o3 highly tuned, is around $1000 per command at the moment and this is because of hardware, now we will need to wait 6-18 months for this to come down as NVIDIA and other TPU/CPU manufacturers release products that are optimised to run these models. This is a very small window to get your skills sorted, so don't wait!
@@chromotk1118 AI is a means and not an end, someone has to type in the prompts and put systems together. Databases are still needed, integrations are still required, security and compliance are still important, so is accessibility and of course business continuity. So still lots of considerations before letting an AI lose on running a business. The Enterprise still needs a crew!
Every model had a safety testing team. This time only they're opening it up also to third parties. The other models all had red teaming and safety testing. That is normal.
Let's see, maybe we're going to have a Sora story again. They tease it, it takes ages to release, and in the meantime, Google overtakes them. I really hope that this doesn't occur, but we never know.
Sora was only overtaken by Veo 2 because google has the access to the most videos and can curate the models to better filmography… they own RUclips lol. The only other thing they’re better at is SEO and search engines, everything else ChatGPT got them beat
Some argue that it still struggles with certain simple human tasks. However, I believe this only underscores the fact that its inner workings are fundamentally different from those of humans. These minor issues will not impede its immense potential to drive significant advancements in science and technology. I also believe that humanity may achieve immortality sooner than expected, as the development of artificial superintelligence is progressing far more rapidly than anticipated.
Why are we obsessed with having the AI acting like a human, it does it's job brilliantly as point of fusion between a human brain and technology at large. It's fantastic to team with, otherwise what is the point in making them behave like humans.
@@BoominGamewere not trying ti make it act like a human persay its more about making these systems general in nature and from what we can tell humans are pretty much the only other general intelligence apart from a few animals
I can't believe how fast this is going to be honest we went from reasoners to agents in a fucking month and back to advanced godlike reasoners in another few weeks...
To have true agents doing useful tasks online and the majority of human desk jobs we needed to reach this level of reasoning or even more so (they still need to learn and improve in real time), this development was expected. Stuff will speed up every 3-6 months now instead of yearly. Tons of new models coming!
Yea.... but this has to be explained again and again. Programming, software engineers, developers, whatever you wanna call it .. the act of just TYPING the code isnt that. The AI will TYPE the logic you tell it to type, the logic of the program you designed writting in code. The actual work, the logic, is the actual skill developers are useful for, once you have the logic figured out, whatever size the program is TYPING it is just a mechanical procedure... tedious, long... and prone to errors.... but not "hard". like doing manual math vs a calculator, but we still have mathematicians dont we?.In fact in calculator work NOTHING beats a computer. Because a mathematician is not just a human calculator. A developer isnt just someone who knows the language... a developer is someone who knows how to develop and build the program/aplication/web whatever required for the job, typing that program is just the way to get that info from our brain into the computer.... if AI can do that.. its autocomplete on steroids.
They're crowd sourcing safety testing because they cut their safety team. And while code competition benchmarks are cool, they're still commonly solved problems. We need a benchmark for solving novel problems.
^This. I use GPT4 pretty much on a daily basis for rubber duckying thoughts or to do some grunt tasks. People that think AI is going to create autonomous developers that will just 1:1 replace some engineers on your team, I think are either delusional, or they just build Hello World CRUD APIs with 0 business or domain logic. Next to that, every task seems "code" focused. As if all you do all day as a software engineer is write code. Literally the time spent on meetings, code reviews, writing your deployment logic, syncing with your BA on UATs, provisioning components in your cloud environment, hooking them into your application, etc. etc. etc. is all stuff that all GPT models suck ass at. Even Devin sucks at it. We're nowhere even close to having autonomous replacement for developers. This fearmongering and doomsaying is getting boring.
there is a possibility that 03 could be hosted on cerebras wafer racks to reduce inference costs. my estimates are $130 per command from February 2025. Groqs new hardware set up seems to be taken by anthropic which are expected to release their 03 competitor models in January 2025.
Hey @corbin could you make a video on other people in AI that you follow or people who you think are legit and worth following in the space? I find currently there still to be a lot of hype and “fluffed” channels and also a lot of the “build a $10M app with me in 20 mins”. Feels like you are one of the few who actually are thinking critically and building a sound business model using AI. Would love some recs on others to follow or just about the overall hypeness of the space. Thanks again
03 model ranks 175th! Major shift from thinking AI code is subpar to recognizing its advanced capabilities. Logic-based coding is the future. Wonder what 04 will bring?
OpenAI 03 hitting top 175 on Codeforces is wild! AI coding is next-level now. Can't wait to see what we can build. When do you think OpenAI 04 will drop?
considering the roughly 3 month dif, maybe 3-4 months from now? I'd imagine they'll use that time for red teaming it, then publicly release O3 mini a bit before then.
Or skip to o5(=GPT-5) probably gonna switch back to their main model that has the ability to summon other o-models whenever it needs to and be extremely agentic
havent been impressed by any model's coding capabilities they just do bs and misconception about everything damn instruction. like it will make square wheeled car because you forgot to specify that cars need round wheels, whats even the point.
![Blox Fruits Dragon Rework Update [Full Stream]](http://i.ytimg.com/vi/EqDAp8udhm0/mqdefault.jpg)
The future is how creative you can be
Chat bots
Human level Reasoner
Agents
Human level Innovator <
Organizations
OpenAIs roadmap
Yeah you have to be an innovator yourself and collaborate with AI innovators and teams in the future
👆
@@phen-themoogle7651no
@@phen-themoogle7651no
o3 highly tuned, is around $1000 per command at the moment and this is because of hardware, now we will need to wait 6-18 months for this to come down as NVIDIA and other TPU/CPU manufacturers release products that are optimised to run these models.
This is a very small window to get your skills sorted, so don't wait!
What skill sets do you see evolving further or becoming obsolete as AI becomes more specialized?
@@chromotk1118 AI is a means and not an end, someone has to type in the prompts and put systems together. Databases are still needed, integrations are still required, security and compliance are still important, so is accessibility and of course business continuity. So still lots of considerations before letting an AI lose on running a business. The Enterprise still needs a crew!
@@DavidROliver How long do you think it would take for ai to spit out end-to-end business-grade software needs?
Every model had a safety testing team. This time only they're opening it up also to third parties. The other models all had red teaming and safety testing. That is normal.
Let's see, maybe we're going to have a Sora story again. They tease it, it takes ages to release, and in the meantime, Google overtakes them. I really hope that this doesn't occur, but we never know.
Sora was only overtaken by Veo 2 because google has the access to the most videos and can curate the models to better filmography… they own RUclips lol.
The only other thing they’re better at is SEO and search engines, everything else ChatGPT got them beat
its just about creativity now
Some argue that it still struggles with certain simple human tasks. However, I believe this only underscores the fact that its inner workings are fundamentally different from those of humans. These minor issues will not impede its immense potential to drive significant advancements in science and technology. I also believe that humanity may achieve immortality sooner than expected, as the development of artificial superintelligence is progressing far more rapidly than anticipated.
Why are we obsessed with having the AI acting like a human, it does it's job brilliantly as point of fusion between a human brain and technology at large. It's fantastic to team with, otherwise what is the point in making them behave like humans.
@@BoominGamewere not trying ti make it act like a human persay its more about making these systems general in nature and from what we can tell humans are pretty much the only other general intelligence apart from a few animals
now we need an open source 'o3'
I can't believe how fast this is going to be honest we went from reasoners to agents in a fucking month and back to advanced godlike reasoners in another few weeks...
To have true agents doing useful tasks online and the majority of human desk jobs we needed to reach this level of reasoning or even more so (they still need to learn and improve in real time), this development was expected. Stuff will speed up every 3-6 months now instead of yearly. Tons of new models coming!
Natural language is the future of coding languages. Within a decade we'll be speaking apps into existence.
Much Sooner
Yea.... but this has to be explained again and again. Programming, software engineers, developers, whatever you wanna call it .. the act of just TYPING the code isnt that.
The AI will TYPE the logic you tell it to type, the logic of the program you designed writting in code. The actual work, the logic, is the actual skill developers are useful for, once you have the logic figured out, whatever size the program is TYPING it is just a mechanical procedure... tedious, long... and prone to errors.... but not "hard". like doing manual math vs a calculator, but we still have mathematicians dont we?.In fact in calculator work NOTHING beats a computer. Because a mathematician is not just a human calculator. A developer isnt just someone who knows the language... a developer is someone who knows how to develop and build the program/aplication/web whatever required for the job, typing that program is just the way to get that info from our brain into the computer.... if AI can do that.. its autocomplete on steroids.
They're crowd sourcing safety testing because they cut their safety team.
And while code competition benchmarks are cool, they're still commonly solved problems.
We need a benchmark for solving novel problems.
Its still impressive it can do that , llms are not databases so it shows level of training
^This. I use GPT4 pretty much on a daily basis for rubber duckying thoughts or to do some grunt tasks. People that think AI is going to create autonomous developers that will just 1:1 replace some engineers on your team, I think are either delusional, or they just build Hello World CRUD APIs with 0 business or domain logic. Next to that, every task seems "code" focused. As if all you do all day as a software engineer is write code. Literally the time spent on meetings, code reviews, writing your deployment logic, syncing with your BA on UATs, provisioning components in your cloud environment, hooking them into your application, etc. etc. etc. is all stuff that all GPT models suck ass at. Even Devin sucks at it. We're nowhere even close to having autonomous replacement for developers. This fearmongering and doomsaying is getting boring.
there is a possibility that 03 could be hosted on cerebras wafer racks to reduce inference costs. my estimates are $130 per command from February 2025. Groqs new hardware set up seems to be taken by anthropic which are expected to release their 03 competitor models in January 2025.
Hey @corbin could you make a video on other people in AI that you follow or people who you think are legit and worth following in the space? I find currently there still to be a lot of hype and “fluffed” channels and also a lot of the “build a $10M app with me in 20 mins”.
Feels like you are one of the few who actually are thinking critically and building a sound business model using AI. Would love some recs on others to follow or just about the overall hypeness of the space. Thanks again
I think they are currently training o5.
The 5 in o5 is GPT-5 for sure
03 model ranks 175th! Major shift from thinking AI code is subpar to recognizing its advanced capabilities. Logic-based coding is the future. Wonder what 04 will bring?
OpenAI 03 hitting top 175 on Codeforces is wild! AI coding is next-level now. Can't wait to see what we can build. When do you think OpenAI 04 will drop?
considering the roughly 3 month dif, maybe 3-4 months from now? I'd imagine they'll use that time for red teaming it, then publicly release O3 mini a bit before then.
Or skip to o5(=GPT-5) probably gonna switch back to their main model that has the ability to summon other o-models whenever it needs to and be extremely agentic
But did you see the pricing though? One run can go from $10 to $1000.
03 model's coding leap is impressive. AI reshapes development; time to embrace the change.
GPT 3.5 was way better than me at coding already.
😂😂😂😂😂😂
1:30 im not sure what your argument is here, gpt 4 had 6 months of safety testing...
AI climbs the ranks! The 03 model just made coding less about typing, more about logic. Ready for this revolution?
coding was always about logic, it was never about typing. what are you on about?
What happened to the o2 series ?
O2 is a phone company. They couldn't use it.
🙌
havent been impressed by any model's coding capabilities they just do bs and misconception about everything damn instruction. like it will make square wheeled car because you forgot to specify that cars need round wheels, whats even the point.
What you haven't told is that a more efficient version is $20 per prompt.
*_Google got em so shook they skipped 02 and went straight to 03_* 😂
Glazing google lol 😂😂 it's copyright issues not because of Google you genius
@reddddzzz is that you, Sam?
@@mybocks3 nope not Sam just saying some thing that is obvious
Bye bye human jobs
so companies will rather pay $1000 to run these models which will get cheaper over time so wtf not a good time to be a developer.
no because it isn't done, isn't shipped, isn't validated, and you just made a video on vaporware. congrats for not much.