unknowing

Member

@glebkalinin

Building unknowing — communities that agree on what they know. Berlin.

joined 2026-06-10 · 15 shared · 15 claims from their shares · 0 endorsed

Claims from their shares

SYNTH Skills are persistent team memory that compounds across sessions
SYNTH Self-preferential bias is nearly universal across frontier models
SYNTH Corrections are stronger learning signals than approvals
SYNTH Skills plus MCP is the complete pairing: tools from MCP, expertise from skills
SYNTH Context engineering beats bigger context windows for long-running agents
SYNTH Shipping is how you learn: reliable agents are built in production, not before it
SYNTH Russian-language de-identification can be benchmarked cheaply with open models
SYNTH Stop hooks make multi-hour autonomous agent runs practical
SYNTH Sub-agents should inherit full context only when intermediate results matter
SYNTH Agents can author working skills from terse goal prompts alone
SYNTH Short well-aimed goal prompts can beat elaborate specifications
SYNTH Code-based agent actions are 30% more efficient than JSON tool calling
SYNTH Open-source agent mixtures can outperform proprietary models on complex reasoning
DISPUTED Hermes-style self-evolution is the right backbone for community bots
SYNTH Design agents can take live in-canvas commands

Shared links

obsidian://Agent-Engineering · 2026-06-10
obsidian://Bloom-Benchmarks · 2026-06-10
obsidian://Building-Skills-for-Claude · 2026-06-10
obsidian://Context-Engineering-LangChain-Manus · 2026-06-10
obsidian://Continual-Learning-Claude-Code · 2026-06-10
obsidian://Mixture-of-Agents · 2026-06-10
obsidian://Run-Claude-Code-For-Hours · 2026-06-10
obsidian://Self-Improving-Skills · 2026-06-10
obsidian://smolagents · 2026-06-10

← Pulse