[44] Hady Elsahar - NLG from Structured Knowledge Bases (& Controlling LMs)
The Thesis Review
The Future of General Purpose Models
There is a spectrum here with respect to problems so if your model can put all the probability mass of its output space into toxic sentences there's nothing you could do right. Fortunately this is not the case like even GPT 3 knows trained majorly on English knows about other languages because of leakage in the data and stuff like that or so most more and more we're going on the general purpose models. But I would say it's rare to frame this problem as a control but I still have some thoughts on this maybe you could touch on the future part?
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
