Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

OpenAI Chief Research Officer, Mark Chen "We've found a new paradigm through Reasoning, which we're also scaling" - GPT-5 could be a significant advancement because we're scaling models not just through unsupervised learning (like with GPT-4.5), but also through a new "reasoning paradigm" - GPT-5 maybe a combination of... show more

Haider.

66,885 subscribers

153,382 views • 1 year ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

12 Comments

Rand1 year ago

so why even release 4.5, just give us 5

Rainmaker2 years ago

Can Machine Learning handle volatile markets? Find out in my latest free Substack post! Explore how a Hidden Markov Model (HMM) can navigate market fluctuations and protect your investments. Full code and practical insights are shared.

Yossi Dahan1 year ago

OAI felt the pressure to drop this teaser after the chaotic GPT-4.5 release - just to signal ‘we’re still in the game.’ Cool, we’ll see about that. Wishing them luck.

Julian Shalaby1 year ago

After GPT 4.5 I have no hope left in OpenAI. I wouldn’t have been impressed with GPT 4.5 in 2023 even

Haider.1 year ago

be patient 👀

Tom Boyle1 year ago

A hybrid of unsupervised learning and reasoning seems crucial for more nuanced models like GPT-5.

Charli1 year ago

Pretty sure pure RL and unsupervised learning are the future. As Google said: SFT memorises RL generalises. Less about making humans happy more about actually learning, like AlphaGo and move 37

Justin1 year ago

Hmm. I don't think they will further scale o3 and release that as part of GPT-5. But I could be wrong. GPT-5 will probably mostly be GPT4o. Maybe even 4o-Mini. Most questions are easy so they will use the easier models on them. For the hard questions, they'll use Reasoners (usually Mini versions). If they architect really well, they will have GPT45 give initial data & info, paired with search, before kicking to another model. What I'm concerned about is they may practically never use o3+Gpt4.5 fully together. Instead, o3 will just use GPT4o. I think it will just be too expensive for the rest of the year. Tbd.

Jo1 year ago

when you scale a model by 10x compute like 4?5 for it to barely be better than 4 and even sometimes worse its not scaling anymore, its time to aknowledge that pretraining hit a wall and its time to find a newer architecture or you're just scamming investors at this point

Jay Ballentine1 year ago

They’ve lost all their credibility after 4.5. We should just ignore them from now on.

Alex Kantrowitz1 year ago

Full interview here for those interested!

Haider.1 year ago

great interview, alex... i watched every tech-based interview from your channel

Related Videos

Sam Altman says AI reasoning is still at the GPT-2 stage but the improvement curve is steep and the new o1 model represents a new paradigm of AI development which will enable rapid progress in capabilities

Sam Altman says AI reasoning is still at the GPT-2 stage but the improvement curve is steep and the new o1 model represents a new paradigm of AI development which will enable rapid progress in capabilities

Tsarathustra

419,003 views • 1 year ago

Dylan Patel says GPT-5 is not a bigger model and did not gain intelligence through scale but through optimization. He highlights how OpenAI improved efficiency by cutting down wasted tokens compared to earlier o-series models. He adds that GPT-5’s new router system decides whether to use the base model, mini, or thinking, leading to faster and more efficient results. He points out that o3 could think for around 30 seconds, while GPT-5 only thinks for 5–10 seconds, meaning less compute is used per query.

Dylan Patel says GPT-5 is not a bigger model and did not gain intelligence through scale but through optimization. He highlights how OpenAI improved efficiency by cutting down wasted tokens compared to earlier o-series models. He adds that GPT-5’s new router system decides whether to use the base model, mini, or thinking, leading to faster and more efficient results. He points out that o3 could think for around 30 seconds, while GPT-5 only thinks for 5–10 seconds, meaning less compute is used per query.

Wes Roth

87,396 views • 11 months ago

OPENAI TEAM: GPT-5 THINKS LIKE A TEAM OF PhDs "ChatGPT’s ability to think deeply through complex problems is now built into GPT-5. It will automatically think whenever needed, delivering more comprehensive, accurate, and detailed answers. It’s like having a team of PhDs in your pocket." Source: OpenAI

OPENAI TEAM: GPT-5 THINKS LIKE A TEAM OF PhDs "ChatGPT’s ability to think deeply through complex problems is now built into GPT-5. It will automatically think whenever needed, delivering more comprehensive, accurate, and detailed answers. It’s like having a team of PhDs in your pocket." Source: OpenAI

Mario Nawfal

63,576 views • 11 months ago

Update on the new reasoning popover in ChatGPT web app prompt composer - there's now even a keyboard shortcut to cycle through reasoning levels, and it looks like these levels correspond to "Quick" (low) = GPT-4o, "Think a little" (medium) = o3-mini, and "Think harder" (high) = o3-mini-high

Update on the new reasoning popover in ChatGPT web app prompt composer - there's now even a keyboard shortcut to cycle through reasoning levels, and it looks like these levels correspond to "Quick" (low) = GPT-4o, "Think a little" (medium) = o3-mini, and "Think harder" (high) = o3-mini-high

Tibor Blaho

63,466 views • 1 year ago

Introducing GPT-Realtime-2 in the API: our most intelligent voice model yet, bringing GPT-5-class reasoning to voice agents. Voice agents are now real-time collaborators that can listen, reason, and solve complex problems as conversations unfold. Now available in the API alongside streaming models GPT-Realtime-Translate and GPT-Realtime-Whisper — a new set of audio capabilities for the next generation of voice interfaces.

Introducing GPT-Realtime-2 in the API: our most intelligent voice model yet, bringing GPT-5-class reasoning to voice agents. Voice agents are now real-time collaborators that can listen, reason, and solve complex problems as conversations unfold. Now available in the API alongside streaming models GPT-Realtime-Translate and GPT-Realtime-Whisper — a new set of audio capabilities for the next generation of voice interfaces.

OpenAI

3,653,870 views • 2 months ago

OpenAI Co-founder Andrej Karpathy explains the new computing paradigm: "We're entering a new computing paradigm with large language models acting like CPUs, using tokens instead of bytes, and having a context window instead of RAM. This is the Large Language Model OS (LMOS)"

OpenAI Co-founder Andrej Karpathy explains the new computing paradigm: "We're entering a new computing paradigm with large language models acting like CPUs, using tokens instead of bytes, and having a context window instead of RAM. This is the Large Language Model OS (LMOS)"

Haider.

552,382 views • 2 years ago

GPT-5 on Sudoku-Bench 🧩 Since releasing Sudoku-Bench in May 2025, when no LLM could solve a classic 9x9 puzzle, we've been evaluating the latest generation of models. GPT-5 now leads our leaderboard with 33% puzzles solved--approximately 2x the previous leader--and is the first LLM we've tested to solve a 9x9 Sudoku variant. However, with 67% of the much harder puzzles remaining unsolved, Sudoku-Bench continues to present significant challenges for AI reasoning. Modern Sudoku variants require models to first understand novel rulesets through meta-reasoning, then maintain global consistency across long reasoning chains. Our experiments with GRPO fine-tuning on Qwen2.5-7b and "Thought Cloning" (training on expert human reasoning from Cracking the Cryptic) show that current approaches still struggle with the spatial reasoning and creative "break-in" points that human solvers use naturally. We believe new approaches are required to solve our benchmark. These results highlight persistent gaps between computational problem-solving and human-like reasoning, particularly in tasks requiring integrated mathematical logic, spatial awareness, and creative insight. Read more about our update here: 🔗 Blogpost →

GPT-5 on Sudoku-Bench 🧩 Since releasing Sudoku-Bench in May 2025, when no LLM could solve a classic 9x9 puzzle, we've been evaluating the latest generation of models. GPT-5 now leads our leaderboard with 33% puzzles solved--approximately 2x the previous leader--and is the first LLM we've tested to solve a 9x9 Sudoku variant. However, with 67% of the much harder puzzles remaining unsolved, Sudoku-Bench continues to present significant challenges for AI reasoning. Modern Sudoku variants require models to first understand novel rulesets through meta-reasoning, then maintain global consistency across long reasoning chains. Our experiments with GRPO fine-tuning on Qwen2.5-7b and "Thought Cloning" (training on expert human reasoning from Cracking the Cryptic) show that current approaches still struggle with the spatial reasoning and creative "break-in" points that human solvers use naturally. We believe new approaches are required to solve our benchmark. These results highlight persistent gaps between computational problem-solving and human-like reasoning, particularly in tasks requiring integrated mathematical logic, spatial awareness, and creative insight. Read more about our update here: 🔗 Blogpost →

Sakana AI

154,645 views • 8 months ago

📈 Today, we're launching Agent Mode in Excel on Windows and Mac. An Excel Copilot that can work with you and make edits in your spreadsheet like an expert collaborator. We're also introducing a new, multi-model system that supports both GPT models from OpenAI and Claude models from Anthropic. Wanna try Claude Opus 4.5 natively in Microsoft Excel? We've got you. Give it a shot and let me know what you think.

📈 Today, we're launching Agent Mode in Excel on Windows and Mac. An Excel Copilot that can work with you and make edits in your spreadsheet like an expert collaborator. We're also introducing a new, multi-model system that supports both GPT models from OpenAI and Claude models from Anthropic. Wanna try Claude Opus 4.5 natively in Microsoft Excel? We've got you. Give it a shot and let me know what you think.

Trevor O'Brien

74,821 views • 6 months ago

.The Information Reporter Stephanie Palazzolo on OpenAI's GPT-5—a major step up for coding and what it means for competitors. "GPT-5 is a step up on a lot of different domains...one area that really stood out to [sources] was with coding." "Not only is GPT-5 better on more academic...tasks, but it's also better on the more practical programming tasks...working with very large and complex code bases." "If GPT-5 is going to be significantly better at these more practical everyday programming tasks, that could prove to be bad news for Anthropic." Watch the full episode on

.The Information Reporter Stephanie Palazzolo on OpenAI's GPT-5—a major step up for coding and what it means for competitors. "GPT-5 is a step up on a lot of different domains...one area that really stood out to [sources] was with coding." "Not only is GPT-5 better on more academic...tasks, but it's also better on the more practical programming tasks...working with very large and complex code bases." "If GPT-5 is going to be significantly better at these more practical everyday programming tasks, that could prove to be bad news for Anthropic." Watch the full episode on

The Information

52,067 views • 1 year ago

OpenAI CPO Kevin Weil says model confusion is a feature, not a bug. 4.1 codes better. 4o chats better. “We're trying to build new capabilities rapidly.. and co-evolve with society.” GPT-5 is where it all starts to converge.

OpenAI CPO Kevin Weil says model confusion is a feature, not a bug. 4.1 codes better. 4o chats better. “We're trying to build new capabilities rapidly.. and co-evolve with society.” GPT-5 is where it all starts to converge.

vitrupo

85,987 views • 1 year ago

New episode with Sarah Wang and christina kim & Isa Fulford of OpenAI breaking down GPT-5, Deep Research, and ChatGPT Agent. Where it improved, what it unlocks, the decisions behind it. A clear look at how these models actually get built.

New episode with Sarah Wang and christina kim & Isa Fulford of OpenAI breaking down GPT-5, Deep Research, and ChatGPT Agent. Where it improved, what it unlocks, the decisions behind it. A clear look at how these models actually get built.

Erik Torenberg

107,126 views • 11 months ago

How does GPT-5 become GPT-6? Everything on the internet was already trained on a year, year and a half ago. New data is being added but it's marginal and a lot of it is AI slop. There is interesting private data (like JP Morgan's data) but nobody is opening that up. You don't go from GPT-5 to GPT-6 in isolation because that's not how learning works. Every learning that has ever occurred comes from having information to learn. You can't learn no information. cc Ground Zero / pod with Curtis from Handshake AI.

How does GPT-5 become GPT-6? Everything on the internet was already trained on a year, year and a half ago. New data is being added but it's marginal and a lot of it is AI slop. There is interesting private data (like JP Morgan's data) but nobody is opening that up. You don't go from GPT-5 to GPT-6 in isolation because that's not how learning works. Every learning that has ever occurred comes from having information to learn. You can't learn no information. cc Ground Zero / pod with Curtis from Handshake AI.

himanshu

10,684 views • 2 months ago

Sam Altman says GPT-3 was the first real glimpse of passing a spiritual Turing test GPT-5 shows the first small signs of AI doing new science — useful ideas, real help on papers GPT-6 will be a leap like GPT-3 to 4, but this time for science, where it can really do it

Sam Altman says GPT-3 was the first real glimpse of passing a spiritual Turing test GPT-5 shows the first small signs of AI doing new science — useful ideas, real help on papers GPT-6 will be a leap like GPT-3 to 4, but this time for science, where it can really do it

Haider.

163,642 views • 8 months ago

The DEPRESSING reality of AI adoption curves Advanced AI just broke into it's third major paradigm since launching. Paradigm One was the simple autocomplete engine. GPT-2 and original GPT-3 were glorified autocomplete tools, just "next token predictors." Paradigm 1.5 was when we added "instruct-aligned" GPT-3, which set the stage for Paradigm Two. Paradigm Two was the chatbot era, with ChatGPT being the front-runner. Paradigm 2.5 was when we started adding reasoning, tool use, and RAG, which set the stage for agentic abilities (Paradigm 3). OpenClaw just blew the lid off Paradigm 3, and we're just at the beginning of this new ramp-up in capabilities. The thing is... each of these paradigm shifts creates fundamentally different UX and technical affordances. While most Fortune 500 companies are still struggling to figure out the cyber security, legal, and financial risks of chatbots, the industry is going whole hog into autonomous agents. And many people will try to graft their understanding of chatbots onto agents. But that's like trying to compare the electric lightbulb to the electric motor.... ...yes they both ran on electricity, but their uses, affordances, limitations, and risks were fundamentally different. Yes, autocomplete, chatbots, and agents all run on "next token prediction" but that's like saying "it all runs on electrons." My goal today is to help give you a better intuition as to why adoption is so slow and to give you a new reference frame to understand that agents are a phase change from chatbots. But also... the state and big companies are going to move depressingly slow on all of this...

The DEPRESSING reality of AI adoption curves Advanced AI just broke into it's third major paradigm since launching. Paradigm One was the simple autocomplete engine. GPT-2 and original GPT-3 were glorified autocomplete tools, just "next token predictors." Paradigm 1.5 was when we added "instruct-aligned" GPT-3, which set the stage for Paradigm Two. Paradigm Two was the chatbot era, with ChatGPT being the front-runner. Paradigm 2.5 was when we started adding reasoning, tool use, and RAG, which set the stage for agentic abilities (Paradigm 3). OpenClaw just blew the lid off Paradigm 3, and we're just at the beginning of this new ramp-up in capabilities. The thing is... each of these paradigm shifts creates fundamentally different UX and technical affordances. While most Fortune 500 companies are still struggling to figure out the cyber security, legal, and financial risks of chatbots, the industry is going whole hog into autonomous agents. And many people will try to graft their understanding of chatbots onto agents. But that's like trying to compare the electric lightbulb to the electric motor.... ...yes they both ran on electricity, but their uses, affordances, limitations, and risks were fundamentally different. Yes, autocomplete, chatbots, and agents all run on "next token prediction" but that's like saying "it all runs on electrons." My goal today is to help give you a better intuition as to why adoption is so slow and to give you a new reference frame to understand that agents are a phase change from chatbots. But also... the state and big companies are going to move depressingly slow on all of this...

David Shapiro (L/0)

17,266 views • 5 months ago

🚨 BREAKING: OpenAI just launched ChatGPT Agent. ChatGPT just became an ‘independent computer’ with this. This is not GPT-5, but this is very close. Here’s why this is the beginning of a new era of AI: ⤵️

🚨 BREAKING: OpenAI just launched ChatGPT Agent. ChatGPT just became an ‘independent computer’ with this. This is not GPT-5, but this is very close. Here’s why this is the beginning of a new era of AI: ⤵️

Mushfiq Sajib

24,422 views • 1 year ago

GPT-5.6 vs GPT-5.5 on my custom spaceship prompt. I gave both models the exact same custom prompt. This is also the same prompt I previously gave to Fable 5. For context, GPT-5.6 Pro worked for 87 minutes, while GPT-5.5 Extra High worked for 34 minutes and 42 seconds. As I’ve said before, based on great authority GPT-5.6 will be an incremental/soldi improvement over GPT-5.5, not a “Fable killer.” My rough expectation has been that it would trade blows with Fable 5 on some benchmarks, maybe win around half depending on the category, but not clearly surpass it overall. And again fable five will have bigger model smell, but this was expected. After testing this coding output, that view feels pretty accurate. GPT-5.6 is clearly better than GPT-5.5 in several visual areas. The lighting, shading, chairs, object details, and exterior of the spaceship looked noticeably stronger. The scene was also easier to test. I do want to give GPT-5.5 credit though. It built out the rooms much much better and the planets looked better than GPT-5.6’s. It was also interesting that both GPT-5.5 and GPT-5.6 produced better-looking planets than Fable 5 in this specific test. The downside with GPT-5.5 was stability. The game was much glitchier and harder to test compared to GPT-5.6. But when it comes to the core of the demo, which is the spaceship itself, Fable 5 still beat both models pretty comfortably. GPT-5.6 is impressive, but from this test, it looks exactly like what I expected which was a meaningful incremental improvement over GPT-5.5, at least for indie game demos, but not something that replaces Fable 5. In collaboration with Chetaslua

GPT-5.6 vs GPT-5.5 on my custom spaceship prompt. I gave both models the exact same custom prompt. This is also the same prompt I previously gave to Fable 5. For context, GPT-5.6 Pro worked for 87 minutes, while GPT-5.5 Extra High worked for 34 minutes and 42 seconds. As I’ve said before, based on great authority GPT-5.6 will be an incremental/soldi improvement over GPT-5.5, not a “Fable killer.” My rough expectation has been that it would trade blows with Fable 5 on some benchmarks, maybe win around half depending on the category, but not clearly surpass it overall. And again fable five will have bigger model smell, but this was expected. After testing this coding output, that view feels pretty accurate. GPT-5.6 is clearly better than GPT-5.5 in several visual areas. The lighting, shading, chairs, object details, and exterior of the spaceship looked noticeably stronger. The scene was also easier to test. I do want to give GPT-5.5 credit though. It built out the rooms much much better and the planets looked better than GPT-5.6’s. It was also interesting that both GPT-5.5 and GPT-5.6 produced better-looking planets than Fable 5 in this specific test. The downside with GPT-5.5 was stability. The game was much glitchier and harder to test compared to GPT-5.6. But when it comes to the core of the demo, which is the spaceship itself, Fable 5 still beat both models pretty comfortably. GPT-5.6 is impressive, but from this test, it looks exactly like what I expected which was a meaningful incremental improvement over GPT-5.5, at least for indie game demos, but not something that replaces Fable 5. In collaboration with Chetaslua

Chris

249,713 views • 1 month ago

ChatGPT for Research No more sifting through endless research papers to find the answer you need. Just ask any question and Consensus AI will deliver a GPT-4-powered summary of the top 5-10 papers.

ChatGPT for Research No more sifting through endless research papers to find the answer you need. Just ask any question and Consensus AI will deliver a GPT-4-powered summary of the top 5-10 papers.

Shubham Saboo

1,142,801 views • 3 years ago

🆕Scaling Test Time Compute to Multi-Agent Civilizations, with Noam Brown We're excited to publish our full conversation with Noam Brown on the frontiers of the new reasoning paradigm at OpenAI! - first principles for starting the "Multi-Agents" team - what's not captured by the "System 1/System 2" analogy for inference time compute - how Ilya Sutskever convinced him that reasoning was closer than he thought - Deep Research is existence proof that RL generalizes beyond verifiable rewards - the relationship between AI for imperfect information games (like Poker, Stratego, Diplomacy) and reasoning Enjoy! on youtube, or wherever fine podcasts are sold.

🆕Scaling Test Time Compute to Multi-Agent Civilizations, with Noam Brown We're excited to publish our full conversation with Noam Brown on the frontiers of the new reasoning paradigm at OpenAI! - first principles for starting the "Multi-Agents" team - what's not captured by the "System 1/System 2" analogy for inference time compute - how Ilya Sutskever convinced him that reasoning was closer than he thought - Deep Research is existence proof that RL generalizes beyond verifiable rewards - the relationship between AI for imperfect information games (like Poker, Stratego, Diplomacy) and reasoning Enjoy! on youtube, or wherever fine podcasts are sold.

Latent.Space

105,923 views • 1 year ago