/Tech4h ago

Critic Claims OpenAI Risks Losing AI Race Over Q* Reasoning Focus

1315516919.1K

Original post

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex#501inTech

Would be so *hilarious* if OpenAI lost the race because they bought the Q*/ reasoning meme (together with The Genius Noam Brown, who now defends it nervously on every turn). Weren't cynical enough about LLMs being ad hoc memoization machines. And Dario was just that cynical.

Igor Kotenkov@stalkermustang

The more I look at Fable across benchmarks, the more questions I have to OpenAI. WHERE. ARE. THE. BIG. MODELS.

What happened to scaling? What happened to the infra/experience they've got with 4.5?

Openai is losing the fields they were constantly winning on, science being the most important one. I thought there'd be an analogue of Project Glasswing but for Science, big results, etc., and yes, technically we got some, but the scale of discoveries is meh. I'm pretty convinced A\ can do almost everything that OAI can in the science field now.

And, more importantly, as far as I understand it, both companies will receive classified government data from Genesis Mission, so there will be no moat (kek).

sad day for OAI fans, including myself.

3:50 PM · Jun 12, 2026 · 15.4K Views

Sentiment

Positive users praise Anthropic's Dario and Chris Olah for quietly advancing reasoning capabilities over OpenAI's Q* focus, while negative users harshly dismiss multiple AI labs' efforts.

Pos

66.7%

Neg

33.3%

3 comments with sentiment.

Cluster Engagement

Posts from X

Most Activity

VIEWS2.1KBOOKMARKS2LIKES17REPLIES2

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex

I may have to retract this

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex

Twitching The vermin knows he has no other edge but muh scale learned that one trick in Baidu, never learned another

4h2.1K172

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex

I don't notice Noam Shazeer rescuing Gemini either btw, they're duking it out with DeepSeek-Flash on MathArena of all things (and losing) Yann… well you know how it's going All Vaswani-et-al 2017 team… where are they? maybe this whole jerking off to Geniuses was very misguided

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex

4h1.6K132

🇸🇴 🕳️ Stochastic Dreams@GenHeres123

Did anyone genuinely expect Dario to pull this off? He spent years looking like the nerdy kid in the back of the class constantly talking about how hard and scary everything is, playing it safe, doing things "the responsible way" while everyone else was flexing. And then he just quietly gets the highest grade in the room.

4h471

Burito@Britoisinsane

@teortaxesTex Yeah. OpenAI has Noam Brown, but Anthropic has Chris Olah. Reasoning capability also depends heavily on abstract, language-independent circuits that emerge through scaling. Dario knows this.

https://transformer-circuits.pub/2025/attribution-graphs/biology.html

4h161

RA@hoangbrian

@teortaxesTex What would be even more hilarious is if they all eggs in one basket (q*)

4h24

Luigi Pagani@Luigi1549898

@teortaxesTex Fable is undeniable the best model by a good margin, but I think you are updating a bit too much, let's not be too much frequentist and too little Bayesian. The coming months will tell

4h1