4 ms·
use of words like 'figure' and 'know' are not applicable. GPT exposes the statistically most likely outcome, from a model. the model doesn't know either: the t
by ggm 4y ago
use of words like 'figure' and 'know' are not applicable.
GPT exposes the statistically most likely outcome, from a model. the model doesn't know either: the training is people, who weight the model and tune the heuristics to satisfy goals THEY impute.
If you train GPT to 'believe' there are 380 degrees in a circle, thats what it 'figures' out from.
There is no believe no figure and no know here.
- stavrianos 4y agoI keep seeing people talking about chatgpt hallucinating when it's wrong, but not when it's right. Maybe I've misunderstood, but isn't it just always hallucinating? It's not like the failure-mode is meaningfully different from the successes, except insofar as whether we agree with it, right?
- aezart 4y agoI agree. A stopped clock is right twice a day, and that's equally true with these LLMs.
- ggm 4y agoHallucinating is no different to believe or know or infer: its the wrong label to use for what it does. Yes, colloquially what it does is hallucinate all the time, and sometimes it lucid. But more factually no, it doesn't hallucinate because there is no "it" there, it's not conscious and you need to have a brain, to hallucinate. There's no "there" there. That is the whole of my point: we're using the wrong labels to describe what is happening. When it comes to explaining and describing "it's like" is one of the WORST ways to go. explanation by analogy or metaphor is a trap. "atoms are like billiard balls BZZZT next" "cells are little bags of water BZZZT next" "panadol 'kills' the pain BZZT no, it doesn't kill anything next"
- pulvinar 4y agoAt what point would it be ok to use these words? Will there be some threshold we'll all recognize? Or maybe a bill of rights to force the issue? I predict acceptance will go the way the ether disappeared-- advancing one funeral at a time.
- starfallg 4y ago'Hallucination' is the term LLM researchers use, and is arguably inaccurate based on the dictionary definition. The actual term for this for behaviour in people is 'confabulation', which is a lot more accurate.
- ggm 4y agoOh well.. if the LLM researchers are going to coin terms of art, well and good but I dislike this intensely because it invites belief it implies AGI and brain when in fact, its synthesis of new state from a model. This coining terms of art thing isn't uncommon. Think "brutalist architecture" and remind yourself its "en brute" == raw from the french. It has nothing to do with how "brutal" people think concrete is.
- airstrike 4y ago> This coining terms of art thing isn't uncommon. Think "brutalist architecture" and remind yourself its "en brute" == raw from the french. It has nothing to do with how "brutal" people think concrete is. That's not the full story. See https://www.tate.org.uk/art/art-terms/b/brutalism https://www.tate.org.uk/art/art-terms/b/brutalism > The term was coined by the British architectural critic Reyner Banham to describe the approach to building particularly associated with the architects Peter and Alison Smithson in the 1950s and 1960s.The term originates from the use, by the pioneer modern architect and painter Le Corbusier, of ‘beton brut’ – raw concrete in French. Banham gave the French word a punning twist to express the general horror with which this concrete architecture was greeted in Britain.
- ggm 4y agoPunning on compsci is all too common. "Peer" in bgp has two meanings. It doesn't help. Heap memory. Stack frames. Data bus. Dale Spender wrote about the unfortunate use of killing processes or aborting runtime jobs.
- SanderNL 4y agoRight. We are the ones applying the label after the fact.
- l33t233372 4y ago> There is no believe no figure and no know here I raise an issue with “no figure,” because that would mean that the model is unable to create novel structures and information. It certainly is able to do that.
- ggm 4y agook. I'll concede on 'figure' if we work on the synonyms like 'compute' or 'derive' (although I struggle with that) So whats a good synonym for 'figure' which doesn't imply intentionality and thought?
- l33t233372 4y agoIt’s hard, because previously the only things that could create novel structures were humans, and humans certainly have intentionality and thought.
- Turing_Machine 4y ago> If you train GPT to 'believe' there are 380 degrees in a circle, thats what it 'figures' out from. And a human wouldn't do that, you're saying? I'm not sure I buy that. 360 degrees per full revolution is purely an arbitrary human invention. Math would work just as well if it had been 380, or 50, or 50,000 (example: trig works just as well in radians as well as in degrees).