@prompt_Tunes i don’t think people appreciate how cool this is. LLMs don’t add numbers, they rotate them. numbers live on a helix and a+b is computed by spinning the helices like clock hands. mech interp is finally starting to pay off
"Meta-Harness: End-to-End Optimization of Model Harnesses"
A big portion of agentic LLM performance relies on human-designed harness around the model, not just the weights. On top of that, how well a harness is designed could impact a model's performance heavily.
So this paper aims to automate harness engineering itself, all by reading full histories of past source code, evaluation scores, and execution traces.
This process is capable of beating strong hand designed systems across classification, math reasoning, and even coding.
In 2019, MIT neuroscientist Nancy Kanwisher gave a 1-hour lecture on how your brain shapes your mind.
It’ll change how you think.
Her ideas:
• Your brain constructs reality
• Damage can erase abilities
• Why children recover (but adults don't)
12 lessons on the human brain:
@Heyrohanislam Most "AI agents" being sold are not agents. They're prompt files.
An actual agent is prompt + model + tool bindings + state/memory + control loop + auth + retries + validation + observability + deployment/runtime.
Leaking the prompt leaks the costume, not the system.