Pinned
Machine Learning Street Talk
@MLStreetTalk
MLST is by Dr. Tim Scarfe @ecsquendor w/ cameos from @DoctorDuggar youtube.com/c/MachineLearn… (early access/priv.discord) - Sponsor us!
- This is really interesting work. Inherent got their agent to reproduce masked out figures in popular ML papers by running the same experiments scaled down, with careful controls in place to maintain construct validity. They GRPO a 27B model to act as a controller for frontier1/ Today, we introduce Faraday, a 27B-parameter AI Scientist that extends the capabilities of coding agents with a layer of scientific intuition. Trained via long-horizon RL, Faraday outperforms Claude Opus 4.8 and GPT-5.5 on the task of replicating research papers. 🧵
- New book out soon - the Mythical Agent Month! (joke) The curse of agentic software engineering is that we keep more of the architecture in flux (in adaptation mode). Before we were forced to crystallise more and focus intelligence in small parts of the architecture at any one
- Adaptive specialised intelligenceso like it starts as a general-purpose harness but technically by the end of the task it has self-evolved into an ARC-AGI-3 specific harness
- The skill surface in an agent application should: > Represents everything that your app/code cannot yet do without needing intelligence. > Adapt over time, attract towards a global consistency. > Be ruthlessly re-abstracted, re-factored and compressed with increasing
