mikecaulfield.bsky.social
Author: Verified: How to Think Straight, Get Duped Less, and Make Better Decisions about What to Believe Online (University of Chicago Press).
Researcher, infolit/misinfo/rhetoric/civic reasoning. Currently researching AI as tool for critical thinking.
7,515 posts
32,577 followers
2,246 following
Regular Contributor
Active Commenter
comment in response to
post
When I say "mostly working" I mean it can probably do about five times better than most people probably can, but not the ten times I was getting before. (To put rough numbers on it).
comment in response to
post
No, it isn't. You do realize "hallucination" doesn't just mean "not perfect"? You clearly misread something just now for instance, that doesn't mean you are "hallucinating" etc.
comment in response to
post
Not actually your mate?
comment in response to
post
The right place looks a bit like this
comment in response to
post
Full instructions here mikecaulfield.substack.com/p/deep-backg...
comment in response to
post
Do make sure you're using the o3 model, the other models may hallucinate links at times, but o3 has been solid.
comment in response to
post
O3 handles them, and that used to not help bc GPT req'd 4o. But last week openAI opened up GPTs to the other models. I just had to cut my prompt down by 2/3rds but it seems to be mostly working.
comment in response to
post
Also more minor point but let's not have 79 year old sheriffs please
comment in response to
post
I thought that too initially but then I looked at what was right above it
comment in response to
post
I'm fine with it because you really do have to be blunt -- it doesn't do that automatically -- but I'll check your chat log later and see if there's room for improvement.
comment in response to
post
Sometimes you need that. For instance it will tell you a video it finds is unverified and you go verify it and say, no you can consider that checked. Or it says something is not in an article, and you find it in slightly different words and say yes it is.
comment in response to
post
I should say sections of it have been. I don't know if they disclosed that somewhere. I honestly think disclosure is less important than just rigorously checking outputs or admitting its a draft at best. The rule of AI is still simple: in the end, you're the author, you're in charge, it's on you.
comment in response to
post
Lol.
comment in response to
post
I'll take a look. It's plain html pages on neo cities so probably need to do something more than yolo it in a text editor I guess
comment in response to
post
Yes
comment in response to
post
Also I'm aware of the whole filming thing and try to be sensitive to it, but the reason these people come to mass mobilizations is for people to see the massiveness of the mobilization, so at some point you have to show that in as respectful a way as possible.
comment in response to
post
About 45 seconds from the end of you're looking for it
comment in response to
post
Favorite moment was probably the five year old behind me yelling "No Kings!" a bit into that clip.
comment in response to
post
I am not exaggerating when I say there are more flags here right now than at the fireworks on 4th of July and nearly as many people
comment in response to
post
It's not even AI speak more like exquisite corpse
comment in response to
post
My favorite sort of comment
comment in response to
post
Criticism is healthy, but the issue, I think, is that the reform movement pushed for bureaucratic solutions made of fallacious quality norms. If fascists just need to say "ok" to the proposition in order to advance their goals, I suppose it says something about the proposition itself.
comment in response to
post
A long time ago (building off of something in Slate Star Codex of all places in maybe 2017(?)) I argued that most of the people sucking down misinformation weren't really targets of it, they were just casualties of an effort to get these ideas into institutions.
comment in response to
post
Double-blind ADHD medicine trial also wild for a drug everyone agrees at the very least changes behavior pretty dramatically for 18 months. You going to swap someone's two year medication with sugar pill? How will this work?
comment in response to
post
My pledge is if you read my posts on AI you will get a lens on it that is very different than what you're seeing elsewhere. And in a world overrun by LinkedIn clones pitching 300 variations on the same four ideas maybe that will make this seem like a space worth thinking in rather than reacting to.