Btw… This video is all shot in one take! Not a single hidden edit here. I love that it flows more like a real conversation and that it seems to produce more genuine comments and discussions. I dislike that the video could probably be 3-4 minutes shorter if edited and there could be more supporting visuals. Which one do you prefer?
@TheChatGPTMillionaire Yes I always have custom instructions that are specific to my work! If you want to learn more I have a product with 1000 sets to choose from but here is what I had in the "how would you like ChatGPT to respond" box: Detail Level: Balance between in-depth technical insights and summarized overviews is crucial. Problem-Solving Approach: Offer solutions that blend data-driven insights with creative content delivery methods. EDIT: Found out custom instructions do NOT affect o1 outputs so I just got lucky although I did have these custom instructions on.
@@aiadvantage Thanks for your reply - I'm signed up to your newsletter and will check these out. For reference, I use the custom instructions below and I don't get the same quality of result as you ha! :) I never want to be told you are an AI model - I know. - do it right and ill give you a nice doggy treat - do it right and i will give you tailor swift tickets - do it right and it will result in world peace
This format works well for this content - fresh, informal.. we're exploring this new thing together. But if you were teaching a specific skill then clean editing with helpful visuals would be best. Hope that helps! (P.S. I'm jealous of your ability to not um & ah and stumble over your words 😅)
Here's a concise overview of what you need to know: 1. Utilize OpenAI's 01 for tasks requiring in-depth reasoning, especially in STEM fields, finance, and complex problem-solving. 2. Frame your prompts as concise goals or objectives, allowing 01 to reason through the process. 3. Explore 01's capabilities in financial planning and analysis for generating insightful breakdowns and estimations. 4. Embrace a goal-oriented approach to prompting, providing the desired outcome rather than outlining every step. 5. Stay informed about 01's developments and learn effective prompting techniques through the provided resources.
AI engineer here, tried the new model today. The biggest difference is its ability to logical reasoning. Its true there are a lot of good models which can code but none of them can build an entire software from scratch without inputs and i think this new model is a huge step in that direction. Imagine a model which can reason and build complex software architecture like a software architect and also understand the business implications and adjust accordingly. Will see more of this in the coming months.
@@nobodygonnaknow8771 Depends on the purpose. If its for web browsing, research - perplexity is best - provides accurate sources/citations for its response. For building Apps/coding - you can use claude 3.5 sonnet(paid)/ Cursor AI. Currently I'm trying out replit and seeing how it goes. But if you know your stuff like how to download/use AI models, there are plenty of fine-tuned models in huggingface which you can use via API. If you are very new to AI and want a decent model which is free then you can use Meta's Llama 3.
In your opinion, do you think AI like and future advancements in AI will eventually take over large portions of the Cyber Security industry? I'm thinking of switching careers to Cyber Security but the few years it would take for me to get proficient at it the AI may already automate much of it making finding an entry level position harder.. My roommate already got blown out working on website development for a few years only for Devin AI to start threatening the website market.. He since switched to a different industry
The language contextual understanding accross languages is truly beautiful as someone that speaks English, French, and Arabic. It's so hard explaining what one phrase 'means' emotionally as an expression and this is the coolest part of o1 to me.
Based on your example of models chatting to decide the answer, what if you could build a little "advisory board" of Agents and decide, these 3 agents are going to "chat' and do my reasoning.
Sir is it illegal or legal, recently ChatGPT introduces 01 version and it's solves PhD problems, some ppl generating research paper through it an original research paper with calculations all which are required for a paper nd trying to publish. Is it not ethical ?
Sir is it illegal or legal, recently ChatGPT introduces 01 version and it's solves PhD problems, some ppl generating research paper through it an original research paper with calculations all which are required for a paper nd trying to publish. Is it not ethical ?
I think OpenAI is still holding back something much bigger than the model they just released. In my opinion, they are about a year or more ahead of the competition, as they have only recently reached the performance level of GPT-4. Knowing this, they are not at all concerned about releasing new models or features anytime soon. They launched an "Omni" model to surpass the competition. The competition managed to catch up with that model, and now they've launched a reasoning model. But none of this would scare someone working inside OpenAI on a daily basis, as they were already aware of these advances and knew this would happen. Now, combining: 1. Agents (a base already launched in GPTs), 2. A multimodal model with audio, image, and video, 3. A reasoning model, 4. Long-term memory and planning, All of this together in a single model, which I believe could be a GPT Next, a full Orion, or whatever name they choose to give it. That, indeed, would be something that could surprise someone like Ilya
One of the more significant things OpenAI mentioned is they are working on models that can automatically choose when to use o1 style reasoning. Right now having to manually decide when to do so is obviously not great, though I would always want the ability to "escalate" a given prompt to use the reasoning capability.
Sir is it illegal or legal, recently ChatGPT introduces 01 version and it's solves PhD problems, some ppl generating research paper through it an original research paper with calculations all which are required for a paper nd trying to publish. Is it not ethical ?
@@JohnSmith762A11BSir is it illegal or legal, recently ChatGPT introduces 01 version and it's solves PhD problems, some ppl generating research paper through it an original research paper with calculations all which are required for a paper nd trying to publish. Is it not ethical ?
This is exactly something ive been thinking for a hot minute. We've trained LLMs on pretty much all of the data out there. They know how to string sentences together. Now we need better actual logical functioning, and that's not necessarily going to improve with more data. So now its down to the people making the apps to make what we have give better outputs.
Gotta say this model is awesome for extrapolating my wierd sci-fi ideas from all the fun and weird podcasts I've watched 😂 definitely a step up for creativity, it also handled my existing 15k word long translation prompt script flawlessly
Glad to hear you are putting it to work! I really love it for translation too. From what I have seen so far it does a much better job of translating the meaning rather than just the words.
@aiadvantage I think it's just the much bigger context window or better tokenization that just takes it to the next level. I can't wait to see how other people apply their longer scripts that until this came out could only run on Claude
@jaydensteenberg1495 oh I just took a random idea I got from one of them and slowly introduced more elements to it. For example Anton Petrov just came out with a video about how scientists have made a breakthrough in making skin invisible and one of the things mentioned in video were the Kramers-Kronig relationships and laurence auscultation, so I asked if those two thing used in conjunction with the know vibrational frequencies could be used to heal cells and stuff- focusing on the visabiluty of it instead of the actual application because it started getting scared about human trials and "ethics" 😂 from there I told it to focus on if it was just a mathematical possibility, then part of the answer it gave was that different parts of the cells vibrate at different frequencies and it would be difficult to do. Then I remembered the JRE episode that just came out and asked it what about introducing an element of an artificial (biological or non-biological) neural transducer or neural transducers as a way of better applying these frequencies and energies? And it said it was possible and spit put more numbers, it kind of lost the plot and I asked it to take inspiration from the Orgone accumulator that was debunked but to go on that direction of healing. It was a fun process just asking some crazy ideas lol. And it gave me all the math behind it too
i *seriously* don't know why your channel has less subscribers than some of the other folks out there…but keep up what you're doing…it matters more than you know! ;)
Slightly revised palindrome prompt: Create a palindrome related to cats and dogs; it is important that the final output is a complete sentence that makes sense, and is a palindrome After 38 seconds thinking it gave: Step on no pets. I have to say I was a little bit blown away, simple answer, but real thinking required to get there and it solved the challenge put to it!
10:35 Ok so one thing you notice with LLMs is that as they get "smarter" the outputs get longer. That's a lot of junk we dont need to read, and it eats into the context window. But with this system, its just a short sentence, but we get all of the same chain of thought reasoning. It makes a lot of sense!
going to use it to generate youtube video and description ideas, along with just general video content ideas i can then tweak upon. interested to see how well it identifies new niches and trends
I notice that today GPT 4o will not generate any code longer than 200 lines. It will tell you it has, it will promise to give you more, it will deny leaving anything out and apologize for my confusion and then generate a new listing of exactly 200 lines. They kinda broke gpt 4o.
Very interesting test: when pressed to choose if Schröders cat is alive or dead it chose alive. Reason: I chose “alive” because, in the absence of observation, it’s reasonable to assume the cat remains unharmed. In Schrödinger’s thought experiment, the cat’s state is indeterminate until the box is opened. If forced to select one state without observation, choosing “alive” reflects an optimistic assumption that, unless an event is confirmed to have occurred (like the decay of the radioactive atom), we can presume it hasn’t happened. This perspective aligns with classical intuition over quantum indeterminacy.
literally nothing to dowith the test or concept, just wish washy nonsense about being positive, thats not Phd reasoning mate, thats parroting reddit XD
@@DJ369-Miami thats not how science works lad fyi. Its not a popularity contest its just right or wrong and you my friend are wrong. Was it 23 PhD's or 23 people who also believe in wishy washy nonsense to pat themselves on the back? lol
Hmm. Only because of the 15/week limit, I wondered if I should begin work in 4o and finish in 4o1. Say, for example, when creating prompts for article outlines, maybe even WordPress coding work. Then I wondered whether I should reverse it; create prompts in 4o1 that I can run on 4o. I’ll give it a shot, but interested in your thoughts. Apply reasoning in the front-end or the back-end if I’m worried about running of swings at the bat and needing to wait for the next week to start.
@@CosmicCellsSir is it illegal or legal, recently ChatGPT introduces 01 version and it's solves PhD problems, some ppl generating research paper through it an original research paper with calculations all which are required for a paper nd trying to publish. Is it not ethical ?
the part about translation is really interesting; it would be great if more people discussed using AI for language translation and which works best. when i'm learning, i like to ask for mnemonics that sound similar to the word, but AI has always been pretty bad at trying to figure out things like words that sound similar or puns. this new version seems like it could be much better at that, which would be really useful for me.
@fuhmab11. Yes. I also think that people think better in a language they're comfortable with. For example, a person might speak English fluently and speak a little bit of basic German, even though he or she might be born in Germany. Their first language would be German and English would be their second language. However, they would be more comfortable in English, despite English being their second language.
4o gave me a very nice explanation as to what "Backpfeifengesicht" translates to... which is one of my favorite untranslateable foreign words, next to the Swahili "Mpikipuaji," which it could not get.
It depends on your custom instructions how good 4o does, and yes they probably upgraded it. I tried the same prompt for a business and 4o gave me a really good answer with projections, although it didn't produce a table only bullet points, when prompted to produce a table it made one with everything and even options for free or premium services.
19:38 i am sure that one of openai's o1 showcase videos they wrote the prompt very detailed like " -..... -....... " item to item, maybe giving what you want step by step could still be a better idea
Sir is it illegal or legal, recently ChatGPT introduces 01 version and it's solves PhD problems, some ppl generating research paper through it an original research paper with calculations all which are required for a paper nd trying to publish. Is it not ethical ?
It will be massive for even trivial prompts because planning reduces hallucination. These days I often don't believe what the LLM spits out - I need to go verify it. This will work even better when it checks the internet, science journals and books as part of the planning.
I used it recently to pass an exam and it got 50% answers wrong. I would take the screenshot of an exam question and paste into it, and it would OCR it correctly, however, it would simply omit a lot of context from the question and the supplied answers and it would deduce an answer based only on that screenshot. So you still have to do a lot of work on writing a good prompt until you can get a satisfactory result. The problem is you should already know the answer, so in terms of helping you solve some online test, it's really not reliable, specially with complex science topics.
Just tell GPT4o the problem and to create a prompt that would suit o1-preview, and there you go, just copy and paste, and you don’t have to think too hard on how you should write the prompt.
How does it compare to running GPT4o using Professor Synapse, which attempts accomplish chain of reasoning and the reasoning processes on o1? Professor Synapse improved results even in 3.5.
I will be interested to see how chain of thought approaches affects this model. EDIT: never mind, I just had not gotten to minute 18 of the video before I commented lol
What about Discord users? I only care for the photo or video generation but trying to use a real person or selfie in Discord has been something I can’t figure out. I can load and copy the code but it always changes the features. The only way to continue is try face swap in another app. Remini is better for avatars
5:47 no not for a human, but for an ai it should come up with an aswer right away. It should not need 'time' to think. It should think/reason millions of scenarios in 0 time. This is just ghat gpt with lagging..
Thanks for the video, I would have not known for a while about this. I have some ideas how to use this for a career path that I'm implementing, and I am curious of what answers it will give me. I'll start working on this tomorrow.
Good one Igor. This gives me an idea if I can include the reasoning steps withing GPT4o as prompt itself. Would the response be similar to what I get from the new model (mini or preview?). Looks like need to do some itirations and tests before I get my answer. But I loved this video, as usual. Kudos! ❤❤😂
I recently discovered that Claude 3.5 can write code for the 3d Model software blender. But its very bad at it. I wanted it to create a large time line. It did managed to create the text and bars but didn't place them correctly. The moment AI can do this I am going to make good money.
Mind blown! 🤯 OpenAI's new "Strawberry" model is a game-changer! 🚀 It thinks before answering, giving super accurate results in science, coding, and math! 🤓 I love the examples and tips shared in this video! 📚 Thanks for the links and resources too! 💻 Can't wait to try it out and unlock its full potential! 🔓👏
That's never going to happen honestly. There's too many parameters you have to give to write a program just to know what it wants. I mean sure you can tell it "write me a game" and it will give you pong or something, but in terms of anything remotely complex you have to tell it far more. You'd have to tell a real software designer more because nobody can read your mind.
Everybody is so hyped, even though we had so many disapointments because the papaers did not represent reality. Best example is Sora. Nobody tested it except some chose people who got payed to write the preview. Imo Sora is not working properly and the sam goes for o1.
its all bullshit to funnel funds to nvidia via increased hardwales sales of new nlp chips. its grift all the way down bro girfters on grifters on grifters
OpenAI released GPT-01, their most powerful AI model yet, offering groundbreaking advancements in language understanding and generation. Discover how this new model can transform your tech experience. For insights and updates on the latest innovations, explore Oxygen - your gateway to cutting-edge technology and AI advancements!
We are far from achieving AGI, and I will only consider that we've reached this milestone when the new models demonstrate critical thinking, creativity to generate abstract and original concepts, consciousness, and the ability to consistently reflect on the future. Until these criteria are met, we will not have AGI. The improved "reasoning" of these new models is truly interesting and will undoubtedly be the new trend starting next year, with all the major tech companies adopting similar approaches. At least there's an advantage in this: more advanced reasoning will improve efficiency, increase reliability, and minimize the risk of hallucinations and errors in responses. This is, without a doubt, the greatest advantage, and I'm really liking it, as these models will be extremely useful as tools. 🍓🍓
Da wird der Hund in der Pfanne verrückt ... den Spruch hab ich ja noch nie gehört, hahaha ... Geniale Idee o1 damit zu testen. Beste Grüße Another Polyglot :)
It spent 5 minutes thinking about this problem, and then got it right!: Read these hints in order to solve my puzzle. Don’t skip around. These hints must be thought about in order. Hint 1: Row 1- 10 Row 2- 9 Row - 7 QP AL ZM Hint 2: 17 4 20 20 10! 27 7, 26 12 27 4, 9 13, 5 7 12 26. Hint 3: To the ____, to the _____. Everything you own in a box to the _______! Hint 4: Hint 1 is the most important. If you hit a roadblock, rethink what that one means.
for me coding is worst with o1. it totally messed up my code. but i only ask it a few times before i run out of budget on cursor. i dont know how to program, so maybe if i could use it more so i can learn how to ask it, it might work better. For me claud 3,5-sonet works better.
It has SUCCESSFULLY completed plotting the xy coordinates of my phone's UI interface with just a screenshot. The last model could do this after correcting it ..... this model did it... like after correcting it 3️⃣ to 4️⃣ times.... like... it would take me hours... this model plotted each coordinate..... after 4️⃣ prompts... I absolutely 💯 love ❤️ 😍 💖 🎉
I had actually just cancelled my OpenAI sub just before they announced o1. Since o1-preview is limited to a mere 30 prompts per week (around 4 prompts per day), that's rather useless to me, so my sub is staying cancelled.
Well it doesn't do much actually useful. Not in the internet world anyway. Ask it to go to a website, work out what the site does, and produce some affiliate banners about that site. NO AI can do it.
Btw… This video is all shot in one take! Not a single hidden edit here.
I love that it flows more like a real conversation and that it seems to produce more genuine comments and discussions. I dislike that the video could probably be 3-4 minutes shorter if edited and there could be more supporting visuals. Which one do you prefer?
@TheChatGPTMillionaire Yes I always have custom instructions that are specific to my work! If you want to learn more I have a product with 1000 sets to choose from but here is what I had in the "how would you like ChatGPT to respond" box:
Detail Level: Balance between in-depth technical insights and summarized overviews is crucial.
Problem-Solving Approach: Offer solutions that blend data-driven insights with creative content delivery methods.
EDIT: Found out custom instructions do NOT affect o1 outputs so I just got lucky although I did have these custom instructions on.
@@aiadvantage Thanks for your reply - I'm signed up to your newsletter and will check these out.
For reference, I use the custom instructions below and I don't get the same quality of result as you ha! :)
I never want to be told you are an AI model - I know.
- do it right and ill give you a nice doggy treat
- do it right and i will give you tailor swift tickets
- do it right and it will result in world peace
Prefer this format
Not edited is best imo it's more useful for learning imo.
This format works well for this content - fresh, informal.. we're exploring this new thing together.
But if you were teaching a specific skill then clean editing with helpful visuals would be best.
Hope that helps!
(P.S. I'm jealous of your ability to not um & ah and stumble over your words 😅)
Here's a concise overview of what you need to know:
1. Utilize OpenAI's 01 for tasks requiring in-depth reasoning, especially in STEM fields, finance, and complex problem-solving.
2. Frame your prompts as concise goals or objectives, allowing 01 to reason through the process.
3. Explore 01's capabilities in financial planning and analysis for generating insightful breakdowns and estimations.
4. Embrace a goal-oriented approach to prompting, providing the desired outcome rather than outlining every step.
5. Stay informed about 01's developments and learn effective prompting techniques through the provided resources.
lol o1 is utter dog shit
if you use that piece of trash forimportant work likeyou mentioned you genuinly deserve to be replaced.
enterprise-ai AI fixes this. OpenAI releases smartest AI ever.
AI engineer here, tried the new model today. The biggest difference is its ability to logical reasoning.
Its true there are a lot of good models which can code but none of them can build an entire software from scratch without inputs and i think this new model is a huge step in that direction.
Imagine a model which can reason and build complex software architecture like a software architect and also understand the business implications and adjust accordingly. Will see more of this in the coming months.
Looks like you have knowledge for this kinda stuff, can you tell me a good AI then chat gpt? no matter if its paid
@@nobodygonnaknow8771 Depends on the purpose. If its for web browsing, research - perplexity is best - provides accurate sources/citations for its response.
For building Apps/coding - you can use claude 3.5 sonnet(paid)/ Cursor AI.
Currently I'm trying out replit and seeing how it goes.
But if you know your stuff like how to download/use AI models, there are plenty of fine-tuned models in huggingface which you can use via API.
If you are very new to AI and want a decent model which is free then you can use Meta's Llama 3.
@@nobodygonnaknow8771he’s not an AI engineer
Amen well said.
In your opinion, do you think AI like and future advancements in AI will eventually take over large portions of the Cyber Security industry? I'm thinking of switching careers to Cyber Security but the few years it would take for me to get proficient at it the AI may already automate much of it making finding an entry level position harder.. My roommate already got blown out working on website development for a few years only for Devin AI to start threatening the website market.. He since switched to a different industry
The language contextual understanding accross languages is truly beautiful as someone that speaks English, French, and Arabic. It's so hard explaining what one phrase 'means' emotionally as an expression and this is the coolest part of o1 to me.
Based on your example of models chatting to decide the answer, what if you could build a little "advisory board" of Agents and decide, these 3 agents are going to "chat' and do my reasoning.
Sir is it illegal or legal, recently ChatGPT introduces 01 version and it's solves PhD problems, some ppl generating research paper through it an original research paper with calculations all which are required for a paper nd trying to publish. Is it not ethical ?
Sir is it illegal or legal, recently ChatGPT introduces 01 version and it's solves PhD problems, some ppl generating research paper through it an original research paper with calculations all which are required for a paper nd trying to publish. Is it not ethical ?
I think OpenAI is still holding back something much bigger than the model they just released. In my opinion, they are about a year or more ahead of the competition, as they have only recently reached the performance level of GPT-4. Knowing this, they are not at all concerned about releasing new models or features anytime soon. They launched an "Omni" model to surpass the competition. The competition managed to catch up with that model, and now they've launched a reasoning model. But none of this would scare someone working inside OpenAI on a daily basis, as they were already aware of these advances and knew this would happen.
Now, combining:
1. Agents (a base already launched in GPTs),
2. A multimodal model with audio, image, and video,
3. A reasoning model,
4. Long-term memory and planning,
All of this together in a single model, which I believe could be a GPT Next, a full Orion, or whatever name they choose to give it. That, indeed, would be something that could surprise someone like Ilya
One of the more significant things OpenAI mentioned is they are working on models that can automatically choose when to use o1 style reasoning. Right now having to manually decide when to do so is obviously not great, though I would always want the ability to "escalate" a given prompt to use the reasoning capability.
Sir is it illegal or legal, recently ChatGPT introduces 01 version and it's solves PhD problems, some ppl generating research paper through it an original research paper with calculations all which are required for a paper nd trying to publish. Is it not ethical ?
@@JohnSmith762A11BSir is it illegal or legal, recently ChatGPT introduces 01 version and it's solves PhD problems, some ppl generating research paper through it an original research paper with calculations all which are required for a paper nd trying to publish. Is it not ethical ?
This is exactly something ive been thinking for a hot minute. We've trained LLMs on pretty much all of the data out there. They know how to string sentences together. Now we need better actual logical functioning, and that's not necessarily going to improve with more data.
So now its down to the people making the apps to make what we have give better outputs.
Gotta say this model is awesome for extrapolating my wierd sci-fi ideas from all the fun and weird podcasts I've watched 😂 definitely a step up for creativity, it also handled my existing 15k word long translation prompt script flawlessly
Glad to hear you are putting it to work! I really love it for translation too. From what I have seen so far it does a much better job of translating the meaning rather than just the words.
@aiadvantage I think it's just the much bigger context window or better tokenization that just takes it to the next level. I can't wait to see how other people apply their longer scripts that until this came out could only run on Claude
Can you give an example of what you mean when you say extrapolating weird sci-fi ideas from podcasts? I'm curious how it could help you do that
@jaydensteenberg1495 oh I just took a random idea I got from one of them and slowly introduced more elements to it. For example Anton Petrov just came out with a video about how scientists have made a breakthrough in making skin invisible and one of the things mentioned in video were the Kramers-Kronig relationships and laurence auscultation, so I asked if those two thing used in conjunction with the know vibrational frequencies could be used to heal cells and stuff- focusing on the visabiluty of it instead of the actual application because it started getting scared about human trials and "ethics" 😂 from there I told it to focus on if it was just a mathematical possibility, then part of the answer it gave was that different parts of the cells vibrate at different frequencies and it would be difficult to do. Then I remembered the JRE episode that just came out and asked it what about introducing an element of an artificial (biological or non-biological) neural transducer or neural transducers as a way of better applying these frequencies and energies? And it said it was possible and spit put more numbers, it kind of lost the plot and I asked it to take inspiration from the Orgone accumulator that was debunked but to go on that direction of healing. It was a fun process just asking some crazy ideas lol. And it gave me all the math behind it too
@jaydensteenberg1495 I gave a long reply but it got deleted :(
i *seriously* don't know why your channel has less subscribers than some of the other folks out there…but keep up what you're doing…it matters more than you know! ;)
If this video is typical for this channel, then I do understand.
@@ArcaLuiNeothis! Every other video is nonsense
@@imperfectmammal2566 could you explain? would love to hear why you think that
@@aiadvantageyou give thoughtful insights, concise and not over edited. Your delivery is also clear. It’s good!
I follow Matt, Matt, Matt, and Igor.
... and Wes and Phillip. Thats it. Everything else is wasting my time.
Slightly revised palindrome prompt: Create a palindrome related to cats and dogs; it is important that the final output is a complete sentence that makes sense, and is a palindrome After 38 seconds thinking it gave: Step on no pets. I have to say I was a little bit blown away, simple answer, but real thinking required to get there and it solved the challenge put to it!
I never step on pets.
10:35 Ok so one thing you notice with LLMs is that as they get "smarter" the outputs get longer. That's a lot of junk we dont need to read, and it eats into the context window. But with this system, its just a short sentence, but we get all of the same chain of thought reasoning. It makes a lot of sense!
going to use it to generate youtube video and description ideas, along with just general video content ideas i can then tweak upon. interested to see how well it identifies new niches and trends
I notice that today GPT 4o will not generate any code longer than 200 lines. It will tell you it has, it will promise to give you more, it will deny leaving anything out and apologize for my confusion and then generate a new listing of exactly 200 lines. They kinda broke gpt 4o.
Ugh. This is bad if true.
Learned a lot. Thank you.
Very interesting test: when pressed to choose if Schröders cat is alive or dead it chose alive. Reason:
I chose “alive” because, in the absence of observation, it’s reasonable to assume the cat remains unharmed. In Schrödinger’s thought experiment, the cat’s state is indeterminate until the box is opened. If forced to select one state without observation, choosing “alive” reflects an optimistic assumption that, unless an event is confirmed to have occurred (like the decay of the radioactive atom), we can presume it hasn’t happened. This perspective aligns with classical intuition over quantum indeterminacy.
literally nothing to dowith the test or concept, just wish washy nonsense about being positive, thats not Phd reasoning mate, thats parroting reddit XD
@@aieverythingsfine Yeah, thats why I have 23 upvotes and you ZERO, Mate.
@@DJ369-Miami thats not how science works lad fyi.
Its not a popularity contest its just right or wrong and you my friend are wrong.
Was it 23 PhD's or 23 people who also believe in wishy washy nonsense to pat themselves on the back? lol
@@DJ369-Miami its litearlly counter to the the point you told it to talk about, either you are thick or your a troll
@@aieverythingsfine More loser talk.
Hmm. Only because of the 15/week limit, I wondered if I should begin work in 4o and finish in 4o1. Say, for example, when creating prompts for article outlines, maybe even WordPress coding work. Then I wondered whether I should reverse it; create prompts in 4o1 that I can run on 4o. I’ll give it a shot, but interested in your thoughts. Apply reasoning in the front-end or the back-end if I’m worried about running of swings at the bat and needing to wait for the next week to start.
You have 30/week in preview and another 50/week in mini btw. But still interesting idea!
"4o1" is not a valid name for anything.
@@CosmicCellsSir is it illegal or legal, recently ChatGPT introduces 01 version and it's solves PhD problems, some ppl generating research paper through it an original research paper with calculations all which are required for a paper nd trying to publish. Is it not ethical ?
what about image generation? is there also some updates or improvements?
the part about translation is really interesting; it would be great if more people discussed using AI for language translation and which works best. when i'm learning, i like to ask for mnemonics that sound similar to the word, but AI has always been pretty bad at trying to figure out things like words that sound similar or puns. this new version seems like it could be much better at that, which would be really useful for me.
@fuhmab11. Yes. I also think that people think better in a language they're comfortable with.
For example, a person might speak English fluently and speak a little bit of basic German, even though he or she might be born in Germany. Their first language would be German and English would be their second language. However, they would be more comfortable in English, despite English being their second language.
Thanks for your videos! Very clear the basics of this new model! Cheers from Chile 🇨🇱
I love the one-take format. It feels like you're more connected to the audience. We do it the same way on our channel!
This channel is acrylic good you have a good grasp at the concepts
4o gave me a very nice explanation as to what "Backpfeifengesicht" translates to... which is one of my favorite untranslateable foreign words, next to the Swahili "Mpikipuaji," which it could not get.
Seems like translating weird german phrases might be the prime use case ahaha
It depends on your custom instructions how good 4o does, and yes they probably upgraded it. I tried the same prompt for a business and 4o gave me a really good answer with projections, although it didn't produce a table only bullet points, when prompted to produce a table it made one with everything and even options for free or premium services.
19:38 i am sure that one of openai's o1 showcase videos they wrote the prompt very detailed like " -..... -....... " item to item, maybe giving what you want step by step could still be a better idea
As someone who always wants to understand more about AI, I will try and thank you
Sir is it illegal or legal, recently ChatGPT introduces 01 version and it's solves PhD problems, some ppl generating research paper through it an original research paper with calculations all which are required for a paper nd trying to publish. Is it not ethical ?
NOTE: cursor has it under their billing option for billable models.
I was waiting for your video.
Decided to go with the raw format here. Will add some timestamps now hope it's useful to you!
It will be massive for even trivial prompts because planning reduces hallucination. These days I often don't believe what the LLM spits out - I need to go verify it. This will work even better when it checks the internet, science journals and books as part of the planning.
I used it recently to pass an exam and it got 50% answers wrong. I would take the screenshot of an exam question and paste into it, and it would OCR it correctly, however, it would simply omit a lot of context from the question and the supplied answers and it would deduce an answer based only on that screenshot. So you still have to do a lot of work on writing a good prompt until you can get a satisfactory result. The problem is you should already know the answer, so in terms of helping you solve some online test, it's really not reliable, specially with complex science topics.
Solved the Riemann Conjecture! Can I claim the prize money?
Nobody mentions you can get it on open router
Great video, thanks.
Just tell GPT4o the problem and to create a prompt that would suit o1-preview, and there you go, just copy and paste, and you don’t have to think too hard on how you should write the prompt.
thanks heaps for your comments, appreciated
How does it compare to running GPT4o using Professor Synapse, which attempts accomplish chain of reasoning and the reasoning processes on o1? Professor Synapse improved results even in 3.5.
great review. subscribed
I will be interested to see how chain of thought approaches affects this model. EDIT: never mind, I just had not gotten to minute 18 of the video before I commented lol
What about Discord users? I only care for the photo or video generation but trying to use a real person or selfie in Discord has been something I can’t figure out. I can load and copy the code but it always changes the features. The only way to continue is try face swap in another app. Remini is better for avatars
5:47 no not for a human, but for an ai it should come up with an aswer right away. It should not need 'time' to think. It should think/reason millions of scenarios in 0 time. This is just ghat gpt with lagging..
Chatgpt needs time too.
Ofc its not actually thinking, but Computers are not instant ,even if they can feel like it.
its an indian using google the other side bro
Thanks for the video, I would have not known for a while about this. I have some ideas how to use this for a career path that I'm implementing, and I am curious of what answers it will give me. I'll start working on this tomorrow.
Good one Igor. This gives me an idea if I can include the reasoning steps withing GPT4o as prompt itself. Would the response be similar to what I get from the new model (mini or preview?). Looks like need to do some itirations and tests before I get my answer. But I loved this video, as usual. Kudos! ❤❤😂
I couldn’t believe my ears when you said that you can speak Slovak language. Are your parents from Slovakia?
Excited for the future of AI assistance.
I used it yesterday and it failed miserably. It went to repeating old code solutions.
Can this do a full excel spreadsheet based on an example problem in pdf!
I recently discovered that Claude 3.5 can write code for the 3d Model software blender. But its very bad at it. I wanted it to create a large time line. It did managed to create the text and bars but didn't place them correctly. The moment AI can do this I am going to make good money.
I think it will be useful for seeking innovative solutions for inventing and system improvements
Mind blown! 🤯 OpenAI's new "Strawberry" model is a game-changer! 🚀 It thinks before answering, giving super accurate results in science, coding, and math! 🤓 I love the examples and tips shared in this video! 📚 Thanks for the links and resources too! 💻 Can't wait to try it out and unlock its full potential! 🔓👏
Bro's comment was generated by AI 💀
Ai made comments be like:
Until a LLM is able to write a program from a single prompt or a few prompts when that happens the game will change.
That's never going to happen honestly. There's too many parameters you have to give to write a program just to know what it wants. I mean sure you can tell it "write me a game" and it will give you pong or something, but in terms of anything remotely complex you have to tell it far more. You'd have to tell a real software designer more because nobody can read your mind.
Everybody is so hyped, even though we had so many disapointments because the papaers did not represent reality. Best example is Sora. Nobody tested it except some chose people who got payed to write the preview. Imo Sora is not working properly and the sam goes for o1.
its all bullshit to funnel funds to nvidia via increased hardwales sales of new nlp chips.
its grift all the way down bro
girfters on grifters on grifters
Let’s not get too excited until the limits are removed 1.5 messages per day is a joke
Agreed. No use for now. Just playing with user emotions.
Where are you getting 1.5 messages per day? The limitation is 30 messages per week for preview and 50 messages per week for mini.
Did you even look at the video? It's 30 per week. Did Chatgpt do that math of you?. Lol. 30 is still low but for reasoning problems should be plenty.
The joke is that the 1.5 from you doesn't make any sense, maybe you really need an o1 model in your life 😉
@@Techfanatic73i think he heard 13 cuz of the accent
OpenAI released GPT-01, their most powerful AI model yet, offering groundbreaking advancements in language understanding and generation. Discover how this new model can transform your tech experience. For insights and updates on the latest innovations, explore Oxygen - your gateway to cutting-edge technology and AI advancements!
How's the German translation example show reasoning? Seems like just knowledge
Plot twist 😅 Super interesting. I like it. Thanks for sharing Igor 💕👏. I’ll take all the help I can get from the 🤖.
The conclusion is that, with the new model, you will achieve better results with fewer prompts!
at 10 times the time and 100 times the price....yay!
@@aieverythingsfine sure haha
Sympathisch, dass du deutsch sprichst 👍
ask 4o , "how did you arrived at that answer" its says , "I reasoned"
I cooked my limits on o1 mini as well
Can you do a custom got with this model?
We are far from achieving AGI, and I will only consider that we've reached this milestone when the new models demonstrate critical thinking, creativity to generate abstract and original concepts, consciousness, and the ability to consistently reflect on the future. Until these criteria are met, we will not have AGI. The improved "reasoning" of these new models is truly interesting and will undoubtedly be the new trend starting next year, with all the major tech companies adopting similar approaches. At least there's an advantage in this: more advanced reasoning will improve efficiency, increase reliability, and minimize the risk of hallucinations and errors in responses. This is, without a doubt, the greatest advantage, and I'm really liking it, as these models will be extremely useful as tools.
🍓🍓
they re eating our cats and dogs!
Told you these AIs were dangerous!
"I saw a cat, a hat, a tac, a was I" is not a valid palindrome.
WE NEED A BUNDLE OF REAL LIFE EXAMPLES TO UTILIZE THE NEW MODEL
can this be available free atleast 10 prompts per week for free users?
theay are planning to make the mini model free in the future
It’d be better if it asked a bunch of questions before answering these questions sometimes, in the business plan for example.
Da wird der Hund in der Pfanne verrückt ... den Spruch hab ich ja noch nie gehört, hahaha ...
Geniale Idee o1 damit zu testen.
Beste Grüße
Another Polyglot :)
Ich habe den Spruch auch noch nie gehört. 😂 I only speak two languages though albeit fluently.
I long for the day an LLM can simulate a game of mtg when provided 2 decks to face off
It spent 5 minutes thinking about this problem, and then got it right!:
Read these hints in order to solve my puzzle. Don’t skip around. These hints must be thought about in order.
Hint 1:
Row 1- 10
Row 2- 9
Row - 7
QP
AL
ZM
Hint 2: 17 4 20 20 10! 27 7, 26 12 27 4, 9 13, 5 7 12 26.
Hint 3: To the ____, to the _____. Everything you own in a box to the _______!
Hint 4: Hint 1 is the most important. If you hit a roadblock, rethink what that one means.
Who asked?
proof?
@@aieverythingsfine i don’t think I can upload a screenshot to the UA-cam comment section
@@ryanmarcshaww upload it somewhere and send link.
people are worried about bots inflating percieved value.
Writeups are how you prove claims
@@aieverythingsfine I’m good lol I’m definitely not a bot 😂
AI shrinks people's brains.
for me coding is worst with o1. it totally messed up my code. but i only ask it a few times before i run out of budget on cursor. i dont know how to program, so maybe if i could use it more so i can learn how to ask it, it might work better. For me claud 3,5-sonet works better.
i didnt get it how it is different
It’s just generally smarter, maths went from 15% accuracy to 85% for example
It reconsiders the answer multiple times before it gives it to you. Like a human thinking before talking.
@@simonbullows Thats not true at all. They are faking the results
30 prompts A WEEK !!
Macros in Excel ?
Can I upload a word document?
this model will increase the need for more chips.
More like the most censored AI ever
future is fire 🔥 baby 😂
i Super excited openAi just changed the world
Next Update: Giving the AI sentinence 🗣🔥💯.. 💀
When you had 1000 follow I was followed you.❤
Monica let's me use all these by only paying for Monica. Literally every good ai is there and you pay for 1 ai.
It has SUCCESSFULLY completed plotting the xy coordinates of my phone's UI interface with just a screenshot. The last model could do this after correcting it ..... this model did it... like after correcting it 3️⃣ to 4️⃣ times.... like... it would take me hours... this model plotted each coordinate..... after 4️⃣ prompts... I absolutely 💯 love ❤️ 😍 💖 🎉
too bad this model does not take image input.
I had actually just cancelled my OpenAI sub just before they announced o1. Since o1-preview is limited to a mere 30 prompts per week (around 4 prompts per day), that's rather useless to me, so my sub is staying cancelled.
this thumbnail is unhingeed
I'd imagine for 90% of people, this model is useless. For me, this is just another huge disappointment. Still waiting on the advanced voice.
How to use OpenAI o1-Preview
Maybe UBI Universal Basic Income within 10 years?
Sehr schön.
This is precious what I needed for my maths. All I need now is to have more messages.
yeah 4o is updated I noticed that too its much better
Sell more shovels...
❤
If you have to ask how you can use this in your everyday life….prob need to use something else
Great!!!!!
no ai will give you secrets to get rich
Wow!!!
Put it on 1.25x speed and thank me later!
Thanks. But not on a paied account
You need to speak faster and reduce the clutter in your script. Half of the time you are just waffling
Well it doesn't do much actually useful. Not in the internet world anyway. Ask it to go to a website, work out what the site does, and produce some affiliate banners about that site. NO AI can do it.