ayu.sh

for the love of the game

[ayu.sh @flomodoro]

you need to just do things

[ayu.sh @flomodoro]

@wshxnv congrats on the cease and desist from openai and the job offer from anthropic

[ayu.sh @flomodoro]

/compact cannot be trusted on its own.
using /compact with a prompt e.g "/compact focus on auth, saving state, forget the debugging we did." is the way

[ayu.sh @flomodoro]

you could say fable 5 one-shotted itself

[ayu.sh @flomodoro]

I have led a toothless life, he thought. A toothless life. I have never bitten into anything. I was waiting. I was reserving myself for later on—and I have just noticed that my teeth have gone.
Jean-Paul Sartre

[ayu.sh @flomodoro]

percentage of repos that I can clone and they just work has gone down significantly

[ayu.sh @flomodoro]

recitation03.pdf

[ayu.sh @flomodoro]

cool
https://t.co/AXWz2tNyEF

[ayu.sh @flomodoro]

https://t.co/hxsV1nROBS

[ayu.sh @flomodoro]

Just built an MCP server for @mataroa_blog, now you can manage your entire blog through natural language in Claude Desktop

[ayu.sh @flomodoro]

Workflow engines like @temporalio are the real agents
Being forced to find the boundary between deterministic and non-deterministic tasks, minimizes hallucinations and improves self-consistency
cc: "The Log is the Agent" @yoheinakajima

[ishy @ishygoel]

- joy sullivan

[ayu.sh @flomodoro]

To avoid making such mistakes in the future, I want a generalised prompt. This is v1.3 of that prompt.
https://t.co/DVMn0b8mZE

[ayu.sh @flomodoro]

While trying to take a decision, I ended up forgetting that the events that resulted in the decision were not independent. I realized that while attending an introductory about how LLMs are bad at causal inference, but good at making the building blocks (DAGs) that aid it.

[ayu.sh @flomodoro]

using LLMs to code gets me 80% of the way
Bigger the change, lower the percentage
Doing multiple validation passes gets me close to the ideal output, 80% -> 96% (80+20*0.8) etc
Nothing beats a set of equivalence class tests

[ayu.sh @flomodoro]

Magic today, edge cases tomorrow

[ayu.sh @flomodoro]

using crewAI is a strong negative signal

[ayu.sh @flomodoro]

Implementing @DSPyOSS provides some quick wins
1. DSPy.Refine - Intelligent retry mechanism with feedback
2. MultiChainComparion - enforce self-consistency through parallel llm calls and voting
3. Organising prompts as modules provides better results (personal anecdotal exp.)

[ayu.sh @flomodoro]

@DSPyOSS implementation provides some instant wiins
1. DSPy Refine - intelligent retries, policy enforcement
2. MultiChainComparison - self-consistency checks via voting mechanisms
Adding FewShot examples should also be a quick win

Still experimenting with GEPA, need 30 examples

[ayu.sh @flomodoro]

The goal of every consulting company is to sell you a json based DSL that they "learned" over implementations

[Dwarkesh Patel @dwarkesh_sp]

Wrote up some flashcards and practice problems to help myself retain what @reinerpope taught.

Hope it's helpful to you too!

Suggest more below and I'll add them.

https://t.co/2gQcO24uIW