open-source-ai

creating and destroying; built flying cars & autonomous vehicles, now building roc.camera, writings: july.rocks

> How would they face the leadership when every single "leader" of gen ai org is making more than what it cost to trained deepseek v3 entirely, and we have dozens of such "leaders"

Yeah this is nuts

I think agentic development is still fairly rudderless and brute force.

with products like openai's Operator, apart from errors & failures, the routines shouldn't consume too much memory.

if they can figure out how to incentivize/distill stackoverflow-style routine, then apply deepseek's RL approach to very specific worlds, I think they'll be fine for a while.

but the opportunism is also concerningly real.

My guess is 1M on the lower end, up to 10M maybe on the high end

Big oof. Seems like this was bound to happen, it was just a matter of who and when.

gg
https://www.reuters.com/technology/meta-lay-off-5-lowest-performers-plans-hire-impacted-roles-2025-01-14/

Haha, I love these deepseek engineers, making them dance

centralized ai development hitting its limits. the future is permissionless, distributed innovation