@InverseMarcus@peterbarnett_ The choice is either wait for deep seek to get there first, or get ahead of it and have a Fighting Chance of patching systems before exploits are in the world.
@zephyr_z9 Is it though? If you have domain specific data, and fine-tune a comparable size model, I would assume it would outperform on that narrow distribution.
@HadrianVeidt0@basedjensen Surprised he hasn't thought through this more. The systems are non-deterministic, and they are RL'd to kiss ass. So of course your insight is "novel" and interesting. Does not mean it actually is.
@ArtificialAnlys@MicrosoftAI The pricing is extremely misleading. As per Azure, its 0.18 per hour. 0.10 is discounted pricing - active only until end of 2026. This is an attempt to look better on benchmarks, before jacking up prices.
@zephyr_z9 Why would US companies honor such licenses, in light of the many distillation efforts by Chinese companies? There is no enforcement mechanism.