michael s galpert

do things that increase serendipity

@runsonai @hnshah

I used Codex to fine-tune Qwen 3.6: /goal I want to increase the efficiency of my qwen 3.6 27b 4bit on this machine, do anything that increases it by at least 30% On my m3 ultra. Guess what...it worked! It found multiple tweaks to make it more efficient.

@steipete @sama

brb calling @sama

I've been using /goal for ~3 days on OpenClaw. - 13 runs. - Gazillion tokens. - Many, many PRs. The lesson isn't "i used /goal a lot." it's that /goal is not a "do my ticket" button. It's a constraint workflow. I want a keep the ship on course. A thread on what actually works 🧵

@blakeir Weekly

the strange (and amazing) thing about this moment is that our intuition for what’s possible keeps going out of date.

Me and my goblins on GPT 5.5 FAST

@ANoffsinger @stripe how long are you in town for?

I'll be @stripe Sessions in SF tomorrow! Give a shout if you're around and let's meetup

ClawCon Chicago x.com/i/broadcasts/1…

ClawCon is going to China next month! First event will be in Shanghai cohosted with @themu_xyz luma.com/clawconshanghai

2 different people i know started creating digital clones(?) of themselves in reels and ig ads and these digital representations look nothing like them but their content still does numbers. hashtag don’t hate the player hate the game.

@photobiogenesis @nickgraynews

Sprouting is one of the most overpowered way to improve food Most of the mineral content of nuts/beans/grains is locked up in insoluble phytate complexes, and the content of these is reduced by 50-80% in sprouting Sprouting reduces oxalate and lectin content by 60-90% in beans and grains On top of that, the content of B-vitamins and polyphenols increases significantly from sprouting Broccoli sprouts contain around 50x the concentration of active compounds like sulforaphane vs. broccoli

@sekachov yo this is great!

since i use obsidian pretty much for anything, including todos and collecting links, i needed a quick way to append any text to existing notes. so, meet markie. i'm too lazy to make a proper intro video, so here's a screen recording with a groovy jazz beat.

ClawCon Copenhagen x.com/i/broadcasts/1…

@swyx happy belated bday bud

@sama @tszzl when i felt the agi

"post-AGI, no one is going to work and the economy is going to collapse" "i am switching to polyphasic sleep because GPT-5.5 in codex is so good that i can't afford to be sleeping for such long stretches and miss out on working"

@kaiostephens @runsonai

I ran Qwen3.6:27b to optimize itself in a recursive loop on my home server. Over 26 hours it went from 2.3tok/s to 84.3tok/s decode. It began on the home server, found there was no NVIDIA GPU, detected a CPU/RAM setup with 24 CPU threads and 93 GiB RAM, a 9060xt 16gb, then installed Hugging Face tooling remotely and started pulling GGUF quantizations. It benchmarked remote llama.cpp / llama-server runs across quantizations and flags: Found existing Qwen3.5-9B-Q8_0.gguf Downloaded / tested Qwen3.6-27B GGUF variants Compared Q6_K, Q5_K_M, Q4_K_M Ran server benchmarks over SSH against localhost:8080 Tested thread count, context, batch size, n_ubatch, --no-mmap, and memory-related flags Researched further speed paths: lower quantization, NUMA, huge pages, native CPU builds, cache/KV, TurboQuant, DFlash, speculative decoding, and automated tuning 1,524 tool calls 367 artifacts 345 memory addition 804 browser-control calls All of this from a model that can run on your computer. You don't need a better model. You need a better harness.

@natolambert @garrytan @davemorin

@garrytan Yup atomproject.ai

America needs to go much harder on open source models

@joshu would love that. Will dm next time

@msg come visit the robots?

in Palo Alto today. What’s good?

Positivity geohot.github.io//blog/jekyll/u…

Andy Grove’s “Only The Paranoid Survive” is a shockingly prescient guide for the AI age.