this post was submitted on 02 Mar 2026
68 points (92.5% liked)

Memes

54806 readers
1253 users here now

Rules:

  1. Be civil and nice.
  2. Try not to excessively repost, as a rule of thumb, wait at least 2 months to do it if you have to.

founded 6 years ago
MODERATORS
 
you are viewing a single comment's thread
view the rest of the comments
[–] HiddenLayer555@lemmy.ml 1 points 19 hours ago* (last edited 19 hours ago)

Interestingly I've gotten the 32B local Deepseek R1 model to say it's Claude as well (in English).

IDK if it's indicative of distillation or "model theft" though (how do you distill a closed model like Claude?) And why would them training on prompt responses from Claude also carry the name over unless Claude says its name every time? From my uneducated guess it's probably more of an indication that LLMs don't actually know anything and the information they're trained on mentions every AI model so they just randomly "pick" the most commonly mentioned ones for what they think they are if you don't tell them. Deepseek is a common Chinese model so it's probably associated with asking names of models in Chinese, same with Claude and English.