Not *exactly* the same question, but when talking with collaborators have been working out some tentative shorthand along the lines of: "let me show you what I have so far, the interface is obviously vibecoded but the model architecture and examples are artisanal".
The Washington Post describing Russia has having "exquisite intelligence capabilities".
The New York Times describing the U.S. military as having "large, exquisite unmanned systems".
Why are American journalists suddenly using the word "exquisite" in a weird, awkward way?
title slide reading: Global Authoring in AI Drama Management Mark J. Nelson Computer Science Department & AU Game Center American University, Washington, D.C.
Also gave a guest lecture in his AI for Interactive Storytelling grad seminar, which was a good opportunity to trial a new talk.
photograph of the Rochester Airport outpost of the Strong Museum of Play arcade exhibit
Had a great short trip to the Rochester Institute of Technology. Thank you to @justusrobertson.bsky.social for hosting!
Donald Knuth asking Claude to update plan.md feels like some kind of time warp
There's a tile like this on each corner entrance and then names in the middle of the park. Richard Branson is the only one I saw in a quick walk past (was doing the Red to Silver transfer to Dulles so didn't have time to stop).
a photograph of a plaque on the ground reading "Jeffrey Epstein Walk of Shame"
a new commemoration in Farragut Square
I wonder if some of that will end up automated by LLMs as a second order effect, not directly by LLMs but because LLM coding makes it cheap enough to finally automate stuff that *could* have been automated 20 years ago from a purely technical perspective. But hey, maybe not!
I'd be curious what he feels pressured by. Is it internal OpenAI disgruntlement or customers? (I doubt it's his own personal regrets.)
The limits of optimal control, from the maximalist and minimalist perspectives.
Here's a tiny bit of hopium:
In 2024 we were planning on adding 160 GW of renewable energy to the grid by 2030.
As of the start of this year, we had added 100 GW and are planning on adding 220 GW more by 2030.
(In 2022 we planned for 60 GW in 2030. We hit that by the end of 2023)
A recent paper (arxiv.org/abs/2602.18671) made me question something basic: do the logits of a language model model the next-token or the full sequence distribution? It really messed with my brain (in a fun way!). I wrote about the paper to clarify my thinking.
vaishnavh.github.io/blog/joint-o...
I would've agreed a few months ago, but isn't Meta's PE-AV a pretrained audio model? It does have a multimodal aspect, but it's still purely an encoder, not an ALM.
It was all about spying on Americans: www.theatlantic.com/technology/2...
that's the WV part of Pittsburgh
A Special issue of Philosophy and the Mind Sciences on Representation in the Neurosciences and AI has just come out and it's full of cool papers: philosophymindscience.org/index.php/ph...
German academic job ads do this too! For some reason they insist on translating "Professor/Professorin" to "Professor (m/f)", to somehow reproduce the gendered German words in a language that doesn't have that. In recent years some then moved from "(m/f)" to "(m/f/a)" or "(any gender)".
Oddest experience so far as interim department chair. Received a letter from a California state prisoner addressed to the department (why Computer Science??). Inside, a hand-drawn diagram explaining how the Kennedy assassination really happened (the perpetrator: "G.H.W.B. CIA").
Coincidentally I saw a talk about that a few weeks ago! Don't really know anything about this area other than the one talk, but it was based on this paper.
I guess technically the filament does sublimate, which contributes to the limited lifespan, but not sure that's what the cover designer was going for.
cover of the book Phase Transitions in Machine Learning
Pedantically annoyed by the cover of this book b/c the operation of an incandescent lightbulb does not really involve phase transitions.
Some of them are even taking and enjoying my AI class, so perhaps "anti-AI" is not exactly the right word, but you know what I mean.
I find virulently anti-AI programmers on the internet sort of annoying, but we have a minority of very anti-AI comp. sci. majors, and in that context it is totally fine to me, maybe even actively good.
V.32bis is all I need!
Not sure if I'm extrapolating too much from a few examples, but I get the impression that some '90s U.S. music scenes were Extremely Online before it was common for the average person to be. Like naming your punk record after a Usenet newsgroup levels of online. Anything good written about it?
About a year later, Gentry unfortunately lost the source code in a hard drive crash, but was able to recover it by getting it back from me, as I still had a copy. Consider sharing your source code with grad students as a form of distributed backup! 2/2
An interesting experience during my PhD: My first grad-school paper involved hooking up an external AI drama manager to the interactive fiction game Anchorhead (Michael S. Gentry, 1998), which was possible because Gentry kindly shared the Inform source code with us. 1/2
a photograph of an open door with snow piled up in the entry way and a fire hydrant
good morning from New Jersey 🥶
