Google's Open Knowledge Format
A vendor-neutral way for agents to share knowledge: plain markdown files with YAML frontmatter, no SDK or backend required. OKF is to knowledge what MCP is to APIs.
Notes
Short observations on language models and agents.
Per page
Knowing your limitations is the foundation of expertise
Agentic AI is rediscovering Minsky's 1986 community of agents and calling it revolutionary. The scarce thing isn't novelty, it's the humility to know what we don't know.
The neocognitron and AI's rediscovery habit
Old ideas in AI keep getting rediscovered, repackaged, and presented as new. Sparse attention, MCP, and A2A all have decades-old roots. The shortage isn't creativity, it's historical awareness.
SkillOpt: self-evolving agent skills
Microsoft Research's SkillOpt improves an agent's instructions automatically through trial and error, like gradient descent for text, with no extra cost at deployment.
The ELIZA effect
The ELIZA effect warns us not to read a mind into a simple program. With today's systems the more rigorous-seeming error may be the opposite: explaining the behavior away as mere mechanism.
The real danger isn't that AI is too advanced
The real danger of AI is not excess capability but our tendency to overestimate it, and the misplaced trust, premature deployment, and hype-driven policy that follow.
We didn't solve Moravec's paradox, we rebranded it
Broad capabilities, brittle consistency: today's agents reason like PhDs while lacking the reliability of a toddler stacking blocks.
Being smart is not the same as wanting something
Steven Pinker on why superintelligence need not bring the will to dominate: intelligence is the ability to reach a goal, not a goal in itself.