This is all a bit theoretical. When Kimi weights go public, we'll have an interesting conversation about unit economics vs V4.I suspect that V4 wins *crushingly*…
Impressive world modeling result, but… where did they get Gemini-Flash's pretraining costs?Induction Labs: We’re introducing imagination models: a new foundation model architecture that unlocks learning from…
worth rereading again, now that Wenfeng reminds us that continual learning is the next frontier (the last frontier before "soft singularity", in his accounting). Also check…
Who is more "safety-conscious" than Anthropic? Who is more value-aligned than half of founding OpenAI? Who seems closest to AGI at this point? ≥50% odds by…
> - 950 has 4. SerDes run on separate dies. This makes manufacturing yield higher (notice D2Ds on left and right).I missed this950 is an interesting…
RT Brangus🔍⏹️Show us the notes.
first impressions is I don't like Opus 5. It's the development of 4.7-4.8's line. Fable is Opus 4.5-4.6 that has grown into its mature shape, replaced…
The reason Anthropic people trust Claude is that they've built Claude *so that it can be trusted*, which combined with their science of training also means…
yes, but all good test sets are inherently good training sets. I think Opus 5's behavior on ARC-3 should decently generalize to tasks that can be…
One unambiguously good thing to come out of this:We will get to learn which models took > 1e26 FLOPsAndrew Curran: A bipartisan group of six House…