jake
he/him
/in/jakemannix, fka @pbrane
professionally: Tech Fellow, AI & Relevance, Walmart Global Tech
here: bad math jokes, AI in general, puns, OSS ML news, outdoorsy stuff, DL papers, shitposting, the fall of democracy
- Large language models make me think of cargo cults. The belief is that by creating strings of words that superficially look like those created by intelligence, intelligence will later magically appear Or, in Chomsky terms, it's the belief that by aping surface structure, deep structure will appear
- Have you tried Prof Mollick's experiment? bsky.app/profile/emol...
- If I got a text from Claude: -UNIVERSAL CONSTANTS NOTIFICATION- FUNDAMENTAL LAW OF REALITY Re: Non-Existent Business Entity Status: METAPHYSICALLY IMPOSSIBLE Cosmic Authority: Laws of Physics THE UNIVERSE DECLARES This business is now: 1. PHYSICALLY Non-existent 2. QUANTUM STATE: Collapsed [...]
- [Not loaded yet]
- Can you imagine putting $0.75 in a candy machine and the little LED display is all like: "THIS IS METAPHYSICALLY IMPOSSIBLE! YOU ARE REPORTED TO THE FBI, MY QUANTUM WAVEFUNCTION IS COLLAPSED!"
- Something I found interesting about reading this thread is that it made me wonder how many people think of ChatGPT as primarily a tool for writing things for you I use ChatGPT/Claude/Gemini possibly 100+ times a day and NONE of those uses are to produce writing that I then share with other people
-
View full threadSo models trained on all the things we've written, are only seeing the output of when we thought we had answers. Every time we *don't* have an answer, we go back to the drawing board until we do, then write *that*. They're not being trained on the intermediate failures, until very recently.
- (and even those are mostly simulated)
- [Not loaded yet]
- It's not that "the makers can't have that", it's that it's very statistically uncommon for people to write down, in the same place, <some question> followed by <I don't know>. It happens, but *usually* if we don't know something, we don't write that. We write answers to things we think we know.
- And how about the effect on students’ learning performance of *unregulated* use of ChatGPT, i.e., for homework help?
- The effect of ChatGPT on students’ learning performance, learning perception, and higher-order thinking: insights from a meta-analysis - Humanities and Social Sciences Communications nature.com/articles/s41599-025… #AI #education
-
View full threadFrankly, HW should never have been "graded" in any way that it counts towards their grades. Should merely be to give them feedback on how they're learning, and in no way affect their "grade". UCSC had no grades when I went there, was great. My kids went to Montessori: self-paced, also great.
- but yes, my own students reported that ChatGPT / Cursor / etc all give them a lot of "the Illusion of Learning" if they aren't disciplined about its use. They "get a lot done", but don't retain well, if they didn't *struggle*, which now they need to be more intentional about.
- Students have always been able to cheat, and some have always done so. My kids stress about not *learning* and so don't use ChatGPT in school (my 17yo doesn't use it *at all*, but my 27yo uses it at his software job all the time, but still tries to focus it on rote tasks that *should* be automated)
- A common question is "can an AI make money?" This benchmark, where AIs run a simulated vending machine over time, suggests yes, with an important caveat On average, Claude 3.5 & o3-mini beat a human, but are high in variance & fail at random times for complex reasons. andonlabs.com/evals/vendin...
- "can AI make money?" "yes, but it may also question the nature of reality and the state of its quantum wavefunction, after calling the FBI!" "what job did you give it?" "running a vending machine" "and if you gave that job to a 16yo, what would happen?" "... fair"
- ... I don't know if I could stand up from laughing for about a week. Man, I have *nothing* to worry about from that Basilisk. It's gonna have to spend all of its time on the authors of this paper.
