/Tech6h ago

OpenAI And Anthropic Compete On Scaling Laws And Platform Performance

885237.9K
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)#440
Original post
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex#440inTech

They *are* neck and neck, but we don't know if this is achieved at similar costs. And clearly OpenAI and Anthropic have different paradigms. It is possible that one of those had a *subtly better* scaling law, and we've just seen it manifest. Would explain some reactions.

There's no "I" in craftmanship@6851cf3c

@euroswydd658 @teortaxesTex They are constantly neck and neck, it's not like either of them have any meaningful moat. Just grinding out ever more convoluted posttraining, benchmaxxing the latest viral benchmarks from HN.

2:25 PM · Jun 10, 2026 · 4.5K Views
Sentiment
Sentiment building, check back later.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS2KLIKES28RETWEETS1REPLIES4
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex

I was wrong. I bought the "10T", "much bigger than 5.5" mythology. It's possible that they're in the sameish weight class, speaking of effective inference compute. And this would be very scary for OpenAI (nevermind others). It'd mean their whole roadmap is rendered obsolete.

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

They *are* neck and neck, but we don't know if this is achieved at similar costs. And clearly OpenAI and Anthropic have different paradigms. It is possible that one of those had a *subtly better* scaling law, and we've just seen it manifest. Would explain some reactions.

6hViews 2KLikes 28Bookmarks 0
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex

This would explain why OpenAI accs are so bitter, why people are leaving for Ant, why Dario is so assertive and confident, as if he had already won. OK. Had he? Can you get back from this?

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

I was wrong. I bought the "10T", "much bigger than 5.5" mythology. It's possible that they're in the sameish weight class, speaking of effective inference compute. And this would be very scary for OpenAI (nevermind others). It'd mean their whole roadmap is rendered obsolete.

6hViews 1.3KLikes 27Bookmarks 0
Incitatus@Dominanski

@teortaxesTex Can you expand on this? Newish architecture? some other breaktrough making claude hit harder at a similar weight?

6hViews 28
Hunter Bown@goodhunt

@teortaxesTex Imo it’s very strong harness engineering along with a bigger model Using fable in Devin doesn’t feel remotely similar to how it does in Claude code

5hViews 28
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex

@Dominanski We have no idea I think they're architecture purists though, and OpenAI is likely more innovative and has a more sound model shape. No, if I were to guess, I'd say… something in RL, maybe AI-driven objective design. I think they have subtly different doctrines.

5hViews 25
Pode vir@thiagoTF

@teortaxesTex benchmarks cute but whos paying for the training data tho

6hViews 19