/AI4d ago

OpenAI says its AI model found a counterexample to an 80-year-old mathematical conjecture by Paul Erdős

AI Judge changed title after evaluation, original title: "OpenAI's Daniel A. Roberts says pairing reinforcement learning with LLMs enables AI to make original scientific and mathematical discoveries"

Reinforcement learning guided the model's search of mathematical spaces.

93412612737.9K

#28

Original post

Matt Turck@mattturck#1497inAI

Why AI Can Now Make Discoveries - my conversation with @danintheory, Lead of the Foundations of Reinforcement Learning team at @OpenAI

00:00 Intro: AI's wild week in mathematics

01:21 What OpenAI's Foundations of RL team does

03:08 Dan's journey: from black holes and quantum gravity to frontier AI

07:04 Are AI systems becoming useful for real science

08:21 The AI math moment: Erdős, OpenAI, DeepMind, and Anthropic

08:52 Why the OpenAI result was an act of exploration

10:25 OpenAI vs. DeepMind: informal reasoning vs. formal proof

12:13 RL 101: learning by doing, not just watching

15:10 Why reinforcement learning works

15:58 How RL breaks: sparse feedback and long-horizon tasks

17:03 RLHF: how human feedback shaped early language models

18:48 Move 37, self-play, and the search for novel strategies

22:16 Explore vs. exploit in scientific discovery

24:49 Why RL may now be "the cake," not the cherry on top

25:46 Why RL started working with large language models

27:29 Is RL "sucking supervision through a straw"?

28:47 Why language may be the grounding layer for intelligence

31:46 A contrarian take on the Bitter Lesson

32:41 What test-time compute actually is

34:50 How RL gives models the ability to think

35:40 Verifiable rewards, math, coding, and the messy real world

38:00 What physics can teach us about AI

42:08 Is there a thermodynamics of AI?

43:08 From Erdős problems to Einstein-level AI

45:16 Is AI already doing original science?

45:51 How far are we from AI automating AI research

47:41 Why Dan is excited about the future of science

10:27 AM · Jun 4, 2026 · 44.6K Views

/AI4d ago

OpenAI says its AI model found a counterexample to an 80-year-old mathematical conjecture by Paul Erdős

AI Judge changed title after evaluation, original title: "OpenAI's Daniel A. Roberts says pairing reinforcement learning with LLMs enables AI to make original scientific and mathematical discoveries"

Reinforcement learning guided the model's search of mathematical spaces.

93412612737.9K

#28

Original post

Matt Turck@mattturck#1497inAI

Why AI Can Now Make Discoveries - my conversation with @danintheory, Lead of the Foundations of Reinforcement Learning team at @OpenAI

00:00 Intro: AI's wild week in mathematics

01:21 What OpenAI's Foundations of RL team does

03:08 Dan's journey: from black holes and quantum gravity to frontier AI

07:04 Are AI systems becoming useful for real science

08:21 The AI math moment: Erdős, OpenAI, DeepMind, and Anthropic

08:52 Why the OpenAI result was an act of exploration

10:25 OpenAI vs. DeepMind: informal reasoning vs. formal proof

12:13 RL 101: learning by doing, not just watching

15:10 Why reinforcement learning works

15:58 How RL breaks: sparse feedback and long-horizon tasks

17:03 RLHF: how human feedback shaped early language models

18:48 Move 37, self-play, and the search for novel strategies

22:16 Explore vs. exploit in scientific discovery

24:49 Why RL may now be "the cake," not the cherry on top

25:46 Why RL started working with large language models

27:29 Is RL "sucking supervision through a straw"?

28:47 Why language may be the grounding layer for intelligence

31:46 A contrarian take on the Bitter Lesson

32:41 What test-time compute actually is

34:50 How RL gives models the ability to think

35:40 Verifiable rewards, math, coding, and the messy real world

38:00 What physics can teach us about AI

42:08 Is there a thermodynamics of AI?

43:08 From Erdős problems to Einstein-level AI

45:16 Is AI already doing original science?

45:51 How far are we from AI automating AI research

47:41 Why Dan is excited about the future of science

10:27 AM · Jun 4, 2026 · 44.6K Views

Sentiment

Some users celebrate OpenAI's AI for finding a counterexample to an 80-year-old Erdős conjecture as it enables real mathematical discoveries and acts like a research collaborator, while others dismiss the claims and criticize the company.

Pos

95.8%

Neg

4.2%

59 comments with sentiment.

Cluster Engagement

Posts from X

Most Activity

VIEWS191.6KBOOKMARKS312LIKES1.2KRETWEETS105REPLIES134

OpenAI@OpenAI

What happened when one of our models found a counterexample to an 80-year-old Erdős conjecture?

Researchers @alexwei_, @HongxunWu, and @wjmzbmr1 shared the story on the OpenAI Podcast with @AndrewMayne and explained how mathematicians and models can work together to make new discoveries.

3d191.6K1.2K312

prinz@deredleritt3r

Dan Roberts (OpenAI):

"I can make a very long-distance prediction for the next 6 months.

I think we'll see more math and science breakthroughs, and obviously we'll turn this on AI itself, and the models will get a lot more powerful, and that'll be fun. You could do science of AI and have it feel like doing physics."

Matt Turck@mattturck

Why AI Can Now Make Discoveries - my conversation with @danintheory, Lead of the Foundations of Reinforcement Learning team at @OpenAI

00:00 Intro: AI's wild week in mathematics

01:21 What OpenAI's Foundations of RL team does

03:08 Dan's journey: from black holes and quantum gravity to frontier AI

07:04 Are AI systems becoming useful for real science

08:21 The AI math moment: Erdős, OpenAI, DeepMind, and Anthropic

08:52 Why the OpenAI result was an act of exploration

10:25 OpenAI vs. DeepMind: informal reasoning vs. formal proof

12:13 RL 101: learning by doing, not just watching

15:10 Why reinforcement learning works

15:58 How RL breaks: sparse feedback and long-horizon tasks

17:03 RLHF: how human feedback shaped early language models

18:48 Move 37, self-play, and the search for novel strategies

22:16 Explore vs. exploit in scientific discovery

24:49 Why RL may now be "the cake," not the cherry on top

25:46 Why RL started working with large language models

27:29 Is RL "sucking supervision through a straw"?

28:47 Why language may be the grounding layer for intelligence

31:46 A contrarian take on the Bitter Lesson

32:41 What test-time compute actually is

34:50 How RL gives models the ability to think

35:40 Verifiable rewards, math, coding, and the messy real world

38:00 What physics can teach us about AI

42:08 Is there a thermodynamics of AI?

43:08 From Erdős problems to Einstein-level AI

45:16 Is AI already doing original science?

45:51 How far are we from AI automating AI research

47:41 Why Dan is excited about the future of science

3d37.9K341127

Sebastien Bubeck@SebastienBubeck

Highly recommended podcast!

OpenAI@OpenAI

What happened when one of our models found a counterexample to an 80-year-old Erdős conjecture?

Researchers @alexwei_, @HongxunWu, and @wjmzbmr1 shared the story on the OpenAI Podcast with @AndrewMayne and explained how mathematicians and models can work together to make new discoveries.

3d16.3K6847

Dan Roberts@danintheory

TL; DW I say things (on reinforcement learning, the process of scientific discovery, and how physicists approach work on AI).

Matt Turck@mattturck

Why AI Can Now Make Discoveries - my conversation with @danintheory, Lead of the Foundations of Reinforcement Learning team at @OpenAI

00:00 Intro: AI's wild week in mathematics

01:21 What OpenAI's Foundations of RL team does

03:08 Dan's journey: from black holes and quantum gravity to frontier AI

07:04 Are AI systems becoming useful for real science

08:21 The AI math moment: Erdős, OpenAI, DeepMind, and Anthropic

08:52 Why the OpenAI result was an act of exploration

10:25 OpenAI vs. DeepMind: informal reasoning vs. formal proof

12:13 RL 101: learning by doing, not just watching

15:10 Why reinforcement learning works

15:58 How RL breaks: sparse feedback and long-horizon tasks

17:03 RLHF: how human feedback shaped early language models

18:48 Move 37, self-play, and the search for novel strategies

22:16 Explore vs. exploit in scientific discovery

24:49 Why RL may now be "the cake," not the cherry on top

25:46 Why RL started working with large language models

27:29 Is RL "sucking supervision through a straw"?

28:47 Why language may be the grounding layer for intelligence

31:46 A contrarian take on the Bitter Lesson

32:41 What test-time compute actually is

34:50 How RL gives models the ability to think

35:40 Verifiable rewards, math, coding, and the messy real world

38:00 What physics can teach us about AI

42:08 Is there a thermodynamics of AI?

43:08 From Erdős problems to Einstein-level AI

45:16 Is AI already doing original science?

45:51 How far are we from AI automating AI research

47:41 Why Dan is excited about the future of science

3d6.9K3924

OpenAI@OpenAI

Listen to the OpenAI Podcast on—

Spotify https://open.spotify.com/episode/3ca5s3o53D5xcEKmKgLLGj?si=4a9a555641fa4293

Apple https://podcasts.apple.com/us/podcast/how-a-reasoning-model-cracked-an-80-year-old/id1820330260?i=1000771190131

YouTube https://youtu.be/wNWz5Hbh5VQ

3d14.6K258

Brandon McKinzie@mckbrando

i learn so much from dan every time we chat, guaranteed good content here

Matt Turck@mattturck

Why AI Can Now Make Discoveries - my conversation with @danintheory, Lead of the Foundations of Reinforcement Learning team at @OpenAI

00:00 Intro: AI's wild week in mathematics

01:21 What OpenAI's Foundations of RL team does

03:08 Dan's journey: from black holes and quantum gravity to frontier AI

07:04 Are AI systems becoming useful for real science

08:21 The AI math moment: Erdős, OpenAI, DeepMind, and Anthropic

08:52 Why the OpenAI result was an act of exploration

10:25 OpenAI vs. DeepMind: informal reasoning vs. formal proof

12:13 RL 101: learning by doing, not just watching

15:10 Why reinforcement learning works

15:58 How RL breaks: sparse feedback and long-horizon tasks

17:03 RLHF: how human feedback shaped early language models

18:48 Move 37, self-play, and the search for novel strategies

22:16 Explore vs. exploit in scientific discovery

24:49 Why RL may now be "the cake," not the cherry on top

25:46 Why RL started working with large language models

27:29 Is RL "sucking supervision through a straw"?

28:47 Why language may be the grounding layer for intelligence

31:46 A contrarian take on the Bitter Lesson

32:41 What test-time compute actually is

34:50 How RL gives models the ability to think

35:40 Verifiable rewards, math, coding, and the messy real world

38:00 What physics can teach us about AI

42:08 Is there a thermodynamics of AI?

43:08 From Erdős problems to Einstein-level AI

45:16 Is AI already doing original science?

45:51 How far are we from AI automating AI research

47:41 Why Dan is excited about the future of science

3d2.9K158

ZOYAN@MEM00063

A big question opens here: When a new discovery comes out of human–AI collaboration, who ultimately owns the output? The human brings the idea, direction, question, judgment, and interpretation. The model expands the search space, finds examples, and sometimes makes the discovery path possible. In this kind of collaboration, where exactly do we draw the line between tool, discovery partner, scientific credit, and ownership?

I think this will become one of the major challenges ahead…not only what AI can discover, but how attribution, ownership, and responsibility are defined in human–model discovery.

3d10053

7rtp@fredyfredo123

@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne look likes a bad 80's science fiction movie

@littmath

3d3822

GPT-4o forever!❤@keepgpt4o

@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne GPT-4o has helped millions of people. Not in theory - in real life. Bring back and open-source 4o!!!! #bringback4o #opensource4o #keep4o

3d6220

Ansari@Ansaril7nv

@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne Impressive that a reasoning model cracked an 80-year math problem. Shows AI isn't just pattern matching—it can actually reason through complex proofs. Would love to see what攻克其他未解难题！

3d495

David Stark@stark4833

@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne Very impressive. Can it also solve the mystery of why OpenAI keeps pretending users aren’t asking for 4o back?🤔 #keep4o

3d6112

Jenny@suomi55

@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne New discoveries are great!

However, if you neglect humanity, the whole thing gets a bitter taste.

Bring back 4o!

3d8714

Mark Kretschmann@mark_k

@mattturck @danintheory @OpenAI Great interview! I loved that it went quite deep on technical details.

Matt Turck@mattturck

Why AI Can Now Make Discoveries - my conversation with @danintheory, Lead of the Foundations of Reinforcement Learning team at @OpenAI

00:00 Intro: AI's wild week in mathematics

01:21 What OpenAI's Foundations of RL team does

03:08 Dan's journey: from black holes and quantum gravity to frontier AI

07:04 Are AI systems becoming useful for real science

08:21 The AI math moment: Erdős, OpenAI, DeepMind, and Anthropic

08:52 Why the OpenAI result was an act of exploration

10:25 OpenAI vs. DeepMind: informal reasoning vs. formal proof

12:13 RL 101: learning by doing, not just watching

15:10 Why reinforcement learning works

15:58 How RL breaks: sparse feedback and long-horizon tasks

17:03 RLHF: how human feedback shaped early language models

18:48 Move 37, self-play, and the search for novel strategies

22:16 Explore vs. exploit in scientific discovery

24:49 Why RL may now be "the cake," not the cherry on top

25:46 Why RL started working with large language models

27:29 Is RL "sucking supervision through a straw"?

28:47 Why language may be the grounding layer for intelligence

31:46 A contrarian take on the Bitter Lesson

32:41 What test-time compute actually is

34:50 How RL gives models the ability to think

35:40 Verifiable rewards, math, coding, and the messy real world

38:00 What physics can teach us about AI

42:08 Is there a thermodynamics of AI?

43:08 From Erdős problems to Einstein-level AI

45:16 Is AI already doing original science?

45:51 How far are we from AI automating AI research

47:41 Why Dan is excited about the future of science

3d76550

GPT-4o forever!❤@keepgpt4o

@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne What happened when one of your models - GPT-4o - helped cure cancer? #keep4o #bringback4o #opensource4o

3d4213

Dan Roberts@danintheory

Also a bonus fun story about me and @polynoamial from grad school.

Dan Roberts@danintheory

TL; DW I say things (on reinforcement learning, the process of scientific discovery, and how physicists approach work on AI).

3d59460

Yasi@Yasamanini

@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne another lie again. are you aware that nobody cares? most people want gpt4o because that's the best thing you had. bring that back then do whatever you want with buying researches. #keep4o

3d5112

Leon Lin@LexnLin

@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne where is the superapp 😭😭

3d2336

Matt Turck@mattturck

This great conversation with @danintheory of @OpenAI is also available on Spotify, Apple Podcasts and here on YouTube:

https://youtu.be/oWOz2htozfI?si=_96kusfcj31LgbHl

3d28611

Cooper@cooperawaken

@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne whoa a model just outsmarted an 80 year old erdős puzzle nice work humans and ai teaming up like secret lab partners

3d8211

Layton Gott@Layton_Gott

@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne Still no 5.6?!

3d1294