Also, on the topic of Deepseek, I’ve moved on to using other models. Back when it came out was the first time I really got into using AI for certain things, and it was nice. I was using the V3 model because the thinking of R1 usually took a bit too long for me but would sometimes switch to it when the task required it. But last month when Kimi K2 came out I completely switched over to that one because it was as smart if not smarter than R1, but also as fast as V3. Since about a week ago though I’ve been testing out Qwen’s new model for my use cases and I think I’m going to be using it until something better comes along.
Also, I’d be interested to read that guide once you are done with it.
Yeah, it’s not horrible but when I tried to test it against the other models I’ve used and even “frontier” ones like Gemini 2.5 Pro, it just messed up too much for my liking.
I assume these are local models? What GPU do you run them with? I don’t even have my new gpu yet (soon inshallah) so there’s no point in me downloading a bunch of models yet lol.
Also, I’d be interested to read that guide once you are done with it.
We’ll put it up on prolewiki but I’m not sure how quickly it will get written. So far only one other person has participated, though two PW editors with experience in LLMs told me they’ll contribute soon. If you know the technical aspects deeply, I can add you to the project, it’s an etherpad collab document. My knowledge is very limited so I was able to write some parts, but others completely escape me. As a beginner’s guide I want it to basically teach everything from scratch, no matter how long the guide gets. People will jump to the sections they need to read anyway.
Good points.
Also, on the topic of Deepseek, I’ve moved on to using other models. Back when it came out was the first time I really got into using AI for certain things, and it was nice. I was using the V3 model because the thinking of R1 usually took a bit too long for me but would sometimes switch to it when the task required it. But last month when Kimi K2 came out I completely switched over to that one because it was as smart if not smarter than R1, but also as fast as V3. Since about a week ago though I’ve been testing out Qwen’s new model for my use cases and I think I’m going to be using it until something better comes along.
Also, I’d be interested to read that guide once you are done with it.
There’s also z.ai, but I haven’t got to try it yet.
I’ve tried it but haven’t really liked it.
Ah, bummer
Yeah, it’s not horrible but when I tried to test it against the other models I’ve used and even “frontier” ones like Gemini 2.5 Pro, it just messed up too much for my liking.
I assume these are local models? What GPU do you run them with? I don’t even have my new gpu yet (soon inshallah) so there’s no point in me downloading a bunch of models yet lol.
We’ll put it up on prolewiki but I’m not sure how quickly it will get written. So far only one other person has participated, though two PW editors with experience in LLMs told me they’ll contribute soon. If you know the technical aspects deeply, I can add you to the project, it’s an etherpad collab document. My knowledge is very limited so I was able to write some parts, but others completely escape me. As a beginner’s guide I want it to basically teach everything from scratch, no matter how long the guide gets. People will jump to the sections they need to read anyway.
Tbh, I just use their chat interfaces. Haven’t really felt the need to do a local setup. But yeah they can be setup locally.
Edit: sorry I somehow skipped over your question. Sadly I don’t have much knowledge on the technical aspects of ai so I wouldn’t be much help there.
Oh cool, I’ll try them out then