Log inSign up
Isaac Kauvar profile banner
user avatar

Isaac Kauvar

@ikauvar
mapping neural circuits of intelligence @anthropicAI prev: phd in cross-species neuroscience (human, mouse, fish) @stanford
California, USA
ivk.io
Joined February 2018
528
Following
780
Followers
RepliesMedia
  • user avatar
    Isaac Kauvar
    @ikauvar
    Jul 7
    One thing that struck me about this workspace is its causal role. Many (but not all!) of a model's behaviors, especially complex ones, depend on thoughts that flow through the J-space, reminiscent of the distinction between behaviors that are flexible vs. automatic for humans.
    user avatar
    Anthropic
    @AnthropicAI
    Jul 6
    New Anthropic research: A global workspace in language models. Of everything happening in your brain right now, only a tiny fraction is consciously accessible—thoughts you can describe, hold in mind, and reason with. We found a strikingly similar divide inside Claude.
    Image
    00:00
  • user avatar
    Isaac Kauvar
    @ikauvar
    Jul 6
    A linear representation of "Am I on the right track?" in a language model can influence backtracking, verbalized confidence, and even how many comments it writes in code!
    user avatar
    Nick Jiang
    @nickhjiang
    Jun 17
    New work: The Value Axis 🎯 How do LLMs choose which path to take mid-task? We find they internally track the chance of reaching their goal along a linear axis, akin to a value function in RL. We show it modulates confidence in math & coding and can be reshaped with DPO and SFT.
    Image
    00:00
  • user avatar
    Isaac Kauvar
    @ikauvar
    Apr 2
    Language models often appear to express emotions. What's going on here?? But perhaps more importantly, does it matter? Excited to share our new research about how emotion-related representations in Claude can shape its alignment-relevant behavior.
    user avatar
    Anthropic
    @AnthropicAI
    Apr 2
    New Anthropic research: Emotion concepts and their function in a large language model. All LLMs sometimes act like they have emotions. But why? We found internal representations of emotion concepts that can drive Claude’s behavior, sometimes in surprising ways.
    Image
    00:00