Loading activity…
@ao_qu18465
PhD at @MIT @mitidss @medialab | Building self-evolving AI & multisensory agents | Prev @ByteDanceSeed_ | First/co-first author: MEM1, CORAL & reef
AI
MIT's course page outlines four areas: context, agent harnesses, models and increasingly complex synthetic tasks.
AI
The presenter says CORAL has reached 1.1K GitHub stars and is used by engineers and researchers at Takeda and TikTok.
AI
An announcement says Reef can build, check and ship changes to an agent's setup from a plain-language request. Examples include adding a /chat mode for faster responses and spoken alerts when work is done.
AI
The announcement invites users to supply tasks and an evaluator to evolve prompts or an entire harness, and says custom harnesses can be integrated by adding an adapter.
AI
The announcement says users can enable LoRA training with a one-line configuration change, without a local GPU.
AI
MIT-affiliated researchers release code for agents that evolve from inference feedback.
AI
MIT PhD candidate shares link to article about his AI research on X.
