Skip Navigation
gnomicutterance Deborah @hachyderm.io

here is my finger .plan:

\#a11y / #accessibility, software, #libraries, #archives, publishing, children's & YA lit, #disability, #W3C

Don't feel obliged to follow back. I delete all my posts after a week. I live on #dreamwidth.

Posts 0
Comments 21
The Google AI isn’t hallucinating about glue in pizza, it’s just over indexing an 11 year old Reddit post by a dude named fucksmith.
  • You just defined a GAN, which (1) good for you! and (2) since it’s one of the gold standards of generative AI with buckets of cash poured into it and we are where we are in terms of safety and accuracy, I assure you that it is, in fact, that hard.

    https://en.m.wikipedia.org/wiki/Generative_adversarial_network

  • The Google AI isn’t hallucinating about glue in pizza, it’s just over indexing an 11 year old Reddit post by a dude named fucksmith.
  • You understand that even LLM’s boosters would describe it the same way I did, right? I hate the damn thing, they adore it, but we all agree on the facts: it’s not a data retrieval system, it’s a statistical pattern generator.

  • The Google AI isn’t hallucinating about glue in pizza, it’s just over indexing an 11 year old Reddit post by a dude named fucksmith.
  • I love coming in and seeing I don’t even need to reply, all y’all are already at it.

  • The Google AI isn’t hallucinating about glue in pizza, it’s just over indexing an 11 year old Reddit post by a dude named fucksmith.
  • Sure, if by "database" you mean "tool that takes every cell in every table, calculates their likelihood of those cells appearing near each other, and then discards the data". Which is a definition of "database" that stretches the word beyond meaning.

    Natural language inputs for data retrieval have existed for a very long time. They used to involve retrieving actual data, though.

  • Slack by default using messages, files etc for building and training LLM models
  • @dgerard I think it's the same DB just with extra tooling and services, but I'm not positive.

  • Slack by default using messages, files etc for building and training LLM models
  • @dgerard @rinze

    Push notifications aren't super reliable, especially on mobile. If you need guaranteed on-time notifications, I wouldn't trust it.

    I've written some automations and webhooks and they're fine. The integration ecosystem isn't really mature; I've disabled a few because they were unstable.

    I run Nextcloud as well and I think of them in sort of the same planet of "I like them but it's not for everybody."

  • Slack by default using messages, files etc for building and training LLM models
  • @dgerard @rinze I host mattermost. It's mostly used by my puzzle hunt team and the various satellite teams that spin up around it (so very intensively used a few times a year), and then the rest of the year it's a quiet social space.

    They really push you toward the corporate offerings; I think those get a lot of dev effort.

    I don't love it but it's fine. My primary gripes are mostly that it so extremely slacklike that all the places where it's lacking compared to slack are very noticeable.

  • The Google AI isn’t hallucinating about glue in pizza, it’s just over indexing an 11 year old Reddit post by a dude named fucksmith.
  • But… how can this be? I have been reliably informed by reply guys surfing in from elsewhere that “LLMs don’t work that way” and “you poor dears don’t understand how tokenizing works” and “it’s just statistically likely that they’ll regurgitate certain text, there’s no plagiarism involved”. Are all our reply guys wrong??? Heaven to Betsy!

  • Stubsack: weekly thread for sneers not worth an entire post, week ending Sunday 26 May 2024
  • This is the point, exactly. It’s not even that it’s self harm, which even the community recognizes. If that it creates active workplace hazards. Even in a completely remote environment, I wouldn’t want to be a software developer with this guy, because if he came up with a perfect algorithm that broke the build, he would keep moving forward, because the point was the algorithm, and not the fact that prod went down.

  • Stubsack: weekly thread for sneers not worth an entire post, week ending Sunday 26 May 2024
  • LWers read that post and said “you might be bipolar!” I read that post and said “I’m going to search your post history and find you ‘acquiring’ ADHD meds, aren’t I?”

    Guess what I absolutely found?

    (To be fair when I looked, from his symptom description it looks like he does have ADHD, but he also talks about mucking with his ritalin dose, taking OTC stimulants, & biohacking, in conversation with people who clearly don’t have ADHD but take the meds.)

  • Remember how ChatGPT totally aced the bar exam? Wow! yeah, turns out that was just a lie
  • I have to create a new awful.systems account and read from there, so I stop seeing these weenies' replies after they get banned. The purified experience is the Awful™️experience.

  • Remember how ChatGPT totally aced the bar exam? Wow! yeah, turns out that was just a lie
  • Speaking as a librarian, you’re a mistaken. It is actually incredibly difficult for most people to know how to verify that a source both exists and says what your unreliable informant claims they say.

    Meanwhile medium sized & large law firms have their own librarians (at least in the US), and in many places lawyers from smaller firms have access to some kind of outsourced research service. The easiest solution in the world is to ask the expert researchers whose cost is billable to the client.

  • Remember how ChatGPT totally aced the bar exam? Wow! yeah, turns out that was just a lie
  • Wikipedia informs me that your diet includes squid, anchovies, mackerel, rockfish, and sardines. Yum yum!

  • Remember how ChatGPT totally aced the bar exam? Wow! yeah, turns out that was just a lie
  • Or when you’re singing karaoke and suddenly break into a chorus of the phone number and email address of a random author of a scholarly paper whom you’ve never met whose paper has nothing to do with the lyrics of “Free Bird“?

  • Remember how ChatGPT totally aced the bar exam? Wow! yeah, turns out that was just a lie
  • How is a human being different from running cat my\_textbook.txt? Checkmate! Give Ken Thompson and Dennis Ritchie billions of dollars, for they have invented artificial general intelligence, and now we may all have robot girlfriends.

  • here's a great and useful rundown of Scott Alexander Siskind's actual awful beliefs, from a centrist who found it difficult to accept until that email leaked
  • holy shit I assumed you meant phrenology in the synecdochic sense like when we say "man, andrew sullivan's gotten out the calipers again, sigh" but no, you meant literal phrenology what the actual.

  • Ilya Sutskever and Jan Leike leave OpenAI
  • Permanently Deleted

  • Stubsack: weekly thread for sneers not worth an entire post, week ending Sunday 19 May 2024
  • I mean can you imagine them at any fandom con?

    …oh wait i can