3 ms·

All because ChatGPT is a language model, not an knowledge model. LLMs are designed to produce output that is similar (in language terms) to that its training s

All because ChatGPT is a language model, not an knowledge model. LLMs are designed to produce output that is similar (in language terms) to that its training saw in similar circumstances ... Yes, LLM work on many indistinct levels; Yes, they refer to the input you provide; Yes, LLMs produce results that are easily mistaken for intelligent or useful; but No, the LLM doesn't understand anything - it's merely producing output similar to that seen in training. https://www.newyorker.com/tech/annals-of-technology/chatgpt-is-a-blurry-jpeg-of-the-web https://www.newyorker.com/tech/annals-of-technology/chatgpt-... gives a really nice analogy. > At this point it seems clear to me that these language models are simply ELIZA combined with a method for auto-generating language rules from data instead of a hand-written script. Agreed - much more complex than ELIZA but not much smarter. And ELIZA fooled some of the people, some of the time; LLMs are fooling more of the people, more of the time. :(

version_five 4y ago
I posted this a couple months ago https://news.ycombinator.com/item?id=33781661 https://news.ycombinator.com/item?id=33781661 [Excitement over Chatgpt is] really not that different from Eliza being cool for a few minutes before being obviously found wanting and got what I though was a funny response of feigned amazement that I could think what way. It's funny to see the wave of hype starting to break.
- drpixie 4y ago
  ... but enormously disappointing to see the way so many people have been taken in. The number and strength of defensive statements on hacker news is appalling. Aren't we supposed to be the smart people who know what's going on ... not the gullible people being taken in. (But, as they say, a salesperson is the easiest person to sell to.)
- sharemywin 4y ago
  The thing is I've gotten useful work out of it. I'm not trying to say it's AGI or perfect, but it saves time on some things. And for coding I goto google and stack overflow less. maybe 8 out of 10 times I'm happy with ChatGPT. Since I mostly use it for coding I can verify pretty quickly if it worked out or not. for me it saves lot of typing.

version_five 4y ago

I posted this a couple months ago https://news.ycombinator.com/item?id=33781661 https://news.ycombinator.com/item?id=33781661 [Excitement over Chatgpt is] really not that different from Eliza being cool for a few minutes before being obviously found wanting and got what I though was a funny response of feigned amazement that I could think what way. It's funny to see the wave of hype starting to break.

drpixie 4y ago
... but enormously disappointing to see the way so many people have been taken in. The number and strength of defensive statements on hacker news is appalling. Aren't we supposed to be the smart people who know what's going on ... not the gullible people being taken in. (But, as they say, a salesperson is the easiest person to sell to.)
sharemywin 4y ago
The thing is I've gotten useful work out of it. I'm not trying to say it's AGI or perfect, but it saves time on some things. And for coding I goto google and stack overflow less. maybe 8 out of 10 times I'm happy with ChatGPT. Since I mostly use it for coding I can verify pretty quickly if it worked out or not. for me it saves lot of typing.