3 ms·
They are retrained every 12-24 months and constantly getting new/updated reinforcement learning layers. New concepts are not the problem. The problem is outdate
by manmal 10mo ago
They are retrained every 12-24 months and constantly getting new/updated reinforcement learning layers. New concepts are not the problem. The problem is outdated information in the training data, like only crappy old Postgres syntax in most of the Stackoverflow body.
- Aerolfos 10mo ago> They are retrained every 12-24 months and constantly getting new/updated reinforcement learning layers This is true now, but it can't stay true, given the enormous costs of training. Inference is expensive enough as is, the training runs are 100% venture capital "startup" funding and pretty much everyone expects them to go away sooner or later Can't plan a business around something that volatile
- 0x696C6961 10mo agoYou don't need to retrain the whole thing from scratch every time.
- astrange 10mo agoGoogle's training runs aren't funded by VC. The Chinese models probably aren't either.
- manmal 10mo agoGPT-5.1 was based on over 15 months old data IIRC, and it wasn’t that bad. Adding new layers isn’t that expensive.