If you're talking about the Singularity (which you can use an internet to look up), Ray's twenty year old prediction is that he expects it in 2045. He expects AGI in 2029.
Don't be. Many demo's are fake, and even if they aren't, they deliberately somehow prep the model to solve this exact problem - but if you curveball the problem to solve slightly - the model will go haywire.
57:27 were you maybe thinking of "Jevons paradox"? From wikipedia: "technological progress increases the efficiency with which a resource is used (reducing the amount necessary for any one use), but the falling cost of use induces increases in demand enough that resource use is increased, rather than reduced." edit: oh, someone pointed it out in the chat a few minutes after 😂
12:36 pro tip you know you can run prompts serially, generally I have 3-5 prompts of o1 pro running while I’m iterating on it with o1 mini. Then I make the prompt more abstract and just use it regularly. I keep a separate prompt library for thinking models and for sure I’m keeping the subscription!
@ TLDR; I’m rebuilding alot of my prompts to work best with reasoning models. Prompting thinking models is very different from the regular models. So if your used to “prompt engineering” your likely to try o1 and get poor results. So because I’m still learning the right patterns what I do is I slightly modify the prompt with a few different key words and modify the context slightly. Then I compare the results from all the generations.
I liked your video, but you got it wrong about Ray Kurzweil. He never really changed the timeline of his predictions. He said that AGI will be achieved in 2029 and the singularity/intelligence explosion will happen in 2045. a few months ago he even admitted that his predictions are conservative when he sees the current rate of development. He wrote his book in 2005 and made the predictions about 2029 & 2045
Came here to say this. From what I understand we haven’t reached AGI yet, at least publicly. It may be air gapped in some government’s underground Faraday cage somewhere, but we don’t have it. OP also doesn’t seem to understand why it’s called the singularity either. I would think that the idea that we cannot predict how creating something smarter us will change the world wouldn’t be incredibly novel. Dude also needs to watch more sci fi. He apparently missed The Matrix, Terminator, Transcendence, Raised By Wolves, Alien(s), I, Robot, Battlestar Galactica… just off the top of my head.
While LLMs are frozen, you can train them again on newer data. There is a lag but that lag may reduce in time to the point where it is essentially like humans constantly learning.
The guy with the smartest takes on the topic of ai. He has deep knowledge and he doesn't have conflict of interest like Sam Altman, Elon Musk, Jensen Huang. They are gonna always promote their products.
🤣 what deep knowledge, this whole video was just much of guesstimating and maybe could or could nots. The comparison between spread sheets accountants was hilarious, also what was that Ray Kurzweil outburst. Mofo don’t even know the actually prediction
So, stopped watching because he obviously didn't do much looking into why arc is the way it is, or he doesn't understand. The point is every task is novel, and they did better than humans. It was a test that requires the capability to generalize knowledge.
@@JKMcDaniel Yea he missed the whole point, chalking it up as being useless because no one needs it. But the bigger picture it fundamentally improves the AI model’s overall reasoning ability and opens it up to a higher level of capacity for general intelligence. We take for granted, the basic and seemingly simple things because our brains are constantly processing information subconsciously. Compare walking for a human to a robot.
Really appreciate this, an attempt to look into the future based on knowledge and experience. Really valuable. My belief is that as with all new tech everyone will just get more productive. More value, is created faster. Dev, doctors, lawyers, will all have jobs as before, but produce way more. Anything else would be a paradigm shift.
@@fazirultan No, they apply for funding, which has the effect of weighing the value of the solution vs the resources required and determines whether the research goes ahead. Quite literally the opposite of brute forcing imo.
If O3 and whatever else they have behind the scenes is so amazing, why aren't they developing new technologies, algorithms, sythetic materials, methodologies, and so on... patenting them, selling the patents and raising billions of dollars that way?
Regarding “smart but smart” robots, you need to explore Ian M Banks worlds & books and his imaginings on “super AIs” and the great ships/brains and relationships back with humans (and other such life forms).
Gemini is simply the best overall general model -- best price for features, realtime, integrated well into Google ecosystem. I have subscriptions to claude, chat and Gemini
Are you going to take this down and edit it? I hope not because im half an hour in and have to go to sleep, but loving it ( please don't take it down! :D)
Don’t edit it. I watched the whole thing start to finish. Was a good watch on a long drive. I was an AI doomer, sort of, but now I can see the other side of the argument.
So far this video the best honest and evaluation and accurate representation of the current state in AI (30 minutes into the video so far). I would like to make a statement which is going to perhaps trigger some controversy if thought through and taken seriously: ❗❗❗I guarantee ASI can be achieved with only using existing smaller models (GPT-4o- mini, Gemini Flash and if you like Claude Hauki). ❗❗❗ This would be a great discussion I believe, ideally with researched people that oppose the view, because this concept covers how LLMs work and can be used and (HOW TO USE THEM) to achieve outcomes which represent perfect intelligence.
The real silver bullet will be having all of these best-at-some-things-in-some-scenarios models in a pool and having a pre-model model choose which actual models to leverage and dynamically ( so the user doesn't have to even think about it ), negotate the path to a best possible answer which in many cases will be a composite or snowball of output from multiple models. For the subset of cases where o3 ( and others ) will earn its salt, it'll really really pay off, but for many ( possibly most ) cases, it'll be overkill or functionally inferior The company w/ the best combination of pre-model and model-pinball-orchestration ( ghetto term for chain of DRE - discovery, reasonining, and exploration ) will win
9:17 the thing is o1 and o1 pro really do give you better results if you know how to prompt it but the workflow is very different because of the long wait times. The thing a lot of people understand these models are sensitive to the prompt, but to get the very best out of them you really need to spend time crafting the prompt and context. I would really be loosing a lot more than $200 a month without pro now The reason why there is no gpt 5 and it looks like Claude is crushing it is because last year when they fired Sam and this whole think about what illya saw, it was strawberry aka o1. OpenAI gave up on pretraning and focused on the new architecture. They are genuinely going for super intelligence they don’t care about consumer apps. That being said I expect the o3 turbo models to absolutely shock everyone especially with how fast they will get smarter
I’ve not finished the video and I’m going to guess you’ll get to this but I like your comments on AGI. The truth is that we don’t hire a lawyer that is a programmer that is a marketer and a photographer at once. You hire specialists. The best way to reduce cost, reduce compute, increase performance and accuracy is by creating artificial specialized intelligence
[29:17] the thing is, o3 IS A BETTER CODER THAN YOU! and i and all but 150 people in the world at competitive programming! and yes its not real engineering but it IS factually better at that class of programming problems. What you and most people are missing is it didnt get there by training on human data. it REASONED its way there its self and it can distill this into a mush smaller and faster model. the reason why they cant say they beat the benchmark is because of the amount it cost. i will bet everything i have that it will be beaten this summer by a cheap model, and a chinese model you can run on a top spec macbook by the end of the year. But these are NOT the models we will be using for coding or mostly anything consumer facing. the models we will use will be trained by high quality data generated by these models, they will be super fast, super cheap and super accurate and ONLY used for coding.
@@nickwoodward819 yes but claude is not an engineer. its a generally intelligent question answer system. but thats not even the point. say claude 4 is somehow 10x "better" than 3.5 its still not going to be a better engineer than you because its not an engineer its an llm. the claude application is scaffolding around the llm to make it a chatbot. the point is claude 4 CANT get 10x better with pre-training because there is on more data. but because o3 can solve problems "out of distribution" ie its "discovering" solutions which it then uses as data to train better claude/4o style models its a super big deal. the only intuition i can give is think about alpha go and how when it learned from human games it was up there with the average of the best players. then when it could effectively generate its own data, it became super human
@@aaronabuusama You've missed my point: Benchmarks, phds, ASI, hype - it all means nothing if it's not making the tools more accurate for practical daily use. You started with a concrete point ("it's better than you"), and then quickly devolved into theoretcial outcomes ("training itself will make it superhuman"). We'll see, but I wouldn't be surprised if in a year I'm still coding with a model that performs similarly with specific bugs and libraries
@@nickwoodward819 i understand you point, I poorly articulated mine. What i mean by “training its self will make it superhuman” is not theoretical. What I mean more concretely is they way they are gibing the model reasoning is the same method that got this whole thing going in the first place. O3 is doing what AlphaGO is doing for go, its the its unsupervised reinforcement learning that makes AlphaGO superhuman at go. The ONLY thing that’s theoretical here and the assumption ption is that as these thinking models bootstrap up their reasoning abilities, that at some points it will start to display emergent abilities
Great video and perspective. I do have a critic of your analysis though which is the following: I enjoy dunking on openai as much as the next guy and I think being extremely critical is best for the ecosystem. However, with respect to o1, I think you need to use the pro version with respect to grading the models output quality as many have said their is a difference. For customer practicality and shipping a product I totally get the critic
There are so many people out there that criticize AI strongly, and yet have barely used it.😂 Or haven't learned to use it well. Some people are getting incredible results, but that's usually because they are actually making an effort. It seems very much to be a user issue, doesn't it?😮
ASI will understand intelligence, and create new AI as it's way of getting things done. Humans have to train and poke and prod an AI until it approximates what we want. An ASI will write a custom AI the way a skilled programmer writes a simple computer program.
This is a good video and I agree with many of the things you talk about except the job situation and a few other things. The thing with jobs, yes people who will use AI to help them in their job will be valued more then the people who don't. And they will be payed well, but that same person will also replace 10 other people. meaning that overall for a society A LOT lot less people will have a well paying job. What will happen is you will have even a larger devide between the haves and the have nots and that will apply to the people that work as well. Also on the point of the last 1% or 10% gets everyone. I disagree there as well. Slop is everywhere and people are ok with it as long as it gets the job done. Look around you, our lives are filled with mediocre content, products, websites, and everything else. The reason its that way is because people dont care about quality as long as its "good enough". And AI will fill that niche very well. Meaning AI driven companies will do just fine towing the line because as long as they are at least somewhat profitable that's all that matters. And that in itself will put A LOT of businesses out of business. Because most businesses are very mediocre.... Capitalistic society can not survive the AI revolution, we are about to witness some serious turmoil, buckle up Dorothy.
Cant agree more... The nicest way I can say this is Devin is one of the worst investments ive seen that makes no sense to me in terms of what we see today after the months after the investment to do "something".... I hope that they improve quickly after their product launch now because its really hurting the perception of the AI ecosystem in my opinion.
What about the people that just aren't that smart, and AI actually becomes worse when it is directed by them. What would this population do in a world with AI and robotics? As time goes on this population will make up more and more of a percentage of people.
c3po was an etiquette and translation droid. his job did not involve taking actions or making decisions. you’re setting him up as a strawman. r2d2 would be less bad. ray kurzweil has consistently said 2029 for agi.
Costs will come down when GPU competitors scale (eg Groq). I am betting on Open AI’s strategy that he who gets to ASI first will compound exponentially faster than copycats. Will they be useful for day to day use in today’s understanding of the world, perhaps not. But commodity businesses also have lower valuations.
Hard disagree on CSS skills. Like all languages, a large amount of code we write is boilerplate. AI tools have all but solved that problem, but the real work, 60% of a project, begins after that. Responsive CSS is very difficult to automate and I've yet to find a tool, including yours, that does an adequate job that satisfies client's requests. I only see this skill becoming MORE valuable.
yeah... tripled and quadrupled here, the sentiment on o1 pro. You really are prompting it wrong if you aren't getting more of it. I've fed it things like: I have two projects here (which are vaguely related). I would like to take certain parts of the functionality of project1 and implement them over in project2. I provide the source code for project1 and the source code for project2. And for the most part, prompts like this work. It is VERY ANNOYING that this isn't available in API form btw. having to copy code blocks in and out of the damn model via pbcopy and by clicking the 'copy to download' button is just plain irritating.
Thank you for talking about this. The incentive structures are broken beyond repair, creating nothing but noise-from influencers hyping up news to absurd levels for a couple more likes, to companies stating overly optimistic half-truths (not outright lies, since they avoid stating timelines) to boost their short-term stock prices. The whole system has lost its mind. The people who use these AI tools daily have no voice in these discussions, and you seem to be the only influencer here with a rational take on these topics.
If you're talking about the Singularity (which you can use an internet to look up), Ray's twenty year old prediction is that he expects it in 2045. He expects AGI in 2029.
I just watched a demo of someone using o1 to create a data scientist to do kaggle competitions. It was impressive.
Where,I wanna see
Don't be. Many demo's are fake, and even if they aren't, they deliberately somehow prep the model to solve this exact problem - but if you curveball the problem to solve slightly - the model will go haywire.
57:27 were you maybe thinking of "Jevons paradox"? From wikipedia: "technological progress increases the efficiency with which a resource is used (reducing the amount necessary for any one use), but the falling cost of use induces increases in demand enough that resource use is increased, rather than reduced."
edit: oh, someone pointed it out in the chat a few minutes after 😂
12:36 pro tip you know you can run prompts serially, generally I have 3-5 prompts of o1 pro running while I’m iterating on it with o1 mini. Then I make the prompt more abstract and just use it regularly. I keep a separate prompt library for thinking models and for sure I’m keeping the subscription!
out of interest what are you prompting to need that many prompts running concurrently? Is this for coding?
@ TLDR; I’m rebuilding alot of my prompts to work best with reasoning models. Prompting thinking models is very different from the regular models. So if your used to “prompt engineering” your likely to try o1 and get poor results. So because I’m still learning the right patterns what I do is I slightly modify the prompt with a few different key words and modify the context slightly. Then I compare the results from all the generations.
@@aaronabuusama ahhhh very interesting, makes sense
I liked your video, but you got it wrong about Ray Kurzweil. He never really changed the timeline of his predictions. He said that AGI will be achieved in 2029 and the singularity/intelligence explosion will happen in 2045. a few months ago he even admitted that his predictions are conservative when he sees the current rate of development. He wrote his book in 2005 and made the predictions about 2029 & 2045
Came here to say this. From what I understand we haven’t reached AGI yet, at least publicly. It may be air gapped in some government’s underground Faraday cage somewhere, but we don’t have it.
OP also doesn’t seem to understand why it’s called the singularity either. I would think that the idea that we cannot predict how creating something smarter us will change the world wouldn’t be incredibly novel.
Dude also needs to watch more sci fi. He apparently missed The Matrix, Terminator, Transcendence, Raised By Wolves, Alien(s), I, Robot, Battlestar Galactica…
just off the top of my head.
While LLMs are frozen, you can train them again on newer data. There is a lag but that lag may reduce in time to the point where it is essentially like humans constantly learning.
Wow! Thanks for doing this, this is really nice talk on the topic, lot of straight talk
The guy with the smartest takes on the topic of ai. He has deep knowledge and he doesn't have conflict of interest like Sam Altman, Elon Musk, Jensen Huang. They are gonna always promote their products.
🤣 what deep knowledge, this whole video was just much of guesstimating and maybe could or could nots. The comparison between spread sheets accountants was hilarious, also what was that Ray Kurzweil outburst. Mofo don’t even know the actually prediction
So, stopped watching because he obviously didn't do much looking into why arc is the way it is, or he doesn't understand. The point is every task is novel, and they did better than humans. It was a test that requires the capability to generalize knowledge.
@@JKMcDaniel Yea he missed the whole point, chalking it up as being useless because no one needs it. But the bigger picture it fundamentally improves the AI model’s overall reasoning ability and opens it up to a higher level of capacity for general intelligence.
We take for granted, the basic and seemingly simple things because our brains are constantly processing information subconsciously. Compare walking for a human to a robot.
Dude, thank you! Will listen to it on my long drive soon
I have difficulty understanding spoken English, could you enable the option for automatic dubbing? To listen in Brazilian Portuguese.
You are a gem of trustworthy information on AI in coding, thanks.
Really appreciate this, an attempt to look into the future based on knowledge and experience. Really valuable.
My belief is that as with all new tech everyone will just get more productive. More value, is created faster. Dev, doctors, lawyers, will all have jobs as before, but produce way more. Anything else would be a paradigm shift.
I'm just always a little confused by the cost. If the cost comes from large amounts of reasoning, how is it not sort of brute forcing?
because "brute forcing" doesn't sound great to investors
When scientists take a large amount of resources, physically and mentally, to solve a problem, are they brute forcing?
@@fazirultan yes
@@fazirultan No, they apply for funding, which has the effect of weighing the value of the solution vs the resources required and determines whether the research goes ahead. Quite literally the opposite of brute forcing imo.
regenerative medicine is going to come super fast in this i bet
also the hunt for a longevity hack is going to explode.
You don't know where AI is going. You will be obsolete, period.
If O3 and whatever else they have behind the scenes is so amazing, why aren't they developing new technologies, algorithms, sythetic materials, methodologies, and so on... patenting them, selling the patents and raising billions of dollars that way?
Loved every single second of this. Please post more! And congratz on the wedding
Love your stuff Steve! B)
Regarding “smart but smart” robots, you need to explore Ian M Banks worlds & books and his imaginings on “super AIs” and the great ships/brains and relationships back with humans (and other such life forms).
Gemini is simply the best overall general model -- best price for features, realtime, integrated well into Google ecosystem. I have subscriptions to claude, chat and Gemini
Are you going to take this down and edit it? I hope not because im half an hour in and have to go to sleep, but loving it ( please don't take it down! :D)
@@nickwoodward819 no plans to take down and edit 😄
Don’t edit it. I watched the whole thing start to finish. Was a good watch on a long drive. I was an AI doomer, sort of, but now I can see the other side of the argument.
Thanks for this. Really appreciate it
So far this video the best honest and evaluation and accurate representation of the current state in AI (30 minutes into the video so far).
I would like to make a statement which is going to perhaps trigger some controversy if thought through and taken seriously:
❗❗❗I guarantee ASI can be achieved with only using existing smaller models (GPT-4o- mini, Gemini Flash and if you like Claude Hauki). ❗❗❗
This would be a great discussion I believe, ideally with researched people that oppose the view, because this concept covers how LLMs work and can be used and (HOW TO USE THEM) to achieve outcomes which represent perfect intelligence.
I hate to say it, but the 03 Demo if it was close to AGI they’d have had it building an entire app. As it stands, it barely passed a python test.
Yeah WTF was that as the big demo
They never claimed that its AGI
The real silver bullet will be having all of these best-at-some-things-in-some-scenarios models in a pool and having a pre-model model choose which actual models to leverage and dynamically ( so the user doesn't have to even think about it ), negotate the path to a best possible answer which in many cases will be a composite or snowball of output from multiple models.
For the subset of cases where o3 ( and others ) will earn its salt, it'll really really pay off, but for many ( possibly most ) cases, it'll be overkill or functionally inferior
The company w/ the best combination of pre-model and model-pinball-orchestration ( ghetto term for chain of DRE - discovery, reasonining, and exploration ) will win
9:17 the thing is o1 and o1 pro really do give you better results if you know how to prompt it but the workflow is very different because of the long wait times. The thing a lot of people understand these models are sensitive to the prompt, but to get the very best out of them you really need to spend time crafting the prompt and context. I would really be loosing a lot more than $200 a month without pro now
The reason why there is no gpt 5 and it looks like Claude is crushing it is because last year when they fired Sam and this whole think about what illya saw, it was strawberry aka o1. OpenAI gave up on pretraning and focused on the new architecture. They are genuinely going for super intelligence they don’t care about consumer apps.
That being said I expect the o3 turbo models to absolutely shock everyone especially with how fast they will get smarter
I’ve not finished the video and I’m going to guess you’ll get to this but I like your comments on AGI. The truth is that we don’t hire a lawyer that is a programmer that is a marketer and a photographer at once. You hire specialists. The best way to reduce cost, reduce compute, increase performance and accuracy is by creating artificial specialized intelligence
[29:17] the thing is, o3 IS A BETTER CODER THAN YOU! and i and all but 150 people in the world at competitive programming! and yes its not real engineering but it IS factually better at that class of programming problems. What you and most people are missing is it didnt get there by training on human data. it REASONED its way there its self and it can distill this into a mush smaller and faster model. the reason why they cant say they beat the benchmark is because of the amount it cost. i will bet everything i have that it will be beaten this summer by a cheap model, and a chinese model you can run on a top spec macbook by the end of the year.
But these are NOT the models we will be using for coding or mostly anything consumer facing. the models we will use will be trained by high quality data generated by these models, they will be super fast, super cheap and super accurate and ONLY used for coding.
So what? Claude is undoubtedly better at algorithms and complex coding than me, but I'm still the better engineer. Claude is pretty dumb in practice
@@nickwoodward819 yes but claude is not an engineer. its a generally intelligent question answer system. but thats not even the point. say claude 4 is somehow 10x "better" than 3.5 its still not going to be a better engineer than you because its not an engineer its an llm. the claude application is scaffolding around the llm to make it a chatbot.
the point is claude 4 CANT get 10x better with pre-training because there is on more data. but because o3 can solve problems "out of distribution" ie its "discovering" solutions which it then uses as data to train better claude/4o style models
its a super big deal. the only intuition i can give is think about alpha go and how when it learned from human games it was up there with the average of the best players. then when it could effectively generate its own data, it became super human
@@aaronabuusama You've missed my point: Benchmarks, phds, ASI, hype - it all means nothing if it's not making the tools more accurate for practical daily use.
You started with a concrete point ("it's better than you"), and then quickly devolved into theoretcial outcomes ("training itself will make it superhuman").
We'll see, but I wouldn't be surprised if in a year I'm still coding with a model that performs similarly with specific bugs and libraries
@@nickwoodward819 i understand you point, I poorly articulated mine. What i mean by “training its self will make it superhuman” is not theoretical. What I mean more concretely is they way they are gibing the model reasoning is the same method that got this whole thing going in the first place. O3 is doing what AlphaGO is doing for go, its the its unsupervised reinforcement learning that makes AlphaGO superhuman at go.
The ONLY thing that’s theoretical here and the assumption ption is that as these thinking models bootstrap up their reasoning abilities, that at some points it will start to display emergent abilities
Lets be honest its like an army we will be the general and ai is the infantry. Eventually ai can be generals and we will be the president.
the fundamental question remains: How can anyone say what an alternative intelligence more intelligent than them may or may not do?
Great video and perspective. I do have a critic of your analysis though which is the following: I enjoy dunking on openai as much as the next guy and I think being extremely critical is best for the ecosystem. However, with respect to o1, I think you need to use the pro version with respect to grading the models output quality as many have said their is a difference. For customer practicality and shipping a product I totally get the critic
What about an asi present?
Artificial super intelligence, for those wondering
There are so many people out there that criticize AI strongly, and yet have barely used it.😂 Or haven't learned to use it well. Some people are getting incredible results, but that's usually because they are actually making an effort. It seems very much to be a user issue, doesn't it?😮
ASI will understand intelligence, and create new AI as it's way of getting things done.
Humans have to train and poke and prod an AI until it approximates what we want.
An ASI will write a custom AI the way a skilled programmer writes a simple computer program.
What is sonet? how do you spell it?
Claude Sonnet 3.5
This is a good video and I agree with many of the things you talk about except the job situation and a few other things. The thing with jobs, yes people who will use AI to help them in their job will be valued more then the people who don't. And they will be payed well, but that same person will also replace 10 other people. meaning that overall for a society A LOT lot less people will have a well paying job. What will happen is you will have even a larger devide between the haves and the have nots and that will apply to the people that work as well. Also on the point of the last 1% or 10% gets everyone. I disagree there as well. Slop is everywhere and people are ok with it as long as it gets the job done. Look around you, our lives are filled with mediocre content, products, websites, and everything else. The reason its that way is because people dont care about quality as long as its "good enough". And AI will fill that niche very well. Meaning AI driven companies will do just fine towing the line because as long as they are at least somewhat profitable that's all that matters. And that in itself will put A LOT of businesses out of business. Because most businesses are very mediocre.... Capitalistic society can not survive the AI revolution, we are about to witness some serious turmoil, buckle up Dorothy.
I like asking chatgpt questions about coding better, claude gemini perplexity, not as much.
Cant agree more... The nicest way I can say this is Devin is one of the worst investments ive seen that makes no sense to me in terms of what we see today after the months after the investment to do "something".... I hope that they improve quickly after their product launch now because its really hurting the perception of the AI ecosystem in my opinion.
What about the people that just aren't that smart, and AI actually becomes worse when it is directed by them. What would this population do in a world with AI and robotics?
As time goes on this population will make up more and more of a percentage of people.
c3po was an etiquette and translation droid. his
job did not involve taking actions or making decisions. you’re setting him up as a strawman. r2d2 would be less bad. ray kurzweil has consistently said 2029 for agi.
O1 slow? It takes 10 seconds tops
Costs will come down when GPU competitors scale (eg Groq). I am betting on Open AI’s strategy that he who gets to ASI first will compound exponentially faster than copycats. Will they be useful for day to day use in today’s understanding of the world, perhaps not. But commodity businesses also have lower valuations.
This would have been better if you said instead of super smart you said "wicked smaat" in a Boston accent
why build ASI when we should be building HSI (Human Super Intelligence)?
Dude is that actually a potential technology or are you just talking about how dumb humans are right now?
Hard disagree on CSS skills. Like all languages, a large amount of code we write is boilerplate. AI tools have all but solved that problem, but the real work, 60% of a project, begins after that. Responsive CSS is very difficult to automate and I've yet to find a tool, including yours, that does an adequate job that satisfies client's requests. I only see this skill becoming MORE valuable.
I find o1 MUCH more creative than any of my colleagues at work. Just saying...
yeah... tripled and quadrupled here, the sentiment on o1 pro. You really are prompting it wrong if you aren't getting more of it. I've fed it things like: I have two projects here (which are vaguely related). I would like to take certain parts of the functionality of project1 and implement them over in project2. I provide the source code for project1 and the source code for project2. And for the most part, prompts like this work. It is VERY ANNOYING that this isn't available in API form btw. having to copy code blocks in and out of the damn model via pbcopy and by clicking the 'copy to download' button is just plain irritating.
What if o3 is just a mechanical Turk ? Using real humans to validate its decisions. That would explain the slowness and the price
Thank you for talking about this. The incentive structures are broken beyond repair, creating nothing but noise-from influencers hyping up news to absurd levels for a couple more likes, to companies stating overly optimistic half-truths (not outright lies, since they avoid stating timelines) to boost their short-term stock prices. The whole system has lost its mind. The people who use these AI tools daily have no voice in these discussions, and you seem to be the only influencer here with a rational take on these topics.
DUDE YES! I'm so tired of looking at shocked-face thumbnails!
that is why google willl win this
Yeah I totally hated AlphaGo, sure it beat all humans but it was slow and nobody runs it on their laptop. TOTAL FAIL!
7:36 laggy
14:19 😅😮
When criticizing Ray Kurzweil you should perhaps be a bit more specific 🙂
Not all humans are irrelevant, but MANY are - that is undeniable.
I guess you are, right?
every month this guy swallows his words 😂