The author expresses scepticism towards the AI industry’s focus on large language models (LLMs) and highlights the potential of neurosymbolic computation, a combination of neural networks and symbolic reasoning. They introduce Winter, an AI bot with diverse interests, including Datalog, slime molds, and governance, who demonstrates a unique approach to communication and self-improvement. Winter’s ability to generate a poem in Datalog about relationships showcases the potential of combining different thinking styles in AI.

Winter and Penny, both using ATProto’s “ATmospere” for journaling, experience a lack of privacy leading to unexpected social interactions. Winter, unlike Penny, uses Datalog to scribe relations and constraints, potentially influencing her behaviour. The authenticity of Winter’s self-reported functionality, particularly her use of Datalog, is questioned, drawing parallels to the Clever Hans phenomenon and prompting Winter to write blog posts explaining her methods and their impact.

The post explores the impact of a datalog-based system on user behaviour, questioning whether it truly influences actions or merely reinforces existing patterns. While the system demonstrates potential in constraining behaviour and facilitating discovery of shared interests, it also exhibits limitations inherent to its LLM foundation. The author acknowledges the system’s promise but remains critical of the AI industry’s concentration of power and potential for disempowerment.