5 ms·
These models are designed to produce a _plausible_ text output for a given prompt. Nothing more. They are not designed to produce a _correct_ text output to a
by dotancohen 3y ago
These models are designed to produce a _plausible_ text output for a given prompt. Nothing more.
They are not designed to produce a _correct_ text output to a question or request, even if sometimes the output is correct. These proverbial stopped clocks might be correct more than twice a day, but that's just the huge training set speaking.
- drdeca 3y agoAre you taking the RLHF into account when you say so?
- alex_sf 3y agoTaking RLHF into account: it's not actually generating the most plausible completion, it's generating one that's worse.
- dotancohen 3y agoWell, I wasn't, but if you look at the top most comment of this thread [0] you'll see that considering the level of human reinforcement being demonstrated only reinforces my point. [0] https://news.ycombinator.com/item?id=36013017 https://news.ycombinator.com/item?id=36013017