Anthropic develops new tool to examine hidden processes in LLM generation
Anthropic develops new tool to examine hidden processes in LLM generation
www.technologyreview.com
Anthropic can now track the bizarre inner workings of a large language model
What the firm found challenges some basic assumptions about how this technology really works.
I liked reading this article. Its cool to really poke into the hidden perplexity behind patterns of 'thought' in llms. They aren't merely simple 'auto complete'.
The finding that claude does math in a different way then it says it does and can anticipate words ahead of generation time are facinating.