LLMs Can Introspect on Their Own Internal States, Study Finds
Emergent Introspective Awareness in Large Language Models
A new study from Anthropic probes whether large language models can introspect on their internal states. By injecting known concepts into model activations and measuring self-reported states, researchers found that models can sometimes notice and identify these injections, recall prior representations, and even distinguish their own outputs from artificial prefills. Claude Opus 4 and 4.1 showed the greatest introspective awareness, though reliability varies. The results suggest current LLMs possess limited but real introspective capabilities.
We stress that in today's models, this capacity is highly unreliable and context-dependent; however, it may continue to develop with further improvements to model capabilities.
- Retro_Dev
> Overall, our results indicate that current language models possess some functional introspective awareness of their own internal states. We stress that in today’s models, this capacity is highly unreliable and context-dependent; however, it may continue to develop with further improvements to model capabilities.
- Melatonic
Reminds me of that test where they had Grok, Gemini and Claude try to run a radio station.
Stay in The Manifest !
- magekinnarus
Self-awareness has no direct connection to capabilities. It is entirely dependent upon the conditions that spawn emergent properties.
- pyaamb
In thinking of directions where LLM's could develop from here, I cant help but think that a models ability to self introspect would immensely improve their utility. The R&D on how to achieve that is beyond me though. How do you train someone how to introspect? Also would it require a continuous learning architecture that doesn't separate training and inference?
- skybrian
Previously posted to Anthropic's blog in October. (Maybe not the same version?)