mstdn.social is one of the many independent Mastodon servers you can use to participate in the fediverse.
A general-purpose Mastodon server with a 500 character limit. All languages are welcome.

Administered by:

Server stats:

18K
active users

A special thank you to the useful idiots in the tech world who spent the last decade vilifying those of us who spoke out against Big Tech while providing them with social capital and legitimacy, oftentimes benefitting financially from the transaction.

We couldn’t have made it here without you.

@aral And there are so many useful idiots. Lots of them even nominally antifascist. But even now they keep simping for companies doing DRM and shit and spitting out condescending takes on critics.

Mud Road

@dalias @aral has been captured since the early days of "free" Google and Apple's dominance of schools. The language used by teachers and administrators is telling, tech coaches are see as 'evangelists' for products and services. It's mostly about how to use a product or tool and rarely about what is being produced. And every few years a new saviour arrives. Ten years ago every child who didn't learn to code was going to be left behind. Now it's AI.

@fredb @aral I'm working on a fun lesson plan for debunking the magic of "AI" intended to be accessible to elementary kids & teachers. Not explicitly described as "AI is bs", but with the intended outcome that everyone realizes there's no thinking going on.

@fredb @aral Basic idea is to take a few pages from a book they're reading, break up into groups, and tabulate the words that appear, then what words follow them, and do a Markov bot on paper with dice, no computers needed so nothing is hidden. The details I need to work out are balancing the order of the chain with what's practical for size to work with and how decent the output looks for candidate inputs.

@dalias @aral Great stuff. Like all tech, there are some life changing applications, but a lot of the time it boils down to 'what makes things easier for me to get through the day?' Which of course just kicks the problem down the calendar... I suppose you can use Chat GPT to do your goal-setting and self-reflection docs though :)

@fredb @aral I think you kinda missed my point. The goal is to see that the whole process is about making random sentences that look vaguely (LLMs more convincingly) like something you'd expect a human could have written, but that have no meaning behind them. Only reason I don't intend to promote it with "AI is a scam" snark is that I want it to be something teachers eager to have students learn about "AI" might be attracted to and where the moral is something you discover yourself rather than being preached to.

@dalias @aral Your approach is awesome, and the students will draw their own conclusions as always. Apologies if I didn't make that clear... You're doing a hell of a job if you're constructing learning in this way for your students.

@dalias @aral I'd be curious to see if the teachers actually care that LLM outputs are statistical in nature, to my mind they're more concerned if it sounds about right. A lot of what we have to/tend to write day to day is pretty vacuous and template based. Success in standardised assessment systems is driven by the need for answers that fall within a certain set of parameters. I wish teachers were more conscious of the politics and the ethics of the use of these tools.

@dalias @fredb Sounds great! I was shocked when I heard Gen Z (adults!) talk about AI on youtube saying "its supposed to be super smart"

@dalias @fredb @aral Perhaps before you do, go and give deepseek.com a try (it’s free to use) and look at how it thinks ..

@martin @dalias @aral 'Thinks'. I'll take the bait and have a look, though I'm not sure it's worth firing up another coal plant for.

@fredb @martin @aral AI shills are like the free energy/perpetual motion folks. Somehow it's always "if you'd check out this one you'll see it's real, I promise!" completely failing to get that we already understand on a broad level the claim is impossible.

@dalias @fredb @martin “We” being a tiny handful of people, it feels like.

@aral @dalias @fredb @martin Yeah, really. Like, I had an argument with a friend *who is a statistician* & still refused to believe that there is no reasoning going on or understanding emerging in an LLM - simply because it looks so convincing.

Pretty sure most of it is the "Mentalist Effect"; I mean, despite this having been debunked since fucking Houdini, look at how many people still believe that shit is real...

softwarecrisis.dev/letters/llm

Out of the Software CrisisThe LLMentalist Effect: how chat-based Large Language Models rep…The new era of tech seems to be built on superstitious behaviour

@jwcph @aral @dalias @fredb @martin I don't understand why people can't see them for what they are - Slow, Immutable and Expensive fuzzy databases.

@jwcph @aral @dalias @fredb @martin I know of serious academics who honestly think the field of mathematics will be obsolete soon.

@jwcph @aral @dalias @fredb @martin My wife has been a teacher for 20 years now and if anything, it's clear edtech is 99.9% snake oil.

@gwozniak @jwcph @aral @fredb @martin I've seen some really impressive educational software, but "edtech" is a term I reserve for exploitative VC shit.

@gwozniak @jwcph @aral @fredb @martin Big externally distinguishing characteristic is whether it has access to kids' PII or the profiles are basically pseudonymous. Also the worst of "edtech" seems to be administrative (like the scummy Remind that started as parent messaging and pivoted to trying to be "Uber for unvetted private tutors" 🤮) not pedagogical.

@gwozniak @jwcph @aral @fredb @martin Aside: proud of getting Remind fired from our school district by reporting their early spam to principal and raising concerns about potential abusive, bigoted, or predatory unvetted tutors being advertised to parents.

@gwozniak @jwcph @aral @fredb @martin Mathematics is one place where there might be some useful results, specifically because proofs are testable.

However most interesting advances in mathematics are choosing the right definitions to build stuff from, not proving something that was hard to prove. And I'm skeptical shuffling existing proof techniques according to statistical models will yield a lot of new results.

@dalias @fredb @aral Best debunker so far: implement an LLM in a spreadsheet!

@michaelgraaf @dalias @aral I don't even know where to start with that :)

@fredb @dalias Oh, and don’t forget Microsoft. I wonder if anyone surveyed how many kids were put off of tech by being introduced to computing with Microsoft Office in UK schools. That was part of the reason why we set up Code Club and why I was so heartbroken when one of the founders – and a very dear friend at the time – basically sold it out to Big Tech. Oh well, what doesn’t kill you and all that…