Conversation
@kentindell I take the point, but that article opens with a terrible load of bollocks.
'Pretty much everyone in tech agrees'. Firstly, most people don't know jack shit, so that's not a good guide (hint: 'pretty much most americans agree climate change doesn't exist').
Secondly, LLMs are fundamentally incapable of doing what people claim of them. They are a clever trick, but if you _actually try to use them_ you see they 'hallucinate' very quickly.
And this 'hallucinating' is inherent to the technique. LLMs literally are not capable of knowing what is correct or not or modelling in any way what they're saying. They're a black box clever autocomplete that can never not be interpolating wildly where data points do not exist (hint: in nearly all creative activity trivial novelty exists).
Anybody sane who _actually tries this_ for more than 5 minutes hits up on the damn thing being dreadfully wrong. 'Pretty much everyone in tech' is assuming that LLMs will somehow magically stop having these issues (that are totally inherent to them) one day.
It's like using Eliza and thinking 'ok if people work on this enough it'll be perfect'. NO!
'Pretty much everyone in tech' (yes this rankled) is susceptible to hype and nonsense.
As to your follow ups - yes people are awful and might try to use this anyway. But for anything other than making chat bots even worse (sigh) this thing is going to arrive dead on arrival.
On the idea it will revolutionise IT work just makes me laugh... Another hint: github copilot has been around for a while and yet 'oddly' this kind of technique hasn't had a very big impact on programming at all. 'I wonder why'.
'Pretty much everyone in tech agrees'. Firstly, most people don't know jack shit, so that's not a good guide (hint: 'pretty much most americans agree climate change doesn't exist').
Secondly, LLMs are fundamentally incapable of doing what people claim of them. They are a clever trick, but if you _actually try to use them_ you see they 'hallucinate' very quickly.
And this 'hallucinating' is inherent to the technique. LLMs literally are not capable of knowing what is correct or not or modelling in any way what they're saying. They're a black box clever autocomplete that can never not be interpolating wildly where data points do not exist (hint: in nearly all creative activity trivial novelty exists).
Anybody sane who _actually tries this_ for more than 5 minutes hits up on the damn thing being dreadfully wrong. 'Pretty much everyone in tech' is assuming that LLMs will somehow magically stop having these issues (that are totally inherent to them) one day.
It's like using Eliza and thinking 'ok if people work on this enough it'll be perfect'. NO!
'Pretty much everyone in tech' (yes this rankled) is susceptible to hype and nonsense.
As to your follow ups - yes people are awful and might try to use this anyway. But for anything other than making chat bots even worse (sigh) this thing is going to arrive dead on arrival.
On the idea it will revolutionise IT work just makes me laugh... Another hint: github copilot has been around for a while and yet 'oddly' this kind of technique hasn't had a very big impact on programming at all. 'I wonder why'.
0
0
1
@kentindell honestly it fucking blows my mind that people don't seem to think it matters to check whether the thing works before writing big brained shit like this.
Like shouldn't that be your first consideration? People forget cold fusion? Or more recently crypto, the thing that was DEFINITELY GOING TO REVOLUTIONISE THE WORD right up until it didn't so down the memory hole?
Also I'm not willing to be nice about this author because he LITERALLY OPENS with that irritating 'most people in tech' thing. But of course these people always have to caveat it at some point because people are actually able to go and try this stuff and SEE THAT IT DOESN'T FUCKING WORK 🤣
So what you usually get is 'trust me bro all that will get fixed' or 'the next version sorts it all' or as in here 'well it doesn't work but I guess we'll see'. Honestly. Fuck off.
Sound familiar? What other deep learning technique have we heard this shit about?
Yes this pisses me off.
Also yes, depressingly (very depressingly) there will be big brained boys who will buy into this bollocks and just try and use it.
In the same way people take Musk at his word, turn on autopilot, shove something under the steering wheel so it thinks you have hands on then fall asleep in the back before that lovely ride into the side of a truck.
Like shouldn't that be your first consideration? People forget cold fusion? Or more recently crypto, the thing that was DEFINITELY GOING TO REVOLUTIONISE THE WORD right up until it didn't so down the memory hole?
Also I'm not willing to be nice about this author because he LITERALLY OPENS with that irritating 'most people in tech' thing. But of course these people always have to caveat it at some point because people are actually able to go and try this stuff and SEE THAT IT DOESN'T FUCKING WORK 🤣
So what you usually get is 'trust me bro all that will get fixed' or 'the next version sorts it all' or as in here 'well it doesn't work but I guess we'll see'. Honestly. Fuck off.
Sound familiar? What other deep learning technique have we heard this shit about?
Yes this pisses me off.
Also yes, depressingly (very depressingly) there will be big brained boys who will buy into this bollocks and just try and use it.
In the same way people take Musk at his word, turn on autopilot, shove something under the steering wheel so it thinks you have hands on then fall asleep in the back before that lovely ride into the side of a truck.
0
0
1
@kentindell :)) tbh I took that one quite literally, instant coffee is the stuff of the devil.
AND THAT WAS GOING TO REVOLUTIONISE COFFEE WASN'T IT?? 🤣
(sorry gone a bit caps mad this morning)
AND THAT WAS GOING TO REVOLUTIONISE COFFEE WASN'T IT?? 🤣
(sorry gone a bit caps mad this morning)
1
0
1
@kentindell I liked that analogy except I think it was far too optimistic. More like 3 cups seem ok, another 3 are obviously crap, the remaining 4 are covered in mould and on fire and only 1 of the 10 was actually coffee.
The one thing here that I think will save us is memeing. And people have already been relentlessly taking the piss out of chat gpt/bing/etc. where it gets things wrong and then 'corrects' itself even wronger (because it's just going on what 'sounds' like a good answer and has no sense of right + wrong).
As usual, humour and taking the piss is the best antidote to bollocks going.
BTW if you want a less caps lock angry man shakes hand at cloud take on this, https://blog.piekniewski.info/2023/04/09/ai-reflections/ is an excellent analysis from an actual expert in the field on the topic.
The one thing here that I think will save us is memeing. And people have already been relentlessly taking the piss out of chat gpt/bing/etc. where it gets things wrong and then 'corrects' itself even wronger (because it's just going on what 'sounds' like a good answer and has no sense of right + wrong).
As usual, humour and taking the piss is the best antidote to bollocks going.
BTW if you want a less caps lock angry man shakes hand at cloud take on this, https://blog.piekniewski.info/2023/04/09/ai-reflections/ is an excellent analysis from an actual expert in the field on the topic.
0
0
1
@kentindell Yeah he's fantastic. I defer to him on this area as the expert.
He was on here, not sure if he still is. I'd go look it up on his twitter account except Elon makes it so I can't do that any more :)
He was on here, not sure if he still is. I'd go look it up on his twitter account except Elon makes it so I can't do that any more :)
1
0
1
@kentindell I mean what's frustrating is that article opens like that. If it said instead 'this stuff probably won't work but even if it does it won't destroy jobs'.
I feel like a lot of the reason people do at least open articles like that is because it draws people in as there's a lot of hype and sadly a lot of delusion around this stuff right now and that's just how you get clicks.
It's remarkable how on some things people are incredibly skeptical but on others the brain plops out and all it takes is a snazzy demo and a bunch of SV liars and the thing is believed.
I do think an orange guy and an emerald heir have a lot to do with this
I feel like a lot of the reason people do at least open articles like that is because it draws people in as there's a lot of hype and sadly a lot of delusion around this stuff right now and that's just how you get clicks.
It's remarkable how on some things people are incredibly skeptical but on others the brain plops out and all it takes is a snazzy demo and a bunch of SV liars and the thing is believed.
I do think an orange guy and an emerald heir have a lot to do with this
0
0
0