註冊並分享邀請連結,可獲得影片播放與邀請獎勵。

Nick Almond
@DrNickA
PhD in Physics. Learning Theorist. Governance Designer. Head of Governance Jito. Chief Scientist Factory Labs. All tweets are just my opinion.
加入 September 2010
7.7K 正在關注    21K 粉絲
Nope, it's a great regulatory capture tactic. Scaling laws dictate that each advancement of the frontier with a new major pre-training run (which coincidentally they've all just done) gets astronomically more expensive. Fable / Astra likely cost $2B to train, it's likely $10B next time and it's up only from there. "pacing the frontier" allows the labs to extract rents from their latest models for longer before a competitor steams past them rendering their multi-billion dollar models obsolete and they get forced into the next run. They simply agree not to do the next run and spend their money and time on product, distribution and lobbying governments to protect their moats.
顯示更多
0
8
204
16
轉發到社區