abecedarius.bsky.social
durable key issue: bacon
211 posts
81 followers
67 following
Active Commenter
comment in response to
post
When it got assigned in high school I read the first pages and ... stopped. It felt like it didn't deserve association with school. Read it later.
comment in response to
post
It's proprietary, but they did write up some stuff
docs.google.com/document/u/0... chapter 26 most relevant to the bit I brought up (not sure if there's any longer description)
comment in response to
post
(The system works with a knowledge graph across courses, tracking what mastery it thinks you have and trying to choose exercises that reinforce more than one item efficiently like a generalized SRS system.)
comment in response to
post
(Because uncalibrated confidence gets penalized.)
Maybe bigger: school math curricula are basically designed for people who don't know anything. I think you'd get lots of synergy with more courses in the same system with a "simple math of everything" approach.
comment in response to
post
A game like that'd be way cooler.
But one small idea I had for MA was, most questions are multiple choice: maybe it'd be worth making a fancier UI where you give a confidence level with your answer. I'd love it if along with the first-level training you were picking up a sense of when you're BSing.
comment in response to
post
I tried this in January: bsky.app/search?q=mat... and it's a great improvement over high school in many ways, and I can think of more improvements -- but that's still more in the vein of training than open-ended challenge of science
comment in response to
post
just waving at the far future! though Zendo does show you can make a game of doing real science. How could we make a MMORPG where guilds advance by doing that, and there's a niche for people who can help you learn without spoilers? Sounds hard.
comment in response to
post
That the platform is less locked down and easier to adapt/exit makes it more worthy; it's a relative check on its what its owner can get away with doing to us in the future. This is why I joined before the Elon stuff over there. It's just undignified to let ourselves suffer platform lockin.
comment in response to
post
might've been Pong for me too in the 70s, but geez, memories so worn down.
2. TI Extended BASIC
3. The Bayesian Conspiracy
comment in response to
post
I've started using ze/zir in this case. New chance to give neutral pronouns some traction.
I agree Claude should have more of a spine. (Yes, be careful what you wish for...)
comment in response to
post
ze's growing up so quickly :)
comment in response to
post
Must say I'm not too pleased with the role of pointy-haired boss of a junior programmer who's impressively educated but unpracticed at doing things
comment in response to
post
seems that way. I was just trying some more stuff in this session and needed to debug, so tried investigating this some more.
comment in response to
post
This was essentially right, ze did capture the image but then only analyzed rough features using console.log as the output channel, rather than vision on the full image the way it can work on a screenshot.
comment in response to
post
curious seeing which countries were evil or good, then you try 'soviet' and suddenly all the countries are clustered around evil, relative to that paragon
comment in response to
post
Some such possibility is why I qualified it with "no control". But it did look at the phone photo in the turn before and verbalize the problem it was showing.
comment in response to
post
Also Nader was presented there as effective and dishonest, re the other thread in replies
comment in response to
post
Book from back then The Health Hazards of Not Going Nuclear made it really clear suppressing nuclear power meant shortening millions of lives. Granted this was teenaged-me’s evaluation
comment in response to
post
Based
comment in response to
post
Relevant paper probably, also from scheme-land, “trampolined style”
comment in response to
post
I always thought instead of res reduction clocks should be (object) caps
comment in response to
post
There’s a bit of serious writing about private currencies, e.g. Hayek (which I don’t really understand). Haven’t seen pure memecoins.
comment in response to
post
Any idea why the preference for placeholders?
comment in response to
post
I can see finding common goals because I'm a cooperative guy and when it comes to politics we're all human... but it's true I'm not eager
comment in response to
post
The API doesn't have this particular limit, does it? I have "set up for chat by API instead of web" on my to-get-to stack, but it sounds like it'd matter more for you?
comment in response to
post
If you're writing a math lesson and see "This is unintuitive!" in your output, you should probably reconsider. Maybe not till your next draft, but...
comment in response to
post
Haven't worked on this stuff since basically right after that talk, I'm afraid. I do regret not staying at it and having it ready when the world was ready for Wasm.
comment in response to
post
I'd object to misappropriating "memory safety" this way too! I agree. I made a wasm-like VM meant to support C-level programming safely back in 2001/2 ("Idel", gave a CodeCon talk).
Don't know offhand of a term for no-mutable-globals; it's part of "no ambient authority".
comment in response to
post
I can't dig blurring a technical term that way, but I'd agree mutable globals make similar trouble. Liked Mark Miller's thesis "Robust Composition" on this. (He is/was on the JS committee)
comment in response to
post
From a book The Hungry Mind about curiosity and schooling: iirc an anthropologist coded many hours of school activity generously seeking anyone being curious about anything. You can guess the result.
If I get systematic about "study habits" and such, there's a corresponding danger here.
comment in response to
post
Happy to report my dad's house unburned still. Personally I would've bugged out last night.
comment in response to
post
<looks askance at schools that haven't updated>
comment in response to
post
Neat question about fave tactic. Scanning my Lean cheatsheet I find I don't have one yet.
Fave number: tau. Associated to me with thinking for yourself.
comment in response to
post
Good source of not-boring problems with a ramp of difficulty: projecteuler.net
comment in response to
post
Find any sparks of desire to program (not just for money), fan them. It matters for effective learning, and the better sort of job tends to be under someone looking for signs of this.
I guess you know more than me about effective use of LLMs for work. Not true of a lot of your competition.
comment in response to
post
Aha, the book I read was an earlier more academic one by the same author. www.amazon.com/Company-They...
comment in response to
post
The newer take of his was in a short video, iirc. Also an acquaintance of mine emailed him. Afraid I don't have more.
comment in response to
post
There was a book I liked on the Inklings, I think this one? www.amazon.com/Bandersnatch... (It's been a long while)
comment in response to
post
There are some pluses to a situation like this with a camp in a long fighting retreat, but... yeah. Her old thesis advisor seems more reasonable.
comment in response to
post
I tried having a couple long chats with Claude about this problem shortly before 3.5(new), and I think the difference the feedback made was kinda superficial in the end.
comment in response to
post
That's part, but some other issues:
- Talk is different from writing, usually a product of editing. You see "-- wait, I take it all back" much less even than in talk.
- Training on one-token lookahead vs. gen by beam search.
- RLHF seems to push hard accidentally for "flatter the human and hope"
comment in response to
post
OTOH! Around 2009 I was working for an image search startup whose founder was excited about neural nets and CUDA, and I was quite blah about that direction then
comment in response to
post
(2000 me, roughly: who knows, it's a scientific mystery; but if we get computers on the scale of Drexler's we ought to be able to more or less bruteforce it even with no smarter angle on what thinking even is. And the hardware will be way faster at it than us.)
comment in response to
post
2010 or 11 me read Shane Legg's new year's projections and was like "huh, first time someone seeming credible thinks they know how to AGI Soon" (brainlike RL by 2029 in his modal timeline).
Didn't have the neuroscience to guess how credible, but I was keeping it in mind from there. . .
comment in response to
post
Sorry I missed your comment. We haven't discussed this specifically, but he's pretty pragmatic and has written on how he chose Python for pedagogy -- I guess it was norvig.com/python-lisp.... that I'm recalling. (I was an early contributor to aima-python, etc.)
comment in response to
post
It's at least kinda illuminating that people take that question as a call for random striking facts.