@karchie@freeradical.zone
@karchie@freeradical.zone avatar

karchie

@karchie@freeradical.zone

Escaped Yooper, neurobiology PhD, software tinkerer, medical imaging nerd, Luddite, skier, crotchety old man

This profile is from a federated server and may be incomplete. Browse more on the original instance.

lilithsaintcrow, to random
@lilithsaintcrow@raggedfeathers.com avatar

So…if I were to learn a programming language (I think I once knew BASIC in fourth grade, waaaay back before Al Gore's Internet), which one would my hacker pals suggest?

karchie,
@karchie@freeradical.zone avatar

@lilithsaintcrow a real answer would start with questions like “why do you want to learn?” and “what do you want to do?” but for most people, Python. I very much like @AlSweigart ‘s book here for beginners: https://automatetheboringstuff.com/

karchie, to animals
@karchie@freeradical.zone avatar

one of those days where I pay for schedule flexibility by needing to work on the weekend and Rose is Not Helping by clearly expressing my opinion of work in general and Java in particular

kellogh, to llm
@kellogh@hachyderm.io avatar

a lot of takes today about how anything you can do with an would be better done with “ML”. yes, always from people who never did ML in their life.

see a claim. check their bio, they’ve never even heard of a binary classifier, yet they’re suddenly ML experts

karchie,
@karchie@freeradical.zone avatar

@kellogh haven’t been online a bunch today so I’ll have to look around for this. Strange circumstances have led me to be something of an LLM skeptic, and I’ll still not call it AI because I have one foot back in the days when the connectionists (which camp I was broadly in) were doing something very different from what was then called AI. But trying to draw a line between LLMs and ML? That’s a trick I’ll just have to see

kellogh, to random
@kellogh@hachyderm.io avatar

periodic reminder that the NYT v OpenAI lawsuit is just billionaire hunger games. NYT is trying to open a new revenue stream via lawsuits.

Even if GPT was reproducing the articles (it's not, in a meaningful way), those would be 3-year old articles. NYT is a news agency, nobody was going to buy 3yo articles anyway. There's no lost revenue.

totally fine if you want to pick sides, just realize that you're not fighting for the little guy

karchie,
@karchie@freeradical.zone avatar

@kellogh so…I agree with the larger point that the NYT is not my friend, and my favorite soapbox topic is how US copyright law works against its stated goals. But…while you have to work a bit to find unambiguous examples, there are clearly attractors in the LLM machinery that spit out copyrighted material. Some, probably much, of what’s going on in there is more like associative memory than like generalization. Either cut parameter count until it’s not memorizing, or license the training data

karchie,
@karchie@freeradical.zone avatar

@kellogh the news business is a mess, but each outlet is responsible for their output. If they reproduce copyrighted material, the law is clear. Yes, you have to work to get the commercial LLMs to repeat specific copyrighted material. But it flows right out if you set things up just so. I don’t like the techdirt article’s blending of the training data and the model produced by training on that data, the former has a reasonable fair use argument while the latter is much harder to justify

karchie,
@karchie@freeradical.zone avatar

@kellogh …and the internet archive I can’t really speak about because I don’t know enough. I sure think it’s useful, and the fact that it still exists suggests that they’ve found a defensible fair use case

karchie, to Colorado
@karchie@freeradical.zone avatar

Laser show Christmas tree projected onto a wall, with falling snow illuminated by the beams

karchie, to LearnJapanese
@karchie@freeradical.zone avatar

deleted_by_moderator

    •
    karchie,
    @karchie@freeradical.zone avatar

    @kellogh the Finnish course is very short, 3 sections vs 5 German or 8 Spanish. It doesn’t get beyond simplest tourist phrases plus the Duo humor (“help! there is a moose in the restroom”). I use it to satisfy curiosity about what the language is like (as I’d heard, really damn different from anything else I’ve seen) and maybe lay some groundwork if I ever decide to learn it for real

    mekkaokereke, to random
    @mekkaokereke@hachyderm.io avatar

    Yesterday I spoke to an MIT sophomore hoping to get an interview for a Google internship. The student has already done one summer internship at another company, enters programming competitions, and has coursework in machine learning, distributed systems, and neuroscience. So far they have submitted over 40 internship applications.🤯

    From all their applications, they've only gotten two OAs (online assessments), and no interviews.

    It is so much harder for kids today than when we were young.

    karchie,
    @karchie@freeradical.zone avatar

    @astory @chiasm @mekkaokereke neuroscience PhD here, writing code for a living these last 17 years. It’s been mostly biomedical-ish and sometimes even brains, so the degree is not entirely irrelevant. Took me 6 years of grad school and 2 postdoc to realize that scientists mostly ask people for money for a living, and that I’d really rather be paid more to build things instead. Nothing but approval for people who figure it out faster than I did.

    karchie, to llm
    @karchie@freeradical.zone avatar

    wrestling with the use of code generation, particularly since it’s being gently (for now) encouraged at work. On the one, it’s obviously a CF in the making; on the other, if the target language is, say, Java, maybe all the badness cancels out? Sure I’d rather have a sufficiently expressive language where I can DWIM for myself, but no, we’re doing this in Java. So the real source is now the prompt, though I would want some versioning on the LLM…no, this is still all going to end in tears

    interfluidity, to random
    @interfluidity@zirk.us avatar

    listening to an NPR segment discussing “Gender Queer” as a “banned book”, i looked up “Gender Queer” on Amazon, and it was there, available to purchase. i looked up “Turner Diaries”, and it was not.

    “Gender Queer” is, I am sure, “banned” from many schools and public libraries (which also, I suspect, don’t carry “Turner Diaries”).

    should we be outraged by the suppression (“banning”) of “Turner Diaries” too, or are we arguing more about the criteria of suppression than the fact thereof?

    karchie,
    @karchie@freeradical.zone avatar

    @interfluidity there’s a lot to dig into here, but the short of it: it’s not the gov’t banning Turner Diaries

    alexelcu, to random

    isn't a thing, sorry.

    Online platforms evolve. A big driver is the fact that people want free stuff, forever, paid essentially by other people's money. While this is justified for healthcare, or social security, helping people to live in dignity and thrive, it's not justified for consuming online content.

    "Enshittification" is the same tired rant against The Man, or capitalism (AKA selling the fruits of your labour for a price agreed by the market).

    karchie,
    @karchie@freeradical.zone avatar

    @alexelcu I am puzzled by (1) e14n is just old fashioned tankie capitalism bashing and (2) it’s really just human nature plus the legal and economic environment that makes everything get worse over time. have this meme from The Office somewhere here on my desk…

    hrefna, to random
    @hrefna@hachyderm.io avatar

    WHY ON EARTH DOES THIS ONE AUTHOR LIKE USING POINTER MATH SO MUCH /rant

    karchie,
    @karchie@freeradical.zone avatar

    @hrefna it took me 4 months in 1991 of chasing down slippery bugs to really learn how pointer arithmetic works so damned if I won’t take every opportunity to show that off

    karchie, to boulder
    @karchie@freeradical.zone avatar

    Biked with the cellist to today, 18 miles of mostly gravel trail that they’ve been wanting to ride since we moved here 7 years ago. had lunch at a favorite Pearl St restaurant, then back to by bus

    karchie,
    @karchie@freeradical.zone avatar

    several people wanted to get on the bus but the free fare offered in July and August had ended. I suspect most of them were looking more for a soft seat and AC rather than to get anywhere in particular. So many things wrong with this situation and I get that there are people whose needs are just hard to meet but we’re failing so many of our neighbors because as a society we’re more interested in punitive consequences than in meeting people where they are. it’s heartbreaking.

    anderseknert, to linux
    @anderseknert@hachyderm.io avatar

    Any users using to manage (at least a few of their) packages? Developing on Mac myself and it feels really convenient being able to target both Mac OS and Linux using a single tool, but I have honestly no idea if anyone on Linux is using brew. Are they?

    If not, what would be the best way to distribute CLI tools for Linux? My apps are mostly written in Go, so basically just a binary without dependencies.

    karchie,
    @karchie@freeradical.zone avatar

    @anderseknert nobody uses brew or will install it to get your stuff. If you build a .deb then you get Debian and family (notably Ubuntu), and you can install without too much hassle on most others. If you go on to build an rpm, that covers a very large fraction of the remaining Linux user base, and what’s left is mostly weenies on, like, Arch and Gentoo, and they’ll complain mightily but they know how to convert .deb and secretly enjoy the complaining (Disclaimer: Arch weenie, myself)

    karchie,
    @karchie@freeradical.zone avatar

    @anderseknert though if everything’s statically linked you might just tar up the bare executable with a README

    karchie,
    @karchie@freeradical.zone avatar

    @anderseknert ah. yeah, no good solution to that. Even for one family (say, Debian) it’s either push to the repos for every variant or tell all your users to add a common repo. It’s a wild world out there

    remixtures, to ai Portuguese
    @remixtures@tldr.nettime.org avatar

    : "The purpose is not for the LLM to know the content of any given story or any given novel - the purpose is for it to see the patterns in the output of collective human intelligence.

    That is, this is not Napster. OpenAI hasn’t ‘pirated’ your book or your story in the sense that we normally use that word, and it isn’t handing it out for free. Indeed, it doesn’t need that one novel in particular at all. In Tim O’Reilly’s great phrase, data isn’t oil; data is sand. It’s only valuable in the aggregate of billions,, and your novel or song or article is just one grain of dust in the Great Pyramid. OpenAI could retrain ChatGPT without any newspapers, if it had to, and it might not matter - it might be less able to answer detailed questions about the best new coffee shops on the Upper East Side of Manhattan, but again, that was never the aim. This isn’t supposed to be an oracle or a database. Rather, it’s supposed to be inferring ‘intelligence’ (a placeholder word) from seeing as much as possible of how people talk, as a proxy for how they think.

    On the other hand, it doesn’t need your book or website in particular and doesn’t care what you in particular wrote about, but it does need ‘all’ the books and ‘all’ the websites. It would work if one company removed its content, but not if everyone did."

    https://www.ben-evans.com/benedictevans/2023/8/27/generative-ai-ad-intellectual-property

    karchie,
    @karchie@freeradical.zone avatar

    @remixtures interesting read and I agree with much of it, except for the quoted passage. one way NNs can produce output with the same statistics as the training data is autoassociation: to encode a copy of training data. one way to avoid this is to narrow the network in the middle to force it to find a more compact representation, but how thorough can that be with the giant LLMs? if there are prompts that produce training text, whatever the intent or the process, that’s still plagiarism

    karchie,
    @karchie@freeradical.zone avatar

    @remixtures well sure the machinery isn’t responsible, and there are details that will no doubt get sorted out in courts, but people built and trained and directed the machine. You can’t defend from a copyright claim by saying the photocopier did it

    kellogh, to ai
    @kellogh@hachyderm.io avatar

    new rule: no one gets to talk about accuracy unless they've at least heard of a confusion matrix, maybe even attempted to learn what it is

    karchie,
    @karchie@freeradical.zone avatar

    @kellogh yes but. that’s great for classification models but 95% of what (other) people (not me) call AI these days are generative and, well. I don’t know how one properly evaluates the output of an LLM. I mean there are tools (hi Perplexity) but it feels to me like this isn’t even art yet much less science

    karchie,
    @karchie@freeradical.zone avatar

    @kellogh and so you get people with actual university jobs feeding the Political Compass questions into LLMs to “evaluate bias” and the WaPo covering their preprint like it’s news and ugh give me strength

    futurebird, to art
    @futurebird@sauropods.win avatar

    Saw this old comic book cover... and ... had a reaction. "Monsters and Heroes #7"

    Cropped close up of the "T-Rex" from the cover. It has a very upright posture and a strange neck.
    Marlin Brando saying "look how they massacred my boy"

    karchie,
    @karchie@freeradical.zone avatar

    @futurebird that is a dinosaur who has Seen Some Things

    th, to random

    Remember when computers looked cool?

    karchie,
    @karchie@freeradical.zone avatar

    @kandid @th ooh, I did a summer undergrad research project on the CM-2 but in C* not *Lisp. With a time machine to curate my career I’d sure go back and fix that. I wonder if there’s anything from *Lisp that would be useful on today’s GPUs.

    karchie,
    @karchie@freeradical.zone avatar

    @kandid understood. My day job is but I always have a REPL at hand for prototyping and experiments and testing. Over the years I’ve built a library including small DSL for things I do often. Coworkers are somewhat aware of it all but I can’t be bothered to seriously evangelize

  • All
  • Subscribed
  • Moderated
  • Favorites
  • •
  • JUstTest
  • GTA5RPClips
  • DreamBathrooms
  • thenastyranch
  • magazineikmin
  • everett
  • InstantRegret
  • Youngstown
  • mdbf
  • slotface
  • rosin
  • anitta
  • kavyap
  • modclub
  • normalnudes
  • cubers
  • osvaldo12
  • tacticalgear
  • Durango
  • khanakhh
  • ngwrru68w68
  • provamag3
  • Leos
  • tester
  • cisconetworking
  • ethstaker
  • megavids
  • lostlight
  • All magazines
    •