GPT-5.6 Sol Pro disproves 20-year Benjamini-Hochberg statistical conjecture · Digg

GPT-5.6 Sol Pro disproves 20-year Benjamini-Hochberg statistical conjecture · Digg

/TECH

The AI model constructed a counterexample using a Gaussian factor model.

Combined views

29.4K

3 posts, first seen 12h ago

259 likes

Commentary on X

Gerard Sans | Axiom 🇬🇧@gerardsans

be careful with your wording. A brick is not an architect. A hammer doesn’t build a house. AI can’t solve anything. You can use AI to create hypothesis. Someone needs to actually verify them. The AI needs prompting and steering. Like any other tool. There’s a massive difference between AI was used and AI solved a problem. The gap is the prompt and the verification process. So no AI didn’t solve anything. Not with the current technology anyways.

6:32 AM · Jul 15, 2026

Anastasios Nikolas Angelopoulos@ml_angelopoulos

@EdgarDobriban Congrats on this incredible result, @EdgarDobriban. Tijana and I tried many times with GPT 5.5 as well with no luck. It is incredible that it is now solved and solved by AI. It really makes you think about what our future role should be as theoretically-minded staristicians.

1:03 AM · Jul 15, 2026

Combined views

29.4K

3 posts, first seen 12h ago

259 likes5 comments

5 comments

79 saves

131 reposts

Sentiment

Positive84.9%15.1%Negative

Positive Read

Many users celebrate GPT-5.6 disproving the Benjamini-Hochberg FDR conjecture because it shows AI advancing complex math through novel insights, while some caution that it overstates tools as true solvers and mourns lost human achievements.

Based on 34 visible X reactions from 58 accounts.

Digg Deeper

Ask a question below.

Published answers will appear here.

Commentary on X

Gerard Sans | Axiom 🇬🇧@gerardsans

be careful with your wording. A brick is not an architect. A hammer doesn’t build a house. AI can’t solve anything. You can use AI to create hypothesis. Someone needs to actually verify them. The AI needs prompting and steering. Like any other tool. There’s a massive difference between AI was used and AI solved a problem. The gap is the prompt and the verification process. So no AI didn’t solve anything. Not with the current technology anyways.

6:32 AM · Jul 15, 2026

Anastasios Nikolas Angelopoulos@ml_angelopoulos

@EdgarDobriban Congrats on this incredible result, @EdgarDobriban. Tijana and I tried many times with GPT 5.5 as well with no luck. It is incredible that it is now solved and solved by AI. It really makes you think about what our future role should be as theoretically-minded staristicians.

1:03 AM · Jul 15, 2026

Apurv Verma@verma_apurv5

@EdgarDobriban Very cool work, thanks for sharing!

11:05 AM · Jul 14, 2026

Mohammed Sharukh A@NoNTr1v1aL

@EdgarDobriban Congratulations! 🎉🥳👏

2:12 PM · Jul 14, 2026

View all

robma@robma1982

@EdgarDobriban 这有什么意义呢？没觉得有什么大的意思

9:11 AM · Jul 15, 2026

GPT5.6 one shot solved (by counter example) the central open question of one of the two most important developments in statistics since 1950.. a paper with 130,000 citations. 1) awesome 2) also great that other scientific fields outside ML are moving fast enough for people to public on X immediately (the models can do peer review too after all) 3) yet another result using combinatorial innovation, using a tool from another field in a way a trained human expert might not have 4) yet another result by finding a counter example to a conjecture… 5) wow the pins have started falling eh..

1:07 AM · Jul 15, 2026

The AI model constructed a counterexample using a Gaussian factor model.

Apurv Verma@verma_apurv5

@EdgarDobriban Very cool work, thanks for sharing!

11:05 AM · Jul 14, 2026

Mohammed Sharukh A@NoNTr1v1aL

@EdgarDobriban Congratulations! 🎉🥳👏

2:12 PM · Jul 14, 2026

View all

robma@robma1982

@EdgarDobriban 这有什么意义呢？没觉得有什么大的意思

9:11 AM · Jul 15, 2026

GPT5.6 one shot solved (by counter example) the central open question of one of the two most important developments in statistics since 1950.. a paper with 130,000 citations. 1) awesome 2) also great that other scientific fields outside ML are moving fast enough for people to public on X immediately (the models can do peer review too after all) 3) yet another result using combinatorial innovation, using a tool from another field in a way a trained human expert might not have 4) yet another result by finding a counter example to a conjecture… 5) wow the pins have started falling eh..

1:07 AM · Jul 15, 2026

79 saves

131 reposts

Sentiment

Positive84.9%15.1%Negative

Positive Read

Many users celebrate GPT-5.6 disproving the Benjamini-Hochberg FDR conjecture because it shows AI advancing complex math through novel insights, while some caution that it overstates tools as true solvers and mourns lost human achievements.

Based on 34 visible X reactions from 58 accounts.

Digg Deeper

Ask a question below.

Published answers will appear here.