3 weeks ago, @runlayer sued Rippling to block our launch of a competing MCP Gateway. They incinerated $ millions of their VC funding, then dropped the whole thing like a hot potato, for nothing (see our filing, below). So today, we’re dropping our new MCP Gateway. It’s better. 1/

Today we're also opening the weights for Muse Glimmer, a great 30B parameter dense model that can run locally. Soon we'll also release the weights for Muse Spark 1.2, our latest foundation model. Meta is a strong supporter of open source and I'm proud of these releases. Congrats to @alexandr_wang and the MSL team for all your great work on these models.

The growth of synthetic data companies has been underwhelming. They underestimated how critical human intelligence is. AI cannot reliably measure the frontier of its own mistakes definitionally. It would be like asking students to all grade their own homework. This translates to reward hacking in synthetic tasks. Labs also in-house purely synthetic data pipelines. Many of these companies experienced a revenue sugar high, then stalled because their data creation had no moat. Durable value comes from solving hard problems. It's extremely challenging to organize millions of human experts to efficiently train frontier intelligence. There are network effects and extreme operational complexity. That is the reason it's valuable.

Anthropic very publicly claiming there needs to be a crackdown on distillation is puzzling. It's a ~$1T company with incredible resources and the "distillation attacks" at the scale they describe would seem easy to detect. Only cost would be giving up the associated API revenue.