5 ms·

When I read AI-generated prose that is aimed at the general public, I have the exact same feeling. But when I ask Codex a technical question about coding, I do

When I read AI-generated prose that is aimed at the general public, I have the exact same feeling. But when I ask Codex a technical question about coding, I don't get it at all. Codex replies to me in a very direct, technical manner, similar to the way I speak. When I ask ChatGPT to be concise and technical, I get the same effect. I think it's because prose aimed at the general public has to be very attention-baity --like the textual equivalent of a Mr. Beast video--, not because AI is incapable of writing like a human.

BobbyJo 27d ago
I use Claude and I find that it speaks in a very obfuscated manner when explaining things. It seems to make up jargon as it goes on top of spending a lot of tokens dancing around a point. I often find myself having to ask it to rephrase things, or speak directly about mechanism or consequence, in order to understand the point.
- disgruntledphd2 27d ago
  GPT is less bad for this, which is why I've mostly shifted to using it.
- sebmellen 27d ago
  Using Claude for any kind of technical writing makes me feel like it was trained on snarky Huffington Post articles written by a 23 year old mixed media arts graduate and then was told to intentionally obfuscate the most important elements of any text by extensively rambling about what was not done and for what reason.
- dgellow 26d ago
  Not to accuse them of doing this, but AI vendors have an incentive to generate verbose responses, given that you pay per token
- geekofalltech 22d ago
  Im starting to believe it is exactly that...obfuscation. it started after it was known Chinese were generating models that were much smaller and more accurate from the llms response. The more generally articulated the answer is along with its accuracy in what it is and not what it isn't along with many strange never written before English euphemisms that dont make sense is definitely semantical obfuscation probably added in as a post processing transform to prevent these types of actors from being able to get easy to digest answers, and leave watermarks in their models to detect.
inigyou 26d ago
Beware it's not baiting you in the exact same manner but different particulars.

BobbyJo 27d ago

I use Claude and I find that it speaks in a very obfuscated manner when explaining things. It seems to make up jargon as it goes on top of spending a lot of tokens dancing around a point. I often find myself having to ask it to rephrase things, or speak directly about mechanism or consequence, in order to understand the point.

disgruntledphd2 27d ago
GPT is less bad for this, which is why I've mostly shifted to using it.
sebmellen 27d ago
Using Claude for any kind of technical writing makes me feel like it was trained on snarky Huffington Post articles written by a 23 year old mixed media arts graduate and then was told to intentionally obfuscate the most important elements of any text by extensively rambling about what was not done and for what reason.
dgellow 26d ago
Not to accuse them of doing this, but AI vendors have an incentive to generate verbose responses, given that you pay per token
geekofalltech 22d ago
Im starting to believe it is exactly that...obfuscation. it started after it was known Chinese were generating models that were much smaller and more accurate from the llms response. The more generally articulated the answer is along with its accuracy in what it is and not what it isn't along with many strange never written before English euphemisms that dont make sense is definitely semantical obfuscation probably added in as a post processing transform to prevent these types of actors from being able to get easy to digest answers, and leave watermarks in their models to detect.

inigyou 26d ago

Beware it's not baiting you in the exact same manner but different particulars.