and writing world. Pro and cons are an ongoing discussion. Your video shows an ethical way to use this IA tool.
It shows the commands and inputs make the difference. Simple commands may be more effective. For example, You showed great examples of commands a bit wordier, so I wonder if shorter commands would be more efficient.
Again I appreciate the way you suggested to use it which contains the ethical components that researchers fear being jeopardize at the use of this new technology
Regarding wordier prompts, non generic prompts help it to make more interesting non generic answers. Hence wordier prompts that detail what the user wants can be beneficial when they are not satisfied with the generic answer.
As a Masters student, I have thoroughly enjoyed your content. I'm looking for a time tracking app that also includes options for goals setting and managing projects. Any recommendations? Maybe a good topic for a video - apps for students?
Me too
toggl time tracker
For gpt 4's analysis of images (not yet out) it was able to solve a university physics problem with a figure as shown in the openAI website. It also understands why memes are funny. It's clear that it can understand an images from a scientific article. I would expect that the probably very expensive 32000 token limit version (maybe even ones with a smaller limit) can explain an article in it's entirety if you feed some definition of terms and maybe even other highly cited articles it references
I'm hoping to get access to win a jelly bean guessing contest
It will be a god send tools for writing meeting minutes using a recording of it. Boom done in 5 minutes 😅😅
thank you, andy :)
This is great - now that ChatGPT exists I know that at least one “person” will read my paper
Be careful with the references, my Bing beta AI (runs on GPT4) gives me mostly real and relevant references, but it mixes them in with "fake" ones when it can't find a reference for some predetermined opinion. It will reference a real paper, but the paper will be irrelevent if you read it, it will not support the data that the AI is claiming it is supporting, it usually won't even mention the piece of data in question. Very subtle things that can mislead you.
The bing AI / edge copilot can read your tabs, so if you have the papers open in different tabs, you can just ask it to do things with it. In one instance, I got Bing AI to extract data from the papers across the tabs, standardise the units from the data, and then package the standardised data into a table for easy comparison. It did it perfectly, the math, the data, and the table.
Great video ! Many thanX
2 quick questions:
Is your piece of Python code available somewhere ? And how to get a local mp3 file analyzed ? Thanks
first of all "pip install openai-whisper", an here the code:
import whisper
from pytube import UA-cam
import os
def create_and_open_txt(text, filename):
# create and write the text to a .txt file
with open(filename, "w") as file:
file.write(text)
os.startfile(filename)
url = input("Enter the UA-cam video URL: ")
yt = UA-cam(url)
audio_stream = yt.streams.filter(only_audio=True).first()
output_path = "UA-camaudios"
filename = "audio.mp3"
audio_stream.download(output_path=output_path, filename=filename)
print(f"Audio downloaded to {output_path}/{filename}")
model = whisper.load_model("base")
result = model.transcribe("UA-camaudios/audio.mp3")
print(result["text"])
create_and_open_txt(result["text"], 'output.txt')
For launch after adding the Pythonxx and Pythonxx/Scripts path; go inside the directory of your script and type inside a "cmd" windows:
py mon_script.py
@@patrickblot3224 thank you ! Much appreciated
Can I let chat gpt wirte about a topic, then change some sentences etc. if necessary. And then search for fitting sources for the parts written. I have problems with writing but i am good at searching. Do you think this would work for a thesis? At the end I would definitely run the paper through a plagiarism scanner.
I'm going to try taking the cc from a podcast or documentary on UA-cam and ask chat gpt to summarize it
12:00 AQUI PARA ESCREVER
It's awesome for an already published work. Although, I wonder how safe it is to upload some of my unpublished work and create an abstract or rough draft of the manuscript with this?
It will give you an initial boost thats all. You won't have to start from the scratch but rather you have something to start with. Based on your input it will give you a professional response or writing but in the end you gotta read it and modify it based on your expertise of the field.
it is not very safe, the information can be seen by a lot of people and can be used in other chats
Anything you upload to it is now public domain. Anything unpublished or confidential should never be fed into it because you relinquish sole ownership of the information in doing so.
I was wondering that too. Will it appear with plaigerism next time on Turnitin?
This is incredible
If I buy the sub, do I automatically get access to all versions??? Or is there a whole separate route for GTP 4?
Im currently writing my thesis and im really on the fence if i should get it - any opinions either way?
Hello, the video to text part was interesting, could you make a basic tutorial or show how one could do that?
this is awesome!! can you share any suggestions on the code you made! thank you
"Then man made the machine in his own likeness."
This is very impressive indeed, but there is a difference between doing actual research and producing lots of useless papers just to boost your h-index
If you identified the two paragraphs as abstracts in the original prompts you may have gotten different results
Dear Dr based on you explanation I found that chatgpt is like a child that we can train it with correct feed and prompt give it. Please correct or confirm my feedback
Chat Geee Peee Teee is absolutely great❤
i came for the beard. man, it is neat and well groomed.
Fabulous ❤
thx
I've tried to do a literature review with the free version and I got a lot of fake papers, titles and even authors (dangerous and impressive at the same time haha)...Did anyone else get an improved experience with the GPT-4?
no, all 10 references made up by chatgpt 4
Looks like its writing your scripts for you pretty good too
Top man.
Useful for writing but I wouldn't extend that to research yet
Is it ok to use chat GPT to correct English grammar and improve language used when writing manuscripts? Will that be an issue with journals?
I don't think so, no. I naturally write like an AI apparently because my own write-ups have been detected for 60% AI content 🙂 It hasn't been a problem though (yet).
Love your Vids, could you link to the Script you used.cheers
i tried AI to help me write an article, but for deep level version, it makes me spent more time to check and change the wrong points than originally wrote a new article.
Hi Andy Stapleton, great video on research! Could you kindly share the real links for the mentioned ChatGPT-4 in the video description or comment section? It would be greatly appreciated for easy access to the discussed resources. Thank you!
I am making discoveries in my field which profoundly challenge status quo ideas. I found that ChatGPT4 is like an amalgam of all the worst skeptics in academia. I have to dance around with it for 2-3 hours to bypass its defense of the status quo position. I also found that at the free tier, its citations were entirely fictional. Given this experience with it, I'm not inclined to shell out for the paid tier. It might provide real citations, but I doubt the bias problem is any better. Has anyone else had this problem with it, that it defends the mainstream position on a controversial subject like a bulldog?
Actually it's not really a fictional citation but it uses author/ editor of that article. Meaning that you have to open the website of the article , then you can see the author's/ editor's name
@@farishakim6759 No, it provides entirely made-up citations, which I have followed through on and found to be non-existing. This is being called "hallucinating."
ChatGPT is terrible at math, I tried both 3.5 and 4 and they suck at basic math problems. Other than that, they’re pretty mediocre at things but it tries to get the job done regardless. You have to continually correct it and input additional prompts because it does a lot of things wrong. It’s still pretty cool, and I’m sure in the future it will improve more, seeing that it has developed this quickly already!
With plug-ins it solves that problem. wolfram alpha