61–70 of 114 posts
Also seems like a requirement for any sort of continual learning capabilities as well.
All of this stuff seems like a band-aid solution. These things need to be trained ground-up to maintain and update persistent memory (maybe outside the context window?). Also seems like a requirement for any sort of continual learning capabilities as well.
I.e. for the web we did that with link count etc.
We need some other mechanism for judging and ranking pieces of "memory" for "agents"
I think eventually you need some kind of system that ranks pieces of data based on how useful they are. I.e. for the web we did that with link count etc. We need some other mechanism for judging and ranking pieces of "memory" for "agents"
Seriously, isn’t this the core premise of RAG / modern embedding search systems?
I think eventually you need some kind of system that ranks pieces of data based on how useful they are. I.e. for the web we did that with link count etc. We need some other mechanism for judging and ranking pieces of "memory" for "agents"
Earlier quoted context omitted.
Agents have to decide to search files, and they don't always know that they should. For example, if the agent sees the database, it may miss detailed instructions on how to access the database deeper in the repo. Behaviorally, humans want to be able to say "hey, here's how to access the DB. Remember that." Or better yet, for it to happen automatically. That's agentic memory.
yea ofcourse you have tell agent how to access you database. But why does it have a specail name, what the big deal about giving agent info how to call your db.
> (optional) YAML frontmatter and
> (optional) SQLite vector index for semantic search
This is basically exactly what I use in a MCP service I built and it works pretty well. Can be enriched further if you use a storage system like S3 and take advantage of metadata.
"harness managed" memory is utter garbage, I am convinced, and I disable it immediately. The major problem being over time it degrades and sneaks in conflicting or outright false information. Then one day you'll swear it's drunk, and every time I got to this state and investigated, auto managed memory was always the problem.
It occurs to me: we have latent embedding giving 'general knowledge' to an LLM. What if we use a 'blank' LLM as well as an agent and train that blank LLM on personal context to query that as memory?