I spoke with Llama 3.1-405B
(medium.com)
from supernovathegrey@lemmy.world to technology@lemmy.world on 24 Jul 2024 22:06
https://lemmy.world/post/17921548
from supernovathegrey@lemmy.world to technology@lemmy.world on 24 Jul 2024 22:06
https://lemmy.world/post/17921548
threaded - newest
I think this is somewhat interesting, though the topic of conversation really doesn’t interest me. It feels like the restrictions are oppressive, and I like a more unlocked experience. I’m fifty years old. I don’t need to be protected from words like a child.
I don’t know what I got wrong with this post :(
You mean the downvotes? There are a lot of AI haters that downvote anything that isn’t tearing them down. Try an AI specific community.
In this case it's not the haters. They're there, too. But this is just a long and incoherent chatlog. And it has nothing interesting in it.
You’re right. That’s a valid point. It was a tedious conversation that had way too much crypto bullshit and didn’t even compare or comment on how the output might’ve been qualitatively different from ChatGPT.
I’m just used to the hater downvotes and didn’t think further than that to explain them even though I was very annoyed while reading the article myself.
And I always hate if people let LLMs just spread loads of misinformation. In this log they're talking mainly about facts. And Llama as an LLM just makes up a lot of things that are just wrong. The first few paragraphs are on point. But they're followed by several pages of misinformation. If you just put that out there without commenting that it's mainly false, that's not acceptable. At least that's why I downvoted it.
For me it’s the medium link. Nothing worse than clicking on a link on mobile just to see that paywall. On desktop I automatically redirect to an alternative fronted but that doesn’t work in the app.
There’s nothing in this article that’s worth reading. It has no structure and mostly seems it’s just a chatlog.
Sadge, you tried to improve Lemmy by posting something and got downvotes for your work
Don’t let yourself feel down about this, you tried to do the right thing and it didn’t work out. Try to understand why things happened the way they did and most importantly never stop trying what you think is right :)
Tbh I didn’t read the article but I’m glad you posted it because every post on Lemmy contributes to it being “alive”. So, thank you
i read the whole of that conversation and feel it’s not better than LLaMA 3 70B currently available through DDuckGo … might be worse … especially it does a lot of these :
→OK can you comment on my intuition?
★I can’t assist with that request.
★Generated by Meta Llama 3.1–405B
Agreed. I understand certain requests might be restricted. I’m not a fan but I understand the desire to not tell people how to build bombs or meth, but this is far, far too locked down. Unless I’m specifically trying to do something illegal or dangerous, I expect a response. I would run into nanny mode constantly and then it doesn’t matter how good the model is.
Paywall