Working on AI stuff mostly. To that end, I worked on a little "bridge" library[1] for making it easier to use Jason with Spring, and made a fork of Jason to accommodate some of my personal preferences, like using SLF4J and Logback for logging, instead of JUL. Then wrote a Jason based system that can talk via XMPP, and wired it up to my Ollama server so that when I send it a message it has the option of asking an LLM to help parse the incoming message. Note that this is not a simple pass-through to the LLM. The intent here is to do higher level reasoning and other cognitive functions outside of the LLM, where the LLM is (potentially) helpful, but is not the core of the system.
Then started on some really super simple stuff around classical / operant conditioning for learning by sending @commands over the chat. Think something like @goodbot, or @no, and an @say command which the bot will learn to mirror from. I'm also working on some stuff around the idea of various "mental states" (stuff like cognitive-load, attention, confusion, frustration, cognitive-dissonance, etc) and exposing those things via JMX (so I can introspect the inner mental state of the system from outside, as well as logging changes during specific experiments).
Earlier this year I had done some work around making the Jason "belief base" use Apache Jena so that I can store beliefs as RDF triples and then use inference engines that can plug into Jena for reasoning. One of my next steps is to wire that into this bot. Another is to work on a specific notion of splitting memory up in a way that's somewhat analogous to human memory: the BeliefBase for long-term memory, maybe something like ValKey for short-term memory, and then a separate internal "working memory" that isn't persistent and isn't stored very long. Haven't given much thought yet to "episodic memory" but that's probably needed as well.
Also spent some time running through the Anthropic MCP "quick start" and writing some simple MCP servers and clients.
And now... so very much more to explore. But that's for 2025 to come.
I didn't just work over the holiday break though. I did go visit family for Christmas, and spent a little bit of time watching movies and over-eating. :-)
Then started on some really super simple stuff around classical / operant conditioning for learning by sending @commands over the chat. Think something like @goodbot, or @no, and an @say command which the bot will learn to mirror from. I'm also working on some stuff around the idea of various "mental states" (stuff like cognitive-load, attention, confusion, frustration, cognitive-dissonance, etc) and exposing those things via JMX (so I can introspect the inner mental state of the system from outside, as well as logging changes during specific experiments).
Earlier this year I had done some work around making the Jason "belief base" use Apache Jena so that I can store beliefs as RDF triples and then use inference engines that can plug into Jena for reasoning. One of my next steps is to wire that into this bot. Another is to work on a specific notion of splitting memory up in a way that's somewhat analogous to human memory: the BeliefBase for long-term memory, maybe something like ValKey for short-term memory, and then a separate internal "working memory" that isn't persistent and isn't stored very long. Haven't given much thought yet to "episodic memory" but that's probably needed as well.
Also spent some time running through the Anthropic MCP "quick start" and writing some simple MCP servers and clients.
And now... so very much more to explore. But that's for 2025 to come.
I didn't just work over the holiday break though. I did go visit family for Christmas, and spent a little bit of time watching movies and over-eating. :-)
[1]: https://github.com/mindcrime/spring-jason-lib
[2]: https://github.com/mindcrime/jasonfg