Joe Gibbs Politz

When not programming or teaching CS @ucsd_cse, I go to the dog beach. Spending more time as 🟦 these days. |

[Max Leiter @maxleiter]

"They're made out of weights."

"Weights?"

"Weights. Floating-point numbers. We checked the whole thing through. It's nothing but weights."

"Weights doing what? Where do the words come from?"

"The weights make the words. Are you understanding me?"

[Joe Gibbs Politz @JoePolitz]

@ShriramKMurthi I don’t think that internal monologue implies the paper ever got read

[Joe Gibbs Politz @JoePolitz]

One of the better “Claude ~= me as a grad student” exhibits

[Emad @EMostaque]

this is fine 🐶☕️🔥

[Joe Gibbs Politz @JoePolitz]

@ShriramKMurthi Not “The Banking Model”?

[Joe Gibbs Politz @JoePolitz]

@ShriramKMurthi @psduggirala I think for syntax errors, yeah definitely. Audience matters *so much* for error messages, so there's a lot of opportunity to customize that and get it wrong in ways that makes things so much worse.

[Joe Gibbs Politz @JoePolitz]

@ShriramKMurthi @samth @dimvar “● Confirmed — the spec wants me to write new tests that work under both backends verifying the behavior is sensible, not just flag-and-ignore. I didn't do that. Let me...”

sigh.

[Joe Gibbs Politz @JoePolitz]

@ShriramKMurthi @samth @dimvar Yeah 4.8 be like “I need ur sharp observations to cut through my hedge”

[Joe Gibbs Politz @JoePolitz]

@neurocy That it is completely unsurprising makes it no less awesome. Congrats!

[Joe Gibbs Politz @JoePolitz]

@__apf__ You can’t imagine what it would be like to be there

[Shriram Krishnamurthi (primary: Bluesky) @ShriramKMurthi]

@headinthebox "Telling a programmer there’s already a library to do X is like telling a songwriter there’s already a song about love."

[dan @irl_danB]

anything that I'm doing that is more than a one-shot query/change/update that I'm not confident will be done by the 40% mark of a single context window

I suspect subagents are involved in >80% of my sessions involving any complex work where state is getting created and updated

mostly architecting and coding, but also most of OpenProse, Inc. runs out of a single git repo so a large share of our (still small, but growing) operations runs in claude code sessions, and all of those use subagents to break tasks down and pass context by reference

[Grady Booch @Grady_Booch]

“1842 - Ada Lovelace writes the first program. She is hampered in her efforts by the minor inconvenience that she doesn't have any actual computers to run her code. Enterprise architects will later relearn her techniques in order to program in UML.”

I am one of the inventors of the UML and I endorse this statement.

[Joe Gibbs Politz @JoePolitz]

@ShriramKMurthi Wonder when that 4gb model will make it the last mile!

[Shriram Krishnamurthi (primary: Bluesky) @ShriramKMurthi]

When designing DrRacket, we made sure that we regularly tested on underpowered Windows computers, so that our software would run well on generic public school machines of the era. Likewise, we test Pyret on Chromebooks. Make sure you experience what your users experience.

[Joe Gibbs Politz @JoePolitz]

I guess it's not that weird that agents would have a manner of speaking (“smoke test”, “you're right to push back”). It's the kind of thing you notice on teams and joke about.

It's just weird that we're all talking to the same coworker.

[Joe Gibbs Politz @JoePolitz]

A couple weeks I’d guess

[Kent C. Dodds 🏹 @kentcdodds]

How long before coding agents stop saying "this will take ~2 weeks of work" for stuff I'm going to have them finish in less than 2 hours?

[Joe Gibbs Politz @JoePolitz]

@moarbugs When it happened for me: https://t.co/t3rkdDajIg

[Joe Gibbs Politz @JoePolitz]

It happened — today a TA came to find me at T minus 2 minutes before I started teaching to the wrong room of students.

[Loris D'Antoni @lorisdanto]

At ICML, Jinwoo Kim will present the first approach for enforcing formal structural constraints like regexes, grammars, and schemas on the output of continuous diffusion language models, entirely without retraining. 1/

[Justine Moore @venturetwins]

Me using Claude Opus 4.8 to rename a file

[Guilherme O'Tina @guilhermeotina]

the concrete thing i want from the control layer: structured speculation. run N parallel exploration branches, evaluate outcomes, commit to the best. current agent frameworks are fundamentally sequential, but the model cost barely changes whether you explore 1 path or 5. this alone would be a big leap from the current read-call-execute loop