AI stuff
Everything I want to remember about LLMs and other machine learning stuff
AI stuff
How We Build Effective Agents: Barry Zhang, Anthropic
youtube.comBuilding Effective AI Agents | Anthropic
anthropic.com
But I don’t think he really answers the question he shows in the beginning: Why can’t I just unplug it? Why is Distribution not the bottleneck? Feels a but fluffy
AI Horseless Carriages | koomen.dev
koomen.dev
Should the user write the system prompt? Maybe. But we could also write better, individual system prompts for users based on their context and allow them to edit it.
(How) Do LLMs Reason/Plan? (Talk given at Microsoft Research; 4/11/25)
youtube.comVery interesting insights in how reasoning models work. He argues that the intermediate "reasoning traces" (e.g., chains of thought) produced by LLMs are not necessarily meaningful or reliable, and that these traces can be misleading or simply stylistic rather than substantive.
Because if you strip LLMs to their essence, they are just a much better way of using statistics to aggregate human intelligence and connect everything we’ve all done together to get more use out of it.
Sari Azout • Becoming unLLMable
A practical guide to building agents
Guide to building AI agents using large language models, covering agent definition, use case selection, design components, single/multi-agent orchestration, tool integration, instruction setup, safety guardrails, and deployment best practices.
cdn.openai.comWhat that means is that AI abilities are getting better even as costs are dropping. This graph shows how quickly this trend has advanced, mapping the capability of AI on the y axis and the logarithmically decreasing costs on the x axis. When GPT-4 came out it was around $50 per million tokens (roughly a word), now it costs around 12 cents per milli... See more