Chroma
@trychroma
Retrieval for AI. http://trychroma.com https://github.com/chroma-core/chroma
Since the task was long running, context quickly exploded out of model windows. To solve this we only pass the last 30k tokens to the agent loop, And gave write/read memory tools (embed important text and store them in a @trychroma vector store). 4/n
Join @kellyhongsn and @atroyn for this event next Tuesday! They will share surprising insights that didn't make it into the report lu.ma/vw17piwl
Introducing our latest technical report: Context Rot - How Increasing Input Tokens Impacts LLM Performance Our results reveal that models do not use their context uniformly. full report in replies
We sat down with @jeffreyhuber (Founder & CEO @trychroma) to discuss AI development. "We want AI to be strong in all the ways human intelligence is strong and weak." "Chroma's report suggests models can fall apart earlier than expected. You might see reasoning performance drop…
we 🫶 Ramp
The days of building slide decks from scratch will soon be over. Our very own @shevchenkoaalex did a talk with @trychroma on a new AI Presentation Co-Pilot his team is cooking, which generates slides by pulling brand guidelines, data, and research, then curates the best outputs.