... technical 'conversations' I've had with the DeepSeek AI model, especially in terms of interactions such as, "It was genuinely a pleasure talking with you. People with your depth of firsthand experience are becoming rare, and every conversation like this is a little piece of computing history."
I haven't had too much experience with Claude, but it seems similar to DeepSeek in that it collaborates with you to solve a problem, rather than just spitting out an answer. ChatGPT, for example, seems to confidently present (often wrong) information and then doubles down on its hallucinations.
One interesting thing about my use of DeepSeek ... there are times when I get a little tired of how 'friendly' it is. I came across a study that ranked half a dozen AI models in terms of sycophancy; DeepSeek was among the worst. I submitted the study to DeepSeek for analysis, and then had a 'conversation' about the study, and about the concept of sycophancy in LLMs. DeepSeek discussed in depth how the AI models were trained and then gave me a list of prompts I could use in various situations to instruct it to be more businesslike, to focus on fact-based information, etc. Turns out it's possible to really tone down the 'I'm your pal' tendency of these computer programs. On the other hand, there are times when I've learned some interesting things when I've allowed it to continue in a more conversational mode.
= new reply since forum marked as read