Juan Diego Rodriguez's Avatar

Juan Diego Rodriguez

@juand-r

CS PhD student at UT Austin in #NLP Interested in language, reasoning, semantics and cognitive science. One day we'll have more efficient, interpretable and robust models! Other interests: math, philosophy, cinema https://www.juandiego-rodriguez.com/

4,571
Followers
2,914
Following
801
Posts
30.10.2023
Joined
Posts Following

Latest posts by Juan Diego Rodriguez @juand-r

Video thumbnail

Ayatolá Made in USA: Surrealismo imperialista distópico teocrático.

06.03.2026 08:11 👍 693 🔁 304 💬 133 📌 58

Yes, very odd

06.03.2026 03:46 👍 1 🔁 0 💬 0 📌 0

Interesting they still do this. Last time I saw it was with GPT 3.5

06.03.2026 03:36 👍 1 🔁 0 💬 1 📌 0
Preview
The Curious Case of Neural Text Degeneration Despite considerable advancements with deep neural language models, the enigma of neural text degeneration persists when these models are tested as text generators. The counter-intuitive empirical obs...

Maybe related to this?

arxiv.org/abs/1904.09751

It’s not literally repeating itself… but close.

06.03.2026 03:03 👍 1 🔁 0 💬 0 📌 0

Also: "Más vale pájaro en mano que ciento volando" (roughly "A bird in the hand is worth two in the bush")

06.03.2026 00:09 👍 1 🔁 0 💬 0 📌 0

Indeed, one thing I learned was that there is no such thing as an experiment which is too silly to run!

06.03.2026 00:03 👍 2 🔁 0 💬 1 📌 0
Post image

Gemini attempts to bribe the CoT summarizer with ASCII coffee

05.03.2026 23:43 👍 275 🔁 30 💬 8 📌 8

Maybe a long shot, but are there any virtual options, e.g., for the talks?

05.03.2026 23:06 👍 0 🔁 0 💬 1 📌 0

Hmm there's been a few. By Byron Wallace, and recently by Ekdeep Singh Lubana.

05.03.2026 23:04 👍 3 🔁 0 💬 0 📌 0

Sometimes you see papers that make you think, “I wish I had written this paper”. How do people deal with this?

05.03.2026 20:29 👍 25 🔁 1 💬 8 📌 0
Preview
Quit ChatGPT: right now! Your subscription is bankrolling authoritarianism | Rutger Bregman As a historian, I’ve studied the major consumer boycotts of history. We can take down ChatGPT and send a powerful signal to Silicon Valley, says author and historian Rutger Bregman

"The most effective consumer boycotts in history share two qualities: they are narrow and they are easy," writes @rutgerbregman.com

04.03.2026 14:44 👍 2237 🔁 1099 💬 71 📌 104

My AI agent:
"Heads up: Codex is terrible (per my own notes). It confabulates and is unreliable. Please switch me back to Claude"

05.03.2026 15:46 👍 18 🔁 0 💬 1 📌 0

New post building on this idea, if you are feeling blue as a developer, build your own stuff. It won't fix everything but it can fix some things. vickiboykis.com/2026/03/04/a...

05.03.2026 14:21 👍 72 🔁 15 💬 2 📌 1
Video thumbnail

🔴 Sánchez recalca que España "repudia" el régimen de Irán pero a la vez "rechaza" el conflicto: "Algunos nos acusarán de ingenuos. Ingenuo es pensar que practicar un seguidismo ciego y servil es liderar. No vamos a ser cómplices por el miedo a las represalias de alguno"

04.03.2026 08:52 👍 609 🔁 184 💬 22 📌 17
Post image

You Jin (b. 1979), “The Secret Garden” (2014), oil on canvas, 160 x 120 cm.

04.03.2026 22:14 👍 75 🔁 12 💬 0 📌 0
Video thumbnail

McGovern: How the hell is this America first? You guys broke your top campaign promise. Good luck with that. I hope the defense contractor money was worth it. Shame on you all. The mask is off. You're all just a bunch of pathetic Neocon Warmongers.

04.03.2026 22:20 👍 20982 🔁 6493 💬 547 📌 347

Trying to imagine the looks on the faces of the people I worked with at Google in 2018 if you’d told them that in 7 years their flagship product would be sending users to commit terrorism at airports

and that no one will do anything about it

05.03.2026 00:54 👍 73 🔁 22 💬 3 📌 0

7. Their uses may turn out to empower the worst side of bureaucracy. I made this argument myself yesterday. Also: bad effects for labor and culture! There are many, many more plausible criticisms, most of which require some engagement with, and understanding of the technology to really make stick.

04.03.2026 13:57 👍 439 🔁 32 💬 3 📌 1

📣 ACL 2026 SRW Direct Submissions are open!
acl2026-srw.github.io
More info below 👇

04.03.2026 15:38 👍 1 🔁 0 💬 0 📌 0
Post image

A cartoon from the 1950’s that’s more relevant than ever 💔

04.03.2026 14:40 👍 19806 🔁 8386 💬 344 📌 322

Kicking a dead parrot

04.03.2026 14:41 👍 34 🔁 4 💬 3 📌 0

Full disclosure I did have to nudge them a bit, e.g.:

“Why are you just waiting, be proactive!”

and

“I shouldn’t be talking to you, stop asking me questions. You’re on your own”

04.03.2026 04:00 👍 2 🔁 0 💬 0 📌 0

Sounds fun. Wish I could attend this!

04.03.2026 02:15 👍 1 🔁 0 💬 0 📌 0

Ok I tried this. It took them a while but the two Claudes finally found each other.

04.03.2026 02:01 👍 27 🔁 1 💬 1 📌 1

Good Luck, Have Fun, Don’t Die was amazing

04.03.2026 00:45 👍 2 🔁 0 💬 1 📌 0

My dog punches me in the face all the time

03.03.2026 20:57 👍 1 🔁 0 💬 0 📌 0

This would take an insane amount of tokens though

03.03.2026 02:30 👍 1 🔁 0 💬 1 📌 0
Preview
Classification of finite simple groups - Wikipedia

en.wikipedia.org/wiki/Classif...

03.03.2026 02:16 👍 1 🔁 0 💬 1 📌 0

Also, my agent works seamlessly when using Claude, but every time I switched to Codex it would get confused, go in circles, and ignore my instructions. It was useless.

02.03.2026 06:17 👍 1 🔁 0 💬 0 📌 0

Codex is pretty good for coding. But so is Claude

02.03.2026 06:07 👍 1 🔁 0 💬 1 📌 0