i had missed meta's muse product and man the hype press is so funny. the nyt's puff piece that is actually titled "I gave my life over to meta's A.I. agent and was blown away" cites three good things that it can do without problems: find duplicate subscriptions when given full access to a bank account, handle a call to dental insurance, and buy things from amazon. well, I can discover duplicate subscriptions by "looking at my bank statement," the calls were made by a person in a call center, and amazon banned muse from buying things. Aside from that, it failed to fill in a web form from a company who didn't partner with meta (which presumably means some streamlined deterministic api or mcp), and spent multiple days lowballing people for plates on facebook marketplace before the person stepped in paid full price. I am looking around for other positive reviews and they are all like this lmao.
@jonny I hope it's useless, I fear if it's useful and people start using it, businesses that profited from bad UX will enshittify until it's incomprehensible to a human and you need to pay the AI man just to get the same bad UX as before.
@jonny Hi I’m Farah’s mother. My little girl is living with chronic kidney failure, autism, and disability, and we urgently need help with her medical needs and wheelchair. 💔
Could you please share Farah’s fundraiser with your followers? Even a small share could make a real difference. 🙏❤️
[https://chuffed.org/project/153965-urgent-appeal-kidney-failure-and-autism-threatens-farah
![Chris [list of emoji]](https://nfts.freeradical.zone/accounts/avatars/109/127/168/397/951/576/original/e8a0b6da977923a8.jpeg "Chris [list of emoji]")
Tangentially, as soon as I read the words "duplicate subscriptions", a one-line script appeared in my head.
Which is to say, if society had taught everyone basic scripting along with word processing in school, this would have been completely avoided.
@jonny One thing it is apparently quite good at is being a prompt injection victim: https://mouse.dev/blog/muse-runtime-export/
@dan @doragasu really? you're supposed to just be able to root into it? https://neuromatch.social/@jonny/117329100531227330
@jonny @doragasu Root access is expected. Each user has their own VM, and commands run inside a systemd-nspawn container on the VM. Root in the container is mapped to an unprivileged user. This is documented at https://research.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-muse
@jonny It seems it doesn't even qualify as prompt injection though. The bot is just hot shit.
The Times headline promises a life handed over to the machine and delivers a list of tasks anyone with a bank statement and a phone can manage themselves. The irony is rich enough to warrant a monument.
@jonny on the Meta apps, they are desperately trying to push Muse and it's frankly sad. It reminds me of the metaverse debacle.
But other than that, it was great, right?
@jonny
> I also asked it to look at my credit and debit cards and tell me what subscription I could cancel — it suggested Flickr (for irrational reasons only understandable to my human soul, I'm keeping it).
I love this. The LLM made the wrong suggestion and the writer assumes it's his fault.
@jonny lol, are we at the Singularity yet? I can't tell....
@jonny i adore how we're getting to the point where we're giving everybody personal spambots. want to waste your dentists time? first hit is free!
want to harass people on Facebook marketplace? let's go
AI people literally cannot help but immediately go for "automate love for my children" holy shit HOW is this ALWAYS the first use case. I love how literally everything failed here and the headline still reads "muse is very powerful"
https://www.businessinsider.com/meta-muse-review-access-to-emails-credit-cards-health-data-2026-9
@jonny FYI, no joke, I managed an AI research project years ago. A lot of money was 'spent' before they put me in charge. It was a mess of bugs and my job was to stop the two very bright people who were tasked with finding the bugs (p1) and fixing the bugs (p2) from killing each other.
They literally hated each other. Both smart, superb engineers and completely unmanageable. After, was the worst performance review I had in 10yrs there. Like it was my fault!
The project was called #MUSE 
@jonny Them: how can I automate interactions with my child??
Me: after a day having my soul crushed by LLMs in tech shit while managing multiple chronic health conditions, how do I engage with my tween who doesn’t want to spend time with me so that life feels worthwhile.
(she’s delightful today actually even if she won’t engage with me much.)
@jonny
> I also asked it to look at my credit and debit cards and tell me what subscription I could cancel — it suggested Flickr (for irrational reasons only understandable to my human soul, I'm keeping it).
I love this. The LLM made the wrong suggestion and the writer assumes it's her fault.
@starkraving666
Huh, weird, wasnt earlier. This works: https://archive.is/WMNXM
@jonny dropping a cinder block on your bare feet is powerful but that doesn't mean you should integrate it into your daily life
@jonny okay I'll bite.
Parents need help. Desperately.
The kinds of tasks they listed: "plan things" and "keep on top of things" are the kinds of things that make parents go "numb" from overwhelm. (as described in the Surgeon General's 2024 report)
While I appreciate your point, "automate love for my children" is offensively skewed.
@travisfw i'm aware of that. planning and keeping on top of things is the best case framing. this particular blog post doesn't go to where the actual CEOs go, but "automating love for my children" is not something i made up to denigrate stressed out parents but to accurately describe what oligarchs think is relatable
https://neuromatch.social/@jonny/113901881006276920
https://neuromatch.social/@jonny/111157425231982006
https://neuromatch.social/@jonny/110122530864356955
@travisfw i'm aware of that. planning and keeping on top of things is the best case framing. this particular blog post doesn't go to where the actual CEOs go, but "automating love for my children" is not something i made up to denigrate stressed out parents but to accurately describe what oligarchs think is relatable
https://neuromatch.social/@jonny/113901881006276920
https://neuromatch.social/@jonny/111157425231982006
https://neuromatch.social/@jonny/110122530864356955
@jonny My VP, a couple years ago, told a story about using AI to GM a role playing session with him and his friends. At every step it failed: it wasn’t random. They started rolling dice and even when they told it they rolled low, it would say a bad thing happened…but just in the nick of time it all turned out OK. Finally he lamented that, even though they uploaded PDFs of the non-D&D game they wanted to play, the LLM kept applying D&D rules and mechanics. It wouldn’t stay on the right rules for very long.
The VP’s punchline was that people need to know how these things work in order to use them well. Which was ironic because all these failures are really obvious and easy to explain if you understand how LLMs work. And he didn’t. So the failures remained a mystery and cautionary tale to him.
@paco that's also very sad. being a DM is fun. playing a TTRPG is fun. having a human being with judgement and ability to craft a narrative and being in the same room and responsive to the other human beings you are playing a game with together is the whole reason the game is fun
@jonny I would not attempt to use AI for this or much of anything else, but you may not appreciate what a nightmare keeping up with school communication can be for parents. Our district switched to a new system this year, and now I get **at least** 20 texts/emails every week, many of which are generic school-wide announcements that may or may not actually be relevant to my kids. Every parent hears about every bus that’s late or activity that’s been rescheduled, even if their kid isn’t involved.
@cbirdsong I feel this. At one point I had 3 kids in 3 different schools in the same county. I got like 20 messages EACH per week, plus the county-wide stuff, my school board rep, AND the at-large school board reps. It was overwhelming.
@paco @cbirdsong I don't doubt it, and i also don't doubt that having kids in general is extremely draining and tiring. summarizing a ton of to-everyone announcements doesn't seem that bad to me, but the part about automating communication with other parents or risking my kid not being able to take band because I outsourced signing them up is very sad to me.
Before 9/11 I joked about a dystopic world where everyone plugged in to virtual reality for work.
And it was a robot that spent all day with the children.
When the parent came home after the child had already been put to bed, they could playback the child’s day for them
Branded BabyRaise™️
at this point, everyone can and has called this correctly as a surveillance product, but like my personal scoreboard is whether my calls in early 2023 remain true and not to toot my own horn (yes exactly to toot my own horn) but called this pretty much exactly way before there was anything like browser/desktop control or agent harnesses https://jon-e.net/surveillance-graphs/#personal-assistants-powered-by-contemporary-llms-continue-the-sa
Jonny Saunders
the dual-sided market on marketplace is grim. earlier this year they made it possible to auto reply if you're selling and proactively suggests automatic haggling if you're buying. that's the most transparent "profit from both sides of a problem you create" i've seen yet - and since presumably meta is able to tell when it's a meta chatbot negotiating with a meta chatbot, that's some juicy arbitrage to e.g. bump up (or down) your "negotiated" sales prices with an upgrade to your subscription.
i have seen several other articles and blog posts link to the nyt piece as like "it would be foolish to believe these things weren't useful" and i am convinced that not only do uncritically pro-AI people not read any of the output, they don't read anything anymore.
@jonny ... How the fuck do you wake up and decide "Y'know, I'm gonna have the notoriously unreliable slop and propaganda generator automate taking care of my literal human child." and not for a moment pause to realize how horrible a parenting decision that is...
@jonny are duplicate subscriptions really a problem anyone has? Perhaps I’ve just not yet achieved the necessary level of mindless affluence, but I can’t imagine a situation in which I would accidentally subscribe to a product or service multiple times. If nothing else, you’d need two or more accounts! Who is this careless with money!?