Facebook says BlenderBot 2.0, its updated AI chatbot, can recall past conversations that span months and search the internet to update itself about chat topics
Last April, Facebook's AI research lab (FAIR) announced and released as — open source its BlenderBot social chat app.
Context & Ripple Effects
This is the second step in FAIR's chatbot arc: after open-sourcing the original Blender model in April 2020 as a research artifact, Facebook is now shipping capabilities aimed at real conversation — long-term memory across months and live internet search so the bot updates itself on what it's talking about.
Both additions attack known weaknesses of static trained models: forgetting context between sessions and going stale on current events. The open-source release means researchers outside Facebook can reproduce the setup, and the coverage shows where it leads — Meta later put the successor, BlenderBot 3, in front of US users specifically to harvest feedback on its capabilities.
First-order effects
- AI researchers get an open-source chatbot whose two new components — months-spanning recall and internet lookup — can be studied and reused without Facebook's infrastructure, since FAIR released the app publicly.
Second-order effects
- Rivals building social or assistant chatbots are pushed toward the same pairing of persistent user memory plus web grounding, because a bot that forgets past chats and answers from stale training data now looks visibly worse on both counts.
Third-order effects
- If the pattern holds through BlenderBot 3's public-feedback deployment, consumer chatbots shift from one-shot trained models to continuously corrected systems — with the company running the largest feedback loop setting the improvement pace.
The trend: Consumer chatbots are evolving from statically trained models into self-updating assistants that combine long-term personal memory with live web search, refined through public use.