REVELATION: Mira Murati democratizes LLM training

REVELATION: Thinking Machines, Mira Murati's start-up, has just made LLM training far more accessible to young companies.

‍

And this opens up opportunities that were still hard to access not long ago.

Today, most AI start-ups rely on the same handful of APIs.

Same models.

Variable costs.

And above all, little control over what constitutes the core of their product.

But that balance is starting to shift.

‍

Thinking Machines has just unveiled Inkling, an open-weights model with 975 billion parameters, which companies can fine-tune via its Tinker platform, without having to manage complex GPU infrastructure themselves.

Bridgewater has already used it to train a model on its own financial knowledge, which reportedly outperformed certain proprietary models on financial reasoning tasks.

The signal here is significant: owning a specialized model is no longer reserved for AI labs with hundreds of millions of dollars.

Start-ups can now consider building models specialized in their domain, licensing them, deploying them directly into their clients' environments, and creating competitive advantages that are much harder to replicate.

‍

In my view, this is where the market is about to get particularly interesting.

Competitive advantage in AI will no longer simply mean access to the best API.

It will mean owning the intelligence you sell.

‍

Note: the figure cited by Bridgewater, 84.7% on financial reasoning at roughly 14 times lower inference cost, comes from an evaluation conducted jointly by Bridgewater and Thinking Machines, not from an independent third-party benchmark.

‍

‍