is it a big model smell you smell?
RESEARCH
-

Divine benevolence behind our model’s state-of-the-art achievement
By
–
Magically achieved sota? That is because we believe in our model and also because of divine benevolence
-
Speed and Scale: The New Imperative in Modern Innovation
By
–
C’est un dicton has-been, il est évident que de nos jours il faut à la fois aller plus vite et plus loin, en même temps… https://t.co/cTP5LZOvxC
— Olivier Rimmel 🕊️ (@OlivierRimmel) 1 août 2025C’est un dicton has-been, il est évident que de nos jours il faut à la fois aller plus vite et plus loin, en même temps…
-

Anthropic Research on AI Alignment and Personality Control
By
–
This is more neat research from Anthropic, providing a lot of ways for careful organizations to shape the personality and guardrails of AI in deeper ways than prompts, including measuring and reducing sycophancy. Also the idea of an "evil vector" is interesting in and of itself.
-
How Much Models Think: The Key Simplification Axis
By
–
I can't share a lot at this point, but clearly how much we we allow the model to think is the main axis of the simplification.
-
The Miraculous Complexity of How Systems Actually Work
By
–
The deeper you dig on how something works, the more of a miracle it becomes that anything works at all.
-
IMO 2025 Gold Model Achieves SOTA on Reasoning Tasks
By
–
The IMO 2025 gold model achieved SOTA on many other reasoning tasks by large margins. I couldn't quite believe it when I first saw the results.
-

Anthropic Fellows Program: AI Research Leadership Opportunities
By
–
This research was led by @RunjinChen and @andyarditi through the Anthropic Fellows program, supervised by @Jack_W_Lindsey
, in collaboration w/ @sleight_henry and @OwainEvans_UK
. The Fellows program is accepting applications: -
Full-Time Researcher Positions Now Available
By
–
We’re also hiring full-time researchers to investigate topics like this in more depth:
-

Persona Vectors Identify Training Data Teaching Bad AI Traits
By
–
Persona vectors can also identify training data that will teach the model bad personality traits. Sometimes, it flags data that we wouldn't otherwise have noticed.