DeepSeek-V3 sometimes identifies itself as ChatGPT when asked which model it is; some speculate that its training datasets may contain text generated by ChatGPT
The report adds an identity-level failure to an emerging record of reliability concerns around DeepSeek: later testing found its chatbot could hallucinate citations and URLs, while a hands-on review described powerful output alongside rampant hallucinations. hallucinated citations and URLs and reported hallucination issues make model self-description a practical trust signal, not a cosmetic detail.
It also arrives as OpenAI has adjusted ChatGPT product communication in response to DeepSeek and other rivals, including an o3-mini update meant to help users understand model reasoning. OpenAI's response to DeepSeek-era competition raises the stakes for clear differentiation between assistants.
First-order effects
DeepSeek-V3 users cannot treat the model's answer to a basic identity question as reliable; the reported ChatGPT attribution is an observable product-quality issue.
The behavior prompts scrutiny of DeepSeek-V3's training-data provenance, but the report's dataset explanation remains speculation rather than established evidence.
Second-order effects
Developers and enterprises evaluating DeepSeek will place more weight on reproducible tests, documentation, and output controls instead of relying on the chatbot's own claims about itself.
The incident gives competing assistant providers another reason to distinguish their models through product transparency and reliability features, alongside raw capability.
Third-order effects
As model outputs increasingly circulate as training material, provenance and attribution may become harder to infer from behavior alone; credible disclosure and independent evaluation become more important differentiators.
If identity errors and citation hallucinations persist across assistants, the market is likely to treat trustworthiness as a deployment requirement rather than a secondary chatbot feature.
The trend: This is one data point in the shift from comparing AI assistants mainly on capability toward evaluating them on provenance, reliability, and operational trust.
I have been using DeepSeek API services for a while now. Here are my experiences: 1. Positive: Their API services are very stable. I never experienced any errors, even when making 1,000 concurrent calls. 2. Negative: They lack professionalism as a service provider. …
This actually reproduces as of today. In 5 out of 8 generations, DeepSeekV3 claims to be ChatGPT (v4), while claiming to be DeepSeekV3 only 3 times. Gives you a rough idea of some of their training data distribution. [image]
It is interesting the degree to which OpenAI is responsible for making its competitors possible: 1) They launched the first public big LLM, showing they are powerful 2) Many models trained on GPT data👇 3) Showed off a lot of easily copied ideas (eg test-time compute is key)
It's not too hard to probe a model and deduce how the various parts of its training were done. Once you know that, making it behave a certain way (eg bypass its safety tuning) can be relatively easily: [image]
In case you are reading today about the Chinese model Deepseek and theories that it may have trained on someone else's data ... https://www.nytimes.com/...
it is (relatively) easy to copy something that you know works. it is extremely hard to do something new, risky, and difficult when you don't know if it will work. individual researchers rightly get a lot of glory for that when they do it! it's the coolest thing in the world.
If DeepSeek-V3 is good because it trained on ChatGPT (which of course it did), why isn't Grok amazing? Why isn't *every* model amazing? Why spend 95% of compute pre-training a new model (which equals 405B on Pile-test btw) if the secret sauce is ~fOrBiDdEn~DaTa~ in the last 5%?
It's funny that SOTA models like deepseek will gladly output they are trained by OAI (1.5 years after gpt-4) but maybe not so funny... how hard could it be to steer the model in post training to not admit it is trained by OAI/GPT-4? It's obvious the model will output this when [i…
I saw a post yesterday that I think showed DeepSeek being asked what version it was (or something like that) and it said that it was ChatGPT trained by OpenAI, implying either stolen data or a stolen model. Was that a fabrication? Can anyone verify?
DeepSeek v3 is an order of magnitude cheaper because it likely trained on frontier model outputs, in obvious violation of ToS ToS laundering by training on DeepSeek outputs is impossible to prevent. Does not bode well for economics of training frontier models