[lebovic]

It's too late to prevent distillation of some capabilities, like writing code or finding vulnerabilities [1].

But an AI lab can continue to produce immense economic value without releasing the model publicly for potential distillation. For example, it could use a model solely in-house to develop therapeutics.

Hopefully there's a future where others can access frontier models, but it's not neccessary if preventing proliferation through distillation is considered more important.

[1]: See the notes on distillation in https://dualuse.dev/posts/export-controls-on-fable

[bandrami]

My long-term prediction for the sector is that frontier models will be so expensive that they will only be available for grant-funded projects at research institutions, like supercomputer clusters were 25 years ago.

[wqaatwt]

Why? Well it depends, most evidence is suggesting that Anthropic and OpenAI are making a lot of money on inference so the question is whether its more profitable for them to sell 100X tokens for Y, or 1X tokens for 100Y. In most industries with high fixed costs and low variable costs and unlimited scalability (like LLM providers) the first option ends up being much more profitable

[bandrami]

Literally nobody is making money on inference

[wqaatwt]

Based on what? There isn’t a lot evidence that’s the case..

Prices on OpenRouter for GLM and other large open models indicate that Anthropic/OpenAI must have pretty high gross margins even if their models are several times more expensive to serve.

It wouldn’t make sense for any provider to host large open models and then loss $10 on every $1 they make since they don’t have infinite VC money or any business model that would justify it.

[bandrami]

If they had high margins they wouldn't be issuing senior debt with a 18.5% coupon payment (and failing to fully subscribe it), nor would they need Elon to give them two months of free compute in order to appear profitable for a single quarter.

[wqaatwt]

We were talking specifically about inference and I don’t think there any indication that their gross margins on the API tokens (if not the personal subscriptions) are negative?

Obviously they have R&D and other fixed expenses that make the company itself highly unprofitable but that’s only semi-tangential.