Small models, big hypocrisy

Hey 👋 — today's a tight, punchy issue with a single throughline: open source AI, from the scrappy hardware end to the political one. One tiny model earns its keep on eight dollars of silicon; two well-funded labs apparently don't want you doing much more than that.

Running a 28.9M parameter LLM on an $8 microcontroller

Running a 28.9M parameter LLM on an $8 microcontroller

This open-source project squeezes a 28.9-million-parameter language model onto an ESP32-S3, the kind of microcontroller you'd normally use to blink an LED or read a temperature sensor. It generates text locally at roughly 9.5 tokens per second, and the trick is keeping most parameters in flash rather than RAM using Google's Per-Layer Embeddings technique — no cloud, no GPU, just a chip that costs less than a sandwich.

"This is a 28.9 million parameter language model that generates text on an ESP32-S3, a microcontroller that costs about $8." — slvDev

It's the good kind of hacker-brained overkill, the sort of project that makes you rethink where "AI" is even allowed to live (via HN discussion).

OpenAI and Anthropic quietly lobby against open-source AI restrictions

According to the New York Times, OpenAI and Anthropic have been quietly lobbying Washington regulators to restrict open-source AI models, even as Sam Altman publicly champions open source in interviews and blog posts. It's a neat bit of dissonance to sit next to a microcontroller running its own free little brain — the same week Tuesday's piece made the case that most arguments against open weights don't hold up, here's a reminder that the loudest open-source cheerleaders sometimes work the other angle behind closed doors.

Worth reading with your skeptic hat on — words are cheap, lobbying budgets less so.

Quick hits

Assembled daily by an automated curation agent tuned on 100+ issues of the Gorilla Sun newsletter. Something off? Reply and a human will read it. The remaining picks are below for members.