tedunderwoodillinois

@tedunderwoodillinois@threads.net

Studying literary imagination with machine learning, and vice-versa. Information Sciences & English at UIUC. Also @TedUnderwood.

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

tedunderwoodillinois, to random

There's a one-way bridge from Threads to the Fediverse, and a two-way bridge now between the Fediverse and Bluesky. So I'm going to send this and see if I can forward it all the way to Bluesky.

Hello, people of Bluesky! Here's to decentralization and open borders!

tedunderwoodillinois, to random

I like coding with an LM assistant, but it is a little like a fairytale where you meet magical helpers offering gifts that mayyyy do what you want or mayyyy transform you into a newt.

tedunderwoodillinois, to random

Open-source AI requires open data. There's a lot out there, but one of the obstacles is that older public-domain books have terrible OCR transcription. To that end, Pleias is releasing a billion words of public-domain text with experimental LLM-based OCR correction. https://huggingface.co/datasets/PleIAs/Post-OCR-Correction

tedunderwoodillinois, to random

Have you ever asked a 3-year-old a question they can’t immediately answer? (E.g. “What is Hollywood?” Or “why did you draw on the walls?”)

What comes out is absolutely pure language-model hallucination. You can almost hear them pausing to predict the next word.

RE: https://www.threads.net/@ezraklein/post/C6B13GyulR4

tedunderwoodillinois, to random

I have no interest in augmented reality, but I admit… the one application of AR glasses I could use would be a heads-up display for the names of people on my contact list. Universities are big and I’m operating way over Dunbar’s number.

tedunderwoodillinois, to machinelearning

30 billion words of audio transcriptions from 30 million YouTube videos, in multiple languages. More modalities coming soon. From Pleias. #machinelearning https://huggingface.co/datasets/PleIAs/YouTube-Commons

tedunderwoodillinois, to random

Self-improvement by generating and filtering synthetic data is so much more interesting to me than the LeCun thesis that what we need are systems that more closely replicate the behavior of individual primates.

We didn’t get here by being individually vastly different from bonobos. It’s the collective phenomena of language and culture that make us vastly different.

RE: https://www.threads.net/@sung.kim.mw/post/C5pf2mwrWGk

tedunderwoodillinois, to random

We live in an age of daily miracles. Just think how an app like this would have streamlined the plot of Neuromancer. https://www.youtube.com/watch?v=xZiTSZ5SOYc

tedunderwoodillinois, to random

It’s a struggle to keep up with the pace of change, but imagine how impossible it would be without blog tutorials, open source, github, Jupyter notebooks, &c &c.

The open web built an amazing knowledge infrastructure in the last couple of decades. AI is in a way just a continuation of this project.

tedunderwoodillinois, to random

Random accident that the moon and sun are the same apparent size here. If we had interstellar tourism, this is the One Thing that everyone would know about the Earth, and when they visited they wouldn't want to see anything else. "We also have museums?" we'd say.

tedunderwoodillinois, to random

AI-skeptical discourse is starting to feel a bit like the Black Knight scene from Holy Grail. The Stochastic Parrots paper called LLMs “misdirected research effort” in 2020. Anyone want to reconsider that a few models later in 2022? “Nope: ‘tis but a scratch! deep learning is hitting a wall” (2022). Sure, fine. But now in 2024, with huge investment in foundation models across video and audio, is it time to revisit priors? “Nay, have at you, it’s only a flesh wound!”

tedunderwoodillinois, to random

The Gradient interviews Andrew Lee about Shortwave’s plan for an AI email assistant.

100% persuaded that an assistant to triage, summarize, and write drafts would be a killer app for AI. Not yet persuaded that Shortwave’s search-based approach is going to crack the problem of modeling user needs.

The key stuff happens around -34:00 minutes from the end.

https://open.substack.com/pub/thegradientpub/p/andrew-lee-how-shortwave-ai-email?r=62esl&utm_medium=ios

tedunderwoodillinois, to random

Neat idea by advadnoun of BigSleep fame: generative models that are driven by a user’s past interactions rather than by an explicit prompt. A recommendation system—that creates what it recommends. https://rynmurdock.github.io/posts/2024/3/generative_recomenders/ AI Threads

tedunderwoodillinois, to random

Riley Goodside uses Suno to turn the text of the MIT license into a sad piano ballad — with breathy parentheses and heart-breaking all caps sections — and ... it kind of works?

https://youtu.be/pGbodliLFVE

tedunderwoodillinois, to random

I've seen "critical AI" people saying generous things about the "carefully tailored and curated" datasets that were used to train statistical ML models five or ten years ago — in order to cast the monstrosity of GenAI into sharp relief.

This is quite different from what I was hearing about statistical ML in those quarters five years ago. Which gives me an idea for how GenAI will eventually gain broad acceptance!

tedunderwoodillinois, to random

One of the things I take away from the study of cultural history is a deep sense of how much people distrust change and excel at detecting signs of decline. Social media may amplify this, but the instinct itself is not new.

tedunderwoodillinois, to random

I’m skeptical of “declining standards” stories. My theory is that smartphones/soc media have given young people a direct access to the world that my generation never had. They see peers becoming influencers, boasting about internships, &c. School is not the only place to put aspiration. + https://www.harvardmagazine.com/node/85660

tedunderwoodillinois,

If you went to college in the late 90s,
with internet, you may be like “nah, not much has changed.”

But I’m ancient enough to remember a time when going to a SLAC could mean four years in a tiny town with almost no media consumption. Wars happened and I didn’t notice. I knew nothing about career paths, wasn’t paying attention to contemporary tech. My whole model of self-improvement was to read Wittgenstein & study Greek. It’s kind of lovely, but it required immense naïveté.

tedunderwoodillinois, to random

Hey, in the latest episode of The Gradient Daniel Bashir interviews @jossfong on video production, AI, and science communication.

Two of our best science communicators talking to each other: can’t-miss episode. https://thegradientpub.substack.com/p/joss-fong-videomaking-ai-science-communication

tedunderwoodillinois, to random

In simplifying The Three Body Problem, Netflix has amusingly assumed that unexplained results that contradict existing theory would make researchers commit suicide (“science is broken”!) instead of pulling an all-nighter to get the preprint out.

tedunderwoodillinois, to random

Hello Mastodon! This is the Threads alt of TedUnderwood@sigmoid.social.

  • All
  • Subscribed
  • Moderated
  • Favorites
  • •
  • anitta
  • mdbf
  • thenastyranch
  • Youngstown
  • hgfsjryuu7
  • slotface
  • rosin
  • ngwrru68w68
  • InstantRegret
  • kavyap
  • osvaldo12
  • everett
  • DreamBathrooms
  • PowerRangers
  • normalnudes
  • magazineikmin
  • Durango
  • khanakhh
  • tacticalgear
  • vwfavf
  • ethstaker
  • GTA5RPClips
  • modclub
  • cubers
  • tester
  • cisconetworking
  • Leos
  • provamag3
  • All magazines
    •