• 0 Posts
  • 89 Comments
Joined 2 years ago
cake
Cake day: June 30th, 2023

help-circle
rss


  • So are they loading every exam in its entirety into the same context window and doing the whole thing with a single prompt? Or letting the LLM take arbitrary actions instead of only returning a value? It seems like they would have had to make some pretty bad decisions for this to be possible, beyond maybe just adjusting the one grade. I wonder what the prompt jailbreak equivalents of sql injection would actually look like




  • That’s a great way to do it, but human attention on your code is a scarce and valuable resource. LLMs are great for the sort of lazy stupid questions where you benefit from a quick answer, but also don’t want to waste someone else’s time on. When you are learning nearly all the questions you’ll have will be like this, your progress is gated on finding the answers, and even if you are taking a class and it’s someone’s job to look at your code and help you understand what’s wrong with it, you have to wait your turn for that and only get so much help.



  • There was actually one published a few days ago that concluded that it can be effective:

    Participants with depression experienced a 51% reduction in symptoms, the best result in the study. Those with anxiety experienced a 31% reduction, and those at risk for eating disorders saw a 19% reduction in concerns about body image and weight.

    However the person who did the study shares your concerns:

    I asked Heinz if he thinks the results validate the burgeoning industry of AI therapy sites. “Quite the opposite,” he says, cautioning that most don’t appear to train their models on evidence-based practices like cognitive behavioral therapy, and they likely don’t employ a team of trained researchers to monitor interactions. “I have a lot of concerns about the industry and how fast we’re moving without really kind of evaluating this,” he adds.

    Also they did another article about difficulties and pitfalls of making these things



  • To me the disadvantage would be, the library likely does many more things than just what you need it for, so there is way more code, so you probably can’t realistically read and understand it yourself before incorporating it. This would lead to among other issues the main thing that irritates me about libraries; if it turns out something in it is broken, you are stuck with a much bigger debugging problem where you first have to figure out how someone else’s code is structured.

    Although I guess that doesn’t apply as much to implementations of common algorithms like OP since the library is probably solid. I would consider favoring LLM code over most anything off npm though.





  • I don’t think it’s actually such a bad argument because to reject it you basically have to say that style should fall under copyright protections, at least conditionally, which is absurd and has obvious dystopian implications. This isn’t what copyright was meant for. People want AI banned or inhibited for separate reasons and hope the copyright argument is a path to that, but even if successful wouldn’t actually change much except to make the other large corporations that own most copyright stakeholders of AI systems. That’s not really a better circumstance.






  • Thanks for the info, I’m on linux mint and after checking these out it isn’t immediately apparent from their websites whether or how I could install them. Still think etcher occupies a niche that alternatives don’t fill, its website directs you straight to installing it, it’s cross platform, and using it is very easy, so it’s something that could reasonably be linked to in various install tutorials.