Name recognition? Advertisement? Federal grant to beat Chinese competition?

There could be many legitimate reasons, but yeah I'm very surprised by this too. Some companies take it a bit too seriously and go above and beyond too. At this point unless you need the absolute SOTA models because you're throwing LLM at an extremely hard problem, there is very little utility using larger providers. In OpenRouter, or by renting your own GPU you can run on-par models for much cheaper.

At least in OpenAI's case, it raises the bar for potential competition while also implying that what they have behind the scenes is far better.

Zuckerberg explains a few of the reasons here:

https://www.dwarkesh.com/p/mark-zuckerberg#:~:text=As%20long...

The short version is that is you give a product to open source, they can and will donate time and money to improving your product, and the ecosystem around it, for free, and you get to reap those benefits. Llama has already basically won that space (the standard way of running open models is llama.cpp), so OpenAI have finally realized they're playing catch-up (and last quarter's SOTA isn't worth much revenue to them when there's a new SOTA, so they may as well give it away while it can still crack into the market)

I understand the rationale behind open sourcing llama.cpp. Because it has a lock-in effect.

But I don't see how open sourcing weights has a lock-in effect. In fact, it seems OpeanAI's open models can be run on llama.cpp. So by offereing them, they make llama.cpp even MORE useful. Instead of driving developers towards their own tech.

I believe it's to create barriers to entry and make the space harder to compete in.

There's still a ton of value in the lower end of the market by capability, and it's easier for more companies to compete in. If you make the cost floor for that basically free you eliminate everyone else's ability to make any profit there and then leverage that into building a product that can also compete at the higher end. This makes it harder for a new market entrant to compete by increasing the minimum capability and capital investment required to make a profit in this space.

Oh, wow. Following this line of thinking, what seems like a noble gesture on the surface is a scorched earth approach in reality.

They don't because it would kill their data scrapping buisness's competitive advantage.

Partially because using their own GPUs is expensive, so maybe offloading some GPU usage

LLMs are terrible, purely speaking from the business economic side of things.

Frontier / SOTA models are barely profitable. Previous gen model lose 90% of their value. Two gens back and they're worthless.

And given that their product life cycle is something like 6-12 months, you might as well open source them as part of sundowning them.

inference runs at a 30-40% profit