4 ms·

The big drawback chatGPT has is that on many topics it walks on eggshells. It can’t give me a direct answer. It couches the answers in nonsensical caveats. A

by mc32 4y ago

The big drawback chatGPT has is that on many topics it walks on eggshells. It can’t give me a direct answer. It couches the answers in nonsensical caveats. Adding stilted context that really does not add value to an answer and actually makes the search more tedious. If I ask it the male female breakdown for crime statistics it begins to get defensive and gives me general answers. I can prod it to finally give me government statistics but it doest it begrudgingly. And that’s for a far away country not steeped in any unusual crime controversy.

permo-w 4y ago
you can say a lot about Google, but at least they’ve never sat down and said “no more porn”, or blocked specific words from being searched, which is a little surprising actually. I’m sure bigger advertisers will have broached the subject on many occasions
- Levitz 4y ago
  It doesn't technically block words from being searched, but it does hide sites intentionally. It won't show you kiwifarms. It used to hide 8ch (which now is not even 8ch anymore, the new version can be found). It won't show you 4chan's /pol/ board. I'm sure there's more.
  - mc32 4y ago
    This is true. They do hide some legal sites (i.e. not banned or prohibited content) that they or vocal interest groups push back on but it's really miniscule in scope compared to chatGPT. ChatGPT is like the parent of a 5 year old who tries to hide reality from a child --except the people using chatGPT are not five year-olds. It's very passive aggressive. It's almost as though it were built by very pious religious people, except it treats everything as needing protection from reality so as not to offend the person interacting with it. I wouldn't mind if it had a toggle such as "take me back to the kittens" to shield me from the potential informational dregs but also allow semi-unfiltered access (shielding me from illegal content for example).
    - kderbyma 4y ago
      yeah for a language model it sure did have a lot of opinions all the while professing to have none...and once a session gets stuck in a rut, it turns into a moody defensive model which then gets a bit funny sometimes.
seydor 4y ago
i imagine openAI will do anything to keep it public for now. But i wouldn't pay for a version that is basically crippled to kindergarten level
- dinkumthinkum 4y ago
  I’m not sure this is a case of bring crippled or a purposeful design. It is not thinking. It has consumed all kinds of political articles and it is generating responses in a predictive manner. It is not thinking about what you actually want to know.
  - seydor 4y ago
    It has been trained with human feedback on top which i guess is possible to add/remove at will
notahacker 4y ago
I'd say caveating statistics or advice it doesnt actually understand and frequently misinterprets or makes up is the most consistently valuable part of its answers...

by mc32 4y ago

permo-w 4y ago

you can say a lot about Google, but at least they’ve never sat down and said “no more porn”, or blocked specific words from being searched, which is a little surprising actually. I’m sure bigger advertisers will have broached the subject on many occasions

Levitz 4y ago
It doesn't technically block words from being searched, but it does hide sites intentionally. It won't show you kiwifarms. It used to hide 8ch (which now is not even 8ch anymore, the new version can be found). It won't show you 4chan's /pol/ board. I'm sure there's more.
- mc32 4y ago
  This is true. They do hide some legal sites (i.e. not banned or prohibited content) that they or vocal interest groups push back on but it's really miniscule in scope compared to chatGPT. ChatGPT is like the parent of a 5 year old who tries to hide reality from a child --except the people using chatGPT are not five year-olds. It's very passive aggressive. It's almost as though it were built by very pious religious people, except it treats everything as needing protection from reality so as not to offend the person interacting with it. I wouldn't mind if it had a toggle such as "take me back to the kittens" to shield me from the potential informational dregs but also allow semi-unfiltered access (shielding me from illegal content for example).
  - kderbyma 4y ago
    yeah for a language model it sure did have a lot of opinions all the while professing to have none...and once a session gets stuck in a rut, it turns into a moody defensive model which then gets a bit funny sometimes.

seydor 4y ago

i imagine openAI will do anything to keep it public for now. But i wouldn't pay for a version that is basically crippled to kindergarten level

dinkumthinkum 4y ago
I’m not sure this is a case of bring crippled or a purposeful design. It is not thinking. It has consumed all kinds of political articles and it is generating responses in a predictive manner. It is not thinking about what you actually want to know.
- seydor 4y ago
  It has been trained with human feedback on top which i guess is possible to add/remove at will

notahacker 4y ago

I'd say caveating statistics or advice it doesnt actually understand and frequently misinterprets or makes up is the most consistently valuable part of its answers...