Joe ❌👑

LLM-based agents don't have to be super-intelligent to cause massive damage. After all, the Black Death killed 1/3 of Europe and no one thinks it was conscious. It suffices to have large numbers of agents pounding away at the Internet, powered by millions of dollars in compute power, trying to reach odd corners to satisfy a goal, using every tactic they copied from Stack Overflow or hacking contests, with no knowledge of the side effects their actions may cause. Internet-connected systems have massive numbers of security holes and really stupid attacks can produce bot farms even before LLM AI came along. Enough life-critical systems are connected to the Internet that a massive, stupid agent-based attack where neither a person nor an agent "decided" to cause harm could produce a disaster. So I don't think AI is going to become super-intelligent and kill us all, but do think that stupid AI could make the Internet unusable at some point.

@benroyce@mastodon.social @sundogplanets@mastodon.social The Germans do this better: they just make compound words using their own words instead of borrowing Greek or Latin. It is less pompous. But a thousand years after the Norman Conquest we are still embarrassed about basic Anglo-Saxon words, so respectable people use French, Latin, or Greek words.

@wjmaggos@liberal.city It appears that YouTube is carrying the trailer, they just are refusing to accept paid ads for it. Which means that few people will find it, but we can share it. It's trending now.

@MaryAustinBooks@mstdn.social In theory, representatives are working when the House is not in session, meeting with constituents and the like. But GOP congresscritters typically refuse to hold town halls or to meet with ordinary constituents (who aren't big campaign contributors), so they can't even make that excuse.

@Thayer@mastodon.social @david_chisnall@infosec.exchange The reason the AI tools have this bias is that they trained them to replicate existing hiring decisions. Well before AI, companies were evaluating resumes with a female or a Black sounding name lower.

@vicfroh@mastodon.social There are huge problems with the Catholics, particularly sexual abuse by priests and thinking women's only purpose is about pumping out babies. Those things are not forgivable. But they do take helping the poor a lot more seriously than, say, evangelical Protestants. The Catholic church closest to us runs a food bank and another one close by is a major supporter of homeless folks, collecting clothing and supplies. The cannibalism is crazy, but at least in the US the Catholics stress the Sermon on the Mount and many, not all, protestant churches don't preach it to the point where their followers find the language foreign, "woke", and that faith is all you need, screw the poor because you have a personal relationship with Jesus (atheist here, recovering Catholic).

@bagder@mastodon.social in a world where people download scripts blindly from some random URL with curl and pipe them into bash, I think it is safe enough for the maintainer to take a holiday instead of working really hard to make a strong door lock even stronger when everyone has other ways of getting in.

@monkeyninja @mekkaokereke I just gave to SPLC openly and don't care if they know. Fuck em.

@mathowie@xoxo.zone @mekkaokereke@hachyderm.io I have submitted a grant request to SPLC from my donor advised fund. Morgan Stanley runs it. It normally takes them a week to process it and I haven't heard anything yet. They have never rejected anything before. So we will see if this is just Fidelity. Edit: Morgan Stanley paid SPLC today. So this is just Fidelity.

@QasimRashid@mastodon.social They object to nonwhite people being proud of who they are. But yes, if a certain recent president had run as a Republican, he would have been Barry Omaha or something.

@neatnik@social.lol DHH is a threat, but Omarchy is not as hardly anyone uses it or has a reason to. It should be shunned, yes, as should he, but he is a long way from taking over, and spreading the word about him will prevent him from gaining more power.

@glyph@mastodon.social It is not, because actually writing software is 1/10 of the work, and if no one on the team understands it no one can maintain it. Agents can help whip up a fast prototype, and that can be helpful. But it will be flawed and unmaintainable.

@futzle@old.mermaid.town I am a white Californian who has worked with Indian colleagues for decades and have been to India five times. I am a big fan of Indian English. My favorite word is "prepone". It is the opposite of "postpone". On one of my trips to our Bangalore office, it was my first day, the jet lag was not great and I was obviously fading. The manager I was visiting kindly asked if he could prepone my cab, that is, ask the driver to come for me earlier than originally scheduled. We all need that word.

@zackwhittaker@mastodon.social What is stopping Flock from just selling their LA camera data to ICE instead? Can LA force them to turn off the cameras?

@mcc@mastodon.social When I was trying to teach myself Rust I was really annoyed at rustfmt for dinging me for extra parentheses. Why should it care? Is there a way to make it shut up about extra parentheses?

@simon@fedi.simonwillison.net Sounds like it will relentlessly spend all of your tokens and take an overkill approach to solving your problems.

@mattblaze@federate.social My DAF is managed by Morgan Stanley. They approved my recent grant request to SPLC (which I made after the indictment). So this is specific to Fidelity, which seems to be imposing a guilty until proven innocent, don't piss off Trump process that they are under no obligation to use.

@futurebird The Pine Barrens, perhaps?

@ai6yr@m.ai6yr.org This explanation, though, is exactly what I assumed was the reason for the LLM's output when the story first came out. Not that any thought or intelligence was involved, but you have a machine that is designed to generate the most likely continuation of the text, with just a bit of randomness tossed in for variety. If you set up all of the elements of an SF story about a rogue AI that resorts to blackmail to keep from getting turned off, and if there are hundreds of stories like that in the training data (which of course there are, they stole every story ever published), that's what you'll get. So, yes, LLMs will act out "evil AI" scenarios if prompted correctly because they are effectively trained to do so, even though the human feedback training they do after the model is built attempts to reduce that tendency.

@cks@mastodon.social @mattblaze@federate.social That might be it, to eliminate what otherwise might be a common error.

@SnoopJ@hachyderm.io @futzle@old.mermaid.town Both variants are commonly used in India.

@briankrebs@infosec.exchange Wow, if there's a "shred log", we are really sure that they deleted all the data and don't retain any copies to blackmail us again with.

@Nonilex@masto.ai We need to win the Senate by 2 (since Fetterman is a lost cause). Tough, but possible.

@mxchara@seattle.pink @DeliaChristina@sfba.social That is because the trick of American conservatives is to convince poorer whites that the liberals and socialists don't intend to help them, but plan to steal their money and give it to "undeserving minories". So in the press, "working class" means working class whites who have fallen for this.

@MaryAustinBooks@mstdn.social I think Flock retains ownership which is why there have been problems getting rid of them (even though cities paid). But I am not sure. The big new business trend is that you can't buy things, you only rent them. Or you have bought an object that only the manufacturer can fix, and they can just turn it into a brick anytime (Internet of things devices).

@MaryAustinBooks@mstdn.social If a city tells Flock to turn a camera off and remove it, if they don't do so within a reasonable time they should just declare the camera to be a trespasser and remove all penalties from anyone who disables and removes it.

@monkeyninja @mekkaokereke As a 501-c3 they might not be able to accept anonymized contributions, though.

@whack@hachyderm.io OpenOffice? That is pretty much dead, almost no development in years. It was forked many years ago, and LibreOffice is the actively maintained branch, with significant improvements, though I think you would still find it to be inferior to Keynote, at least it would be less terrible.

@mattblaze@federate.social I subscribed to his RISKS forum back in the early 80s (yes, I am old).

@DeliaChristina@sfba.social This is Politico, which shills for the GOP, trying to discredit Democratic efforts to counter GOP gerrymandering. There are two answers people taking part in the poll (which is a push poll) can give: don't oppose GOP gerrymandering, or counter it even if minority representation is lost. Now, this doesn't mean that Democrats don't suck in a lot of ways. But I think you are responding in the way Politico's owners want you to respond.

@futurebird Maybe the imperfect copy of the journal was made by an obsessed stalker, now determined to gaslight you for reasons you need to figure out.

@catsalad And her admin is in a huge fight with your admin...

@Alice Fortunately you're posting this in the fediverse where we don't care for that shit, and not on some other platform, or someone would take your message as a challenge and make that video of you.

@ariadne@social.treehouse.systems Yes, and initially writing the code is only a small portion of the work. To support and build on that code, you need team members who understand it in detail, who know the "why" of the design, who have clue about what compromises were made and what needs to be fixed. Outsourcing that to an LLM means that you get a result that is difficult to maintain, that no one on your team really understands the details of.

@liaizon@social.wake.st If someone needs to be in charge, pick someone whose career has been with libraries. Not a corporate shill or a spook.

@Amoshias@esq.social @DeliaChristina@sfba.social The question made perfect sense, if you remember that Politico is a GOP owned organization that wants to discredit Democrats.

@scottsantens@hachyderm.io Yes, means testing means you have to have more cops to do the means testing. Make Elon pay millions more in taxes and give him a little bit back in UBI.

@avilewis@mstdn.ca Might want to read down in the article, where he talks about doubling trade with non US partners, while only saying he "remains open" to deeper integration, while at the same time making deals to reduce US dependence. So, maybe. But headlines are ragebait.

@jwz@mastodon.social Hasn't it been the first Friday of every month lately?

@freyjfreyj @mattblaze Morgan Stanley. They processed my DAF grant to SPLC after the indictment without any fuss.

@mattblaze It is possible for you to transfer your entire donor advised fund to a new manager.

@georgepenney We love you, George. Thanks for being here.

@DeliaChristina@sfba.social Oh, I definitely agree with you on that (Black folks are the most reliable D votes). But Politico is trying to fuck with us, again.

@deathkitten @jwz I suppose it is possible that it's deliberate fuckery but probably they just weren't paying attention.

@infobeautiful@vis.social This is not information, it's misinformation. "Same size" does not mean the same as "Same %".

@indivisibleteam@mastodon.social Hi, Indivisible. This is a non corporate space. You can say KEEP YOUR FUCKING HANDS OFF OUR VOTE here and no algorithm will penalize you. And yes, I will be in the streets on October 17.

@mattsheffield@mastodon.social I think you are missing something here. They aren't using the word "alignment" the way the AI doomers do, to refer to making sure that the model does what the owners of the model want it to do. The letter is saying that the goals of the *companies* running the models are not in alignment with what mathematics needs. For example, the recent episode where the company rushed to claim victory before the proof could be reviewed and digested, which lets them score a point but results in rushed and worse communication and hurts the collaborative process.

@lauren@mastodon.laurenweinstein.org Yes, if I recall correctly, other than questions the only statements Eliza would generate were of the form "Earlier you said X" where X was always something the human participant actually typed.

@pmonks@sfba.social @cammerman@mstdn.social @mhoye@cosocial.ca I'm curious to see some of the limericks that are generated by these instructions.

@Teri_Kanefield@mastodon.social So, Congress should step in, even if it means it can be said to own the resulting mess. Regardless of political consequences, we need to stop the bleeding. If Trump's supporters want to argue otherwise, let them explain just how Trump could have possibly finished and won. With what? Nuclear weapons?