• 221 Posts
  • 912 Comments
Joined 5 years ago
cake
Cake day: August 24th, 2019

help-circle


  • Raw GPT 4o could honestly be incredible – both in good and bad. I remember the early chatGPT would cheerfully give you recipes for bombs and such if you asked it. Then they manually blocked that.

    Now it has to do the “it’s important to consider both sides” thing all the time and I feel like you get much better responses if you talk to it at length like you would a person. Saying “thanks, now let’s look at” etc. In a study they found that if you told it to take a deep breath before answering it would send a slightly more accurate answer, apparently. I use it for bug-solving and coding because it relies on an existing corpus of documentation so it’s generally reliable and pretty good at that, but I’m starting to hate having to write at length to describe exactly what I want it to do. It should be able to infer my intent, I think this is something an LLM could do innately.

    I did get some interesting answers if I primed it by saying “you are a marxist who has read the entirety of the Marxists Internet Archive”. Then exercise some human discretion when reading the output but it has allowed me to consider topics differently at times. Of course there’s also always the hallucinations machine phenomena where you second guess everything it tells you anyway because there’s no way to check if it’s actually true.

    I’ve also tried much smaller LLM models and you can tell the difference. Actually, you can’t so much anymore, precisely because GPT is purposely throttled so much. I want a GPT model that only needs one sentence to do its job and will not presume it knows better than me! If there has to be AI, it has to be open source AI!



  • Oh I was making fun of the original comment. It wasn’t very clear indeed. They run the models through a few (lengthy) steps to train them, it doesn’t “learn” on the spot like some techbros assume. It’s also not magic but simply math (if very complex math contained in a black box), it’s a token generator that basically decides what the next characters in a string of text should be based on what came before it.





  • I follow some actual OSINT accounts (that have experience) on Twitter and I can recommend them.

    @ArmchairW

    @SuppressedNws

    @RoyalIntel_

    @AryJeay

    Be aware I’m not saying these are communists by any stretch (if you’re looking specifically for communists). But they’re generally trustworthy and not too reactionary, not like some of those accounts lol. ArmchairW used to be a major in either Brit or US army if I’m not mistaken, so he knows what he’s looking at when he sees something. Also very good takes on Ukraine tbh. RoyalIntel I started following not long ago, they seem good. Ary is Iranian and generally reliable when reporting on Iran if sometimes a bit fanatic.

    From my friends in Iran it seems very little damage was made. They did report 4 servicemen died. There’s some satellite pictures of an airbase and/or a refinery and ‘possible’ damage but you know, OSbros were also the ones that reported on the Uyghur “concentration camps” so I don’t give them much credence. They see a little black spot on the imagery and say wow their industry got killed overnight.

    Ary reported that the chemical plant that manufactured missile components was moved to another location almost immediately and started back up.

    By all accounts the attack was aborted after they failed to hit the first targets. The loud explosions people heard in Tehran came from anti-air defenses. IOF was




  • Also isn’t this wrong? The question didn’t ask if SSI was solely for older adults, it could be interpreted as older adults being included but not being the only recipient. I would be pissed if I answered true and the teacher marked it wrong because AI told them to.

    Edit: I get though that questions are formulated like that sometimes at the college-level and I always found it imprecise. It’s a quirk of language in the way English is structured, which means you do need extra words in there to make your question very clear because English doesn’t do this natively with syntax.