I keep forgetting just how easy it is for people with no responsibility or understanding of what responsibility looks like to act like every act of caution is stupidity/weakness. The MAGA "just bomb Iran! woo!" types, the "protests are for violence" saboteurs, the "just give everyone zerodays" moron
Posts by the fool
well, you sure showed us by, er, deluding yourselves about reality? sure I guess that makes as much sense as any other anti-ai position
Things have costs. They take time. There are no magic wands in real life. If you’re releasing something no one has ever built before you SHOULD be cautious rather than reckless
Like for the first time a tech company takes that seriously and people like you think it must be a scam
For the last time, “Mythos” doesn‘t mean “Cthulhu”
”if it’s powerful enough to find vulnerabilities at concerning scale, it must be able to trivially fix all vulnerabilities everywhere“ was your belief. It’s nonsensical
A local offline agent plus a dump of Wikipedia is astonishingly useful when the internet isn’t around
Is this how you get back at him? Saying stupid things online? Bring shame to the Cybersecurity family name?
They’re not saying that. That isn’t true. Your brain is binary and doesn’t understand aspects of the world that don’t meet your black and white thinking
That’s really more a you problem than anyone else’s I guess
what a strange prediction to make
I just added Letta to a system using the obsidian/wiki pattern (recently popularized by Karpathy) and it's sort of interesting to see the seams. I'm actually wondering if different sometimes-out-of-sync memory systems are interesting for certain kinds of properties of systems like this
the problem is your mental model only has room for two scales of model, "meh" or "godlike", with not a lot of room for the messy realities between
like do you have any friends or family in cybersecurity you could grab coffee with
if my partner works in a biolab with smallpox, does it follow I should think everyone should get a smallpox lab at home
muskets could kill people, I still think some regulations of automatic weapons makes sense and people shouldn't have nukes
"it's too powerful to be released" isn't a FACT question. it's a judgement question. you can't "lie" if you say "I think this is too dangerous to release" and you end up being wrong
I mean if you want to talk epistemics - how do you know that this is bad. Teachers are wrong all the time but we don't think of them as useless (or, I don't, and I think it would be hard to argue infallibility is necessary for imparting knowledge)
If only stupidity couldn't do so much damage...
Just a reminder that Ed Zitron gets paid by AI companies to do PR
The irony on this timeline is too intense. I'm not rated for this level of insanity
in the collective fever dreams of bluesky, my friend
on second thought regret the use of "tribalistic". war over ideology is the essence of "civilized" peoples...
the funny thing is that yes, they're twitter personified in a way and also they're the same people who left it when it became X, even though in many ways it's more what they want (unbridled tribalistic conflict in the arena of Too Online)
"children are people" is one of my most controversial takes
someone called it my "tag" and I genuinely didn't know what they meant
handle, username, screen name... sure
There's a very, uh, vocal contingent on bsky that believe if anyone says something about AI that isn't "it's a world-devouring plagiarism hallucination stochastic parrot that doesn't work at all" then they're being paid to say that
This immediately made me wonder what I still need to learn about that's late
every rando I see is "lol this is PR" while every security professional I know is withdrawing their money from banks and burying gold in their back yard while prepping bug out bags
ah I'm mainly trying to see what I can run locally (yes, I know, it's inefficient and slow but damnit I'm a tinkerer and I've never had one of those gas guzzlers so I'm going to treat myself)
I'm not anti-AI but I'm absolutely on team "let's not hand them more powerful things to use against us!"
I'm even, though I disagree on basically everything about it, tentatively supportive of the "stop datacenter production" bill if it helps people get more serious. (I don't think it will though)
Maybe I should make a log of every value of $X and why it turned out to be a bad idea
I haven't dug too deep either but also heard that some of the systems that should make Gemma4 faster aren't even enabled in the public models (MTP/SSD? something like that? grain of salt and all)