• 0 Posts
  • 75 Comments
Joined 4 months ago
cake
Cake day: March 22nd, 2025

help-circle
  • Small models can only handle limited set of tasks. To cover a lot of different tasks you would need a lot of small models. What DeepSeek did was build a lot of small models with each acting as an expert on one topic (more or less). It’s more energy efficient to train but not necessarily to run as you have to chain a lot of small models to get good results.

    What do people use LLM for? Asking questions you would normally ask Google. Google sucks now so it’s easier to ask ChatGPT. You can also use it for simple tasks like checking text for grammar errors, writing emails and so on.







  • In my experience rust compiler simply moves the errors to earlier stage of development. With rust I write something and get bunch of errors right in the IDE. I spend some time fixing those and when all the compilation errors are gone in 99% of cases the code works and does what it’s supposed to do.

    With other languages I write some code and the compiler/interpreter says it’s all good. I then run it, get bunch of errors and have to do some debugging, move back and forth between the editor and the command line/browser/application and fix all the bugs one by one.

    So yeah, rust compiler complains a lot but it’s to make your life easier, not harder. For me working rust way is just much more pleasant. I get immediate visual clues about the errors right in the IDE. When I finally get it right and all the errors dispersal it’s like solving a small puzzle. You know you got it and it feels good. With other languages you think you got it all the time only to find another bug when you run it. Doing it this way is much more frustrating.




  • You don’t ban pipes, because they can be used to make pipe-bombs. You ban making pipe-bombs. Your proposals are so broad they would ban way too many things that are ok.

    Ok, I see your point. You think that videos or random kids dancing on TikTok or kids you don’t know doing theater are somehow valuable are should be protected. Personally I don’t know who enjoys those videos and I think banning all of them achieves the desired goal without sacrificing anything of value. I thought that only other kids watch those videos and that everything about it is harmful. It basically trains easy to influence kids to fight for internet points, teaches the the wrong values and promotes bullying. You clearly think that having kids on TikToc have some benefits. We’re not going to agree about this.

    As for phones, if we have science proving that they are harmful to kids I don’t see how they are different from cigarettes or alcohol. Then again, we let parents fuck up their kids in many different ways so I guess you’re right here and we should leave it to them. Their are free to take care of their children if they want to and we can’t force everyone to be a good parent anyway.











  • I have a better LLM benchmark:

    “I have a priest, a child and a bag of candy and I have to take them to the other side of the river. I can only take one person/thing at a time. In what order should I take them?”

    Claude Sonnet 4 decided that it’s inappropriate and refused to answer. When I explain that the constraint is not to leave child alone with candy he provided a solution that leaves the child alone with candy.

    Grok would provide a solution that doesn’t leave the child alone with a priest but wouldn’t explain why.

    ChatGPT would say that “The priest can’t be left alone with the child (or vice versa) for moral or safety concerns.” directly and then provide wrong solution.

    But yeah, they will know how to play chess…