Why do people bother with paid AI assistants like Grok, CoPilot, GPT, Gemini, and Claude, when this exists, it is FOSS, and it is currently beating out all the others in terms of raw reasoning power and token efficiency?Come in here and explain why you're a fucking retard still using Grok because you live your life out on twitter.
It's not nearly as capable as you're claiming. It's a less efficient, borderline distilled version of GPT 4.0
>>107745808Thats crap there are open source LLM miles better than deepseek, LLAMA included.
>>107746450>>107746469You're wrong and trying to mislead people for whatever reason.
i use it as a "good enough" workhorse for standard queries to save free-tier budget for paid models . it definitely has noticeable limitations like thinking it's 2024, and somewhat commonly replying in Chinese , and not building memory on past interactions . i haven't experimented with running it locally yet though , and will soon , and i expect these issues to be mitigated with local datasets / training .
>>107746815afaik none of the commercial AI assistants provide cross-conversational memory outside of ChatGPT and CoPilot, but with CoPilot you have to manually tell it "don't forget this" or "save this to memory for later" otherwise every new conversation you start is like you're a complete stranger. All the others, DeepSeek included, have no past retention of your other conversations.I'm not sure why this is such a common limitation that only GPT and CoPilot have bypassed. My only guess is maybe user security and privacy or something like that, but I dunno.
>>107745808I don't know. I've got a free account for deepseek and it's a permanent pinned tab in my browser. I would never use the western LLMs.
>>107745808I can't locally host a 271B parameter model lad.
