Comment by echelon

> Interesting to see your observation where I have observed the opposite: posts that share big news about open-weight local models have many upvoted comments arguing local models shouldn’t be taken seriously and promoting the SOTA commercial models as the only viable options for serious developers.

When I argue this, my point is that FOSS shouldn't target the desktop with open weights - it should target H200s. Really big parameter models with big VRAM requirements.

Those can always be distilled down, but you can't really go the other way.