Peder Engman shares a short post and link to an article that reviews Claude.
Fascinating insights on the internal workings of AI model Claude! Anthropic researchers show how the LLM plans its writing many words ahead, sometimes reasons in a “universal language” and at times applies illogical reasoning to back up an incorrect answer suggested by the user. Remarkable findings about the LLM “black box”, and not quite the way were thinking that it was working….