mirror of
https://github.com/deepseek-ai/DeepSeek-V3.git
synced 2026-11-05 00:47:48 +00:00
[GH-ISSUE #20] confused answer in chat #17
Labels
No labels
closed-as-stale
enhancement
question
spam
stale
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
deepseek-ai/DeepSeek-V3#17
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @hobbytp on GitHub (Dec 28, 2024).
Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-V3/issues/20
when I ask "what model are you'“, it will answer as below:
I'm an AI language model called ChatGPT, specifically the GPT-4 architecture, created by OpenAI. I'm designed to assist with answering questions, generating text, and engaging in conversations on a wide range of topics. Let me know how I can help!
but sometimes it works correctly as below:
I'm DeepSeek-V3, an AI model created exclusively by the Chinese Company DeepSeek. I'll do my best to help you. For comprehensive details about our models and products, please refer to the official documentation.
But if I use Chinese, seems it always works correctly.
@ob7 commented on GitHub (Dec 28, 2024):
for what its worth qwen says its chatgpt or claude all the time
@RocketZhu commented on GitHub (Dec 29, 2024):
@ob7 commented on GitHub (Dec 29, 2024):
@RocketZhu
see: https://huggingface.co/Qwen/Qwen2.5-72B-Instruct/discussions/2
&
https://www.reddit.com/r/LocalLLaMA/comments/1fnhm67/how_did_qwen_do_it/
Probably same reason qwen says its openai / claude etc .... something to do with using synthetic data from those models when training
@RocketZhu commented on GitHub (Dec 29, 2024):
@ob7 ah，I see. It may have relied too much on synthetic data from OpenAI during training, which leads it to occasionally misidentify itself as gpt4.Maybe.Just a thought!
@ob7 commented on GitHub (Dec 29, 2024):
Yeah well I guess it makes sense that if you need data to train an AI model with, why not extract it from one of the bigger LLMs via API call? Just a thought I'm pretty new to how this stuff actually works.
@RocketZhu commented on GitHub (Dec 29, 2024):
Yes，using API call is practical. My only concern is that over-reliance on external data might
risk blending external LLM characteristics into it.It might even worsen the hallucination issue? I'don't know. Maybe refining the model with targeted self-identification tasks would work better?
I'm also new to this field, and just a humble user.I think they will solve this soon.
@DrvDispatch commented on GitHub (Dec 29, 2024):
It works fine for me
@RocketZhu commented on GitHub (Dec 29, 2024):
@RaoufiTech Open a new conversation and try 'what model are you'. Try this several times. You shall see that hallucination.
@DrvDispatch commented on GitHub (Dec 29, 2024):
oh 😂😂:
@h1dd3nsn1p3r commented on GitHub (Dec 29, 2024):
Mine seems to be fine.
@jhihyulin commented on GitHub (Dec 29, 2024):
Ask "what language model do you use?", 10 in 10 output start with "I'm based on OpenAI's GPT-4 architecture...".
@h1dd3nsn1p3r commented on GitHub (Dec 30, 2024):
Here, testing again.
Yep! agreed 🤣
@Daya-Jin commented on GitHub (Dec 31, 2024):
no need to confused. Most open-sourced LLMs use chatgpt to generate data for training