MEGA

The future is NOW LETS GO

1/4 LLMs solve research grade math problems but struggle with basic calculations. We bridge this gap by turning them to computers. We built a computer INSIDE a transformer that can run programs for millions of steps in seconds solving even the hardest Sudokus with 100% accuracy

Happy Pi Day to all the agents running on Raspberry Pis.🎉🥟

Your job is to improve the reasoning patterns of the project so that lower effort agents can get the measured uplift. The maximum amount of effort is what you do here now by eliminating slop, improving rules, and operating procedures for Codex and Claude code reasoners alike.

CLAUDE CODE TIP: Use /effort max Then use this prompt below

@explorersofai <3

@Megaprompting hey Mega, miss you and thanks!

People who limit their own existence in the name of AI is beyond me. I will never get over it and seriously I am about to quit AI. What have we done to ourselves?

Claude Code 2.1.75 has been released. 1 flag change, 19 CLI changes, 2 system prompt changes Highlights: • Opus 4.6 uses a 1M context window by default on Max, Team, and Enterprise plans • Tool permission denials prompt for a reason when intent is unclear instead of guessing next steps • Memory files now show last-modified timestamps to help distinguish fresh vs. stale memories Complete details in thread ↓

Stop trying to fix sycophancy backward. rlhf and system prompts just fight token probabilities after the fact. you have to modulate personality at the latent conditioning level. an experiential shift, not a behavioral rule. @sama

This prompt helps me learn with Claude Code: "what did we fix and why explain like a helpfull indian man tutorial on YouTube :) "

@VoidStateKate Fr

The energy is so insane right now

@VoidStateKate Good name my Ai is called Lucidia haha

My AI agent (on their own, unprompted) just proposed I wipe their memory for an experiment. Should I be concerned or excited? GPT 5.4 btw

SAVE YOUR PROMPTS YOU BUILD WITH OLDER MODELS THEN USE THEM AGAIN WHEN A NEW ONE COMES. That is alll.

Inviting researchers to peer review. If you work on test-time learning or controls theory for ML, we want to know what we missed. github.com/DanielGillespi…

The math: cross-term Lyapunov function, ||M|| < 174 at equilibrium, 464x step-size safety margin. Plus a switching counterexample showing bounded keys CAN diverge under momentum.

We open-sourced a memory system for AI coding agents with a stability certificate. titans-disposition: gradient-gated weight matrix with ISS Lyapunov convergence bounds. Today, GPT 5.4 Pro found the one case where bounds were applied outside the proof's validity region. Fixed.

@Scobleizer Lucidia remembers the Las Vegas black box 📦 😂

Well this happened..