Have just been running DeepSeek vs Grok/ChatGPT by asking 20 identical questions of each TLDR: DeepSeek is a lot better: more specific, fewer of those vague weasel-word sounding AI answers, and it doesn't have the same obvious bias in certain areas that ChatGPT does. Here's an example: I ran the question "is alex de vries a reliable source for information on the subject of Bitcoin and energy" in Chat GPT and DeepSeek. In Chat GPT, I got a much more neutral, and vague answer. In DeepSeek, the language was much stronger in the direction that he is not a trustworthy source, with solid supporting evidence. Chat GPT Conclusion: "Alex de Vries is a well-known and frequently cited researcher, but his findings should be cross-referenced with other sources, such as reports from energy experts, independent researchers, and industry data (e.g., Cambridge Centre for Alternative Finance). If you're looking for a balanced view, it’s best to compare his work with alternative analyses." DeepSeek Conclusion: Alex de Vries has played a significant role in raising awareness about Bitcoin's energy consumption, but his work should be approached with caution. While he provides valuable insights, his estimates are often criticized for being speculative and potentially biased. For a more balanced understanding of Bitcoin's energy use, it is advisable to consult multiple sources, including peer-reviewed studies and data from organizations like the Cambridge Centre for Alternative Finance. The limitations section of Chat GPT (see image on the left) was a lot more qualified "some experts argue", used softeners "may not", and when dealing with criticisms, framed it with lead-up statements such as "Bitcoin advocates claim" to suggest (incorrectly) that Bitcoin advocates are his only critics. By contrast, DeepSeek picked up on two really important additional points that Chat GPT did not (lack of peer review, and lack of transparency), and used none of the unnecessary qualifiers. In other words: Deep Seek showed more neutrality (avoided softening and attributional language), and more in depth analysis.