Sign in

David Gleich

@dgleich.bsky.social
118 followers 79 following 440 posts

Purdue CS. I like matrices, graphs, networks, tensors if they fit, energy efficient stuff, Julia (I’m ex-Matlab), Laura (my wife), data, clean air, 3D printing, vaccines (especially for kids), cooking, hiking, skiing. I’m quiet so I don’t say too much.

PostsRepliesMedia
David Gleich @dgleich.bsky.social · 08/10/2026
Crazy. I ran a cost estimate for a solar and battery system. If you use current low end prices it ties our current kWh price from grid (.18 kWh) assuming replacing batteries every 10 years and panels every 25 over 50 years. But needs 30-40 kw of panels. So not feasible to build.
000
David Gleich @dgleich.bsky.social · 07/10/2026
The matching result is also funny. Run a “fallback” algorithm if the graph is less than 10^(30000000000000000000000) vertices.
060
David Gleich @dgleich.bsky.social · 07/10/2026
Me: Ahh! new “almost-linear time matching algorithm” … can we implement it? Codex: No. It just uses a normal algorithm for anything with less than ~10^(30000000000000000000000) vertices. Me: Oh, my graphs aren’t that big.
A dialogue with Codex analyzing the new algorithm from OpenAI. It reports it will need graphs with 1 followed by 30 sextillion zeros in order to work.
151
Reposted by David Gleich
Stefan Neumann @neumannstefan.com · 07/10/2026
So, the OpenAI results include: - A proof the unique games conjecture - L = RL - Matrix multiplication in n^(9/4) time - Almost-linear-time exact matching - Integer multiplication faster than n log n - Quasipolynomial algorithms for mean-payoff and parity games Overview: github.com/openai/math/...
github.com
math/overview.pdf at main · openai/math
Contribute to openai/math development by creating an account on GitHub.
1197
David Gleich @dgleich.bsky.social · 06/10/2026
I love how a verbed-noun “gate” has now replaced the standard “check” or “test” in technical vernacular. Thanks Anthropic!
000
Reposted by David Gleich
Andrew C 💯 @arc315.bsky.social · 05/10/2026
In a sane world those would be treated as cyber crimes and the people who are responsible for those agents would be charged.
041
Reposted by David Gleich
George Pearkes @peark.es · 01/10/2026
Modern economic development is about maximizing complexity. To make things more complex, you need abstraction. Abstraction requires organization and communication.
235621
David Gleich @dgleich.bsky.social · 30/09/2026
Quality isnt so hard. It’s in the eye of the beholder and inherently subjective. We can even disagree on quality! That’s fine. What’s hard is that we tell ourselves we are objective and passive observers of “science / truth” and these viewpoints conflict. Conferences need more words. maybe more soon
010
Reposted by David Gleich
Brett Karlan @oldjerryfodor.bsky.social · 29/09/2026
Since I became a professor, the only movies I can sit through are dumb, loud action movies. Everything else I can't get through. Depressed ruminations about the human condition? Buddy, that's my day job! Now, Sandra Bullock keeping a bus going above 50mph so it doesn't explode? I'm interested..
3959034
David Gleich @dgleich.bsky.social · 29/09/2026
I’m annoyed by all the academics stressing “quality over quantity” as a tonic for the LLM deluge of results / crap. People have been saying this for years. Either setup a compact with some language on putting this in hiring documents/recommendations/etc. or stop saying it.
2132
David Gleich @dgleich.bsky.social · 27/09/2026
ChatGPT gave me bad advice on how to fix an overly bitter ipa.it suggested adding salt. I asked about lemon. Did a taste test. Lemon was better. “Super intelligent” but ignorant about human experience.
010
David Gleich @dgleich.bsky.social · 26/09/2026
Ask it to turn it into a tauri app. Loving the tauri app I have for presentations… (see eigendeck.dev
eigendeck.dev
Eigendeck — Math and demos, together
A desktop presentation tool for technical talks: embedded interactive demos, Jupyter notebooks, and LaTeX math in one deck — so the math, explanation, code, and live demos live together.
020
Reposted by David Gleich
Jonathan Ladd @jonathanmladd.com · 23/09/2026
I always get a lot of grief online when I write that the solution to America's problems cannot be "Democrats win every election forever." That's not democracy. The solution is having at least two parties where the major center-right and center-left parties accept liberal democracy.
1614720
David Gleich @dgleich.bsky.social · 23/09/2026
I guess some folks are ambitious at taking rights away from others. Not that kind of ambition!
000
David Gleich @dgleich.bsky.social · 23/09/2026
I miss ambition in government! Let’s set amazing and maybe even borderline unrealistic goals and see what we can do.
100
David Gleich @dgleich.bsky.social · 23/09/2026
Thankfully I’m a no on the third bullet point.
000
Reposted by David Gleich
Dean Eckles @eckles.bsky.social · 23/09/2026
"Why is There so Little Money in U.S. Politics?"
171
David Gleich @dgleich.bsky.social · 23/09/2026
This feels like a joke!
010
David Gleich @dgleich.bsky.social · 23/09/2026
Clean air kits are great! Super fun to put together. Not just for school!
031
David Gleich @dgleich.bsky.social · 22/09/2026
Oh interesting. I haven’t been able to get that to work. Maybe worth trying again. What model?
110
David Gleich @dgleich.bsky.social · 22/09/2026
Hah. Funny to think about grad students in these sentences instead. First drafts I have to fix! The difference with grad students is that their humanity actually makes me care. (And they learn and get better and generally are fun and smart…)
030
David Gleich @dgleich.bsky.social · 22/09/2026
Is there some magic I need to understand to get LLMs to help with writing? I find them so terrible at this I feel like I’m wasting time trying to use them.
250
David Gleich @dgleich.bsky.social · 21/09/2026
Nothing like a sternly workshopped and wordsmithed statement to get the point across!
010
David Gleich @dgleich.bsky.social · 21/09/2026
Thanks johan. I come from a small ugrad school with a strong honor code (at that time) regarding cheating etc. this seems to be a missing component of these debates too.
010
Reposted by David Gleich
Carl T. Bergstrom @carlbergstrom.com · 19/09/2026
Between the human verification, and the three factor authentication and so forth, I can transfer my life savings to a random bitcoin account easier than I can read a Nature paper my library subscribes to.
938456
David Gleich @dgleich.bsky.social · 19/09/2026
True for me too. My university login is more locked down than my banks.
041
David Gleich @dgleich.bsky.social · 17/09/2026
They took the json output use cases, formalized common cases them, and 100x-ed them. Solid work. I bet nvidia buys them soon.
190
David Gleich @dgleich.bsky.social · 17/09/2026
Yeah. The ceo said it at hn. Crazy this isn’t in the docs though!
110
David Gleich @dgleich.bsky.social · 17/09/2026
Couldn't find it in the docs (site:typesafe.ai "Bernoulli"), unless you count HackerNews as docs! news.ycombinator.com/item?id=4971...
news.ycombinator.com
100
David Gleich @dgleich.bsky.social · 17/09/2026
Re: Jev. I bet they used the word ‘noul’ so they could track usage of their tool in codebases. docs.typesafe.ai/primitive… I can’t find any actual explanation of this choice. Claude thinks it’s part of ‘Ber-noul-li’ as in a Bernoulli trial. I can’t find an actual explanation anywhere.
110
David Gleich @dgleich.bsky.social · 17/09/2026
Coding agents are such a blessing to work with new APIs.
020
David Gleich @dgleich.bsky.social · 15/09/2026
Wild – if the data are accurate on the M6 perf. The M6 is as fast as the original M1 ultra in multicore CPU perf.
010
David Gleich @dgleich.bsky.social · 15/09/2026
Why are there so many rumors about llm efforts? These seem super leaky! These fields are not really that big at all.
000
David Gleich @dgleich.bsky.social · 14/09/2026
I was wondering why they “cannot” do this investigation (initial post) but it turns out they can. Compare web.archive.org/web/20260… to web.archive.org/web/20260…
010
David Gleich @dgleich.bsky.social · 14/09/2026
My sister is big on Instagram. I doubt many people would believe the vile DMs people send her. It’s so important to call this behavior out as @peark.es does. She posts some of them publicly too. Platforms do little to help.
040
Reposted by David Gleich
George Pearkes @peark.es · 14/09/2026
Here's a person that thinks posting the deranged cursing at me in a DM from someone is "leaking" a DM. What kind of moron do you need to be to think spewing hatred at someone in DMs is somehow privileged from public view? Very, very stupid people.
182739
David Gleich @dgleich.bsky.social · 14/09/2026
LinkedIn has the worst discourse on LLMs.
000
David Gleich @dgleich.bsky.social · 14/09/2026
And I was going to post something this morning about opus 5 being really bad….
000
David Gleich @dgleich.bsky.social · 14/09/2026
Super cool!
031
Reposted by David Gleich
Kevin Reuning @reuning.bsky.social · 14/09/2026
I'll post more about this tomorrow, but I've spent the last few weeks trying to do some social network analysis on the most recent "OpenAI Swarm" You can do some interesting things with the data to identify how the different "Agents" interacted. kevinreuning.com/blog/ai_swar...
kevinreuning.com
OpenAI Swarm – Kevin Reuning
095
Reposted by David Gleich
John Voorheis @johnvoorheis.bsky.social · 13/09/2026
What if another name for "jagged frontier" is "not a general purpose technology"?
083
David Gleich @dgleich.bsky.social · 10/09/2026
Is anyone else surprised Open AI can’t answer about the impact of user-info for training their LLM? It seems like searching for “the forcing function is smooth option c and d in Fefferman” (or simple variations) in user-derived training data is a feasible task. I can’t imagine it is that common.
000
David Gleich @dgleich.bsky.social · 10/09/2026
Does anyone else seem like they can feel the impact of rope in llm writing? It loves to use a lot of words to say little.
000
David Gleich @dgleich.bsky.social · 08/09/2026
What harness are you using?
100
David Gleich @dgleich.bsky.social · 08/09/2026
I tried to use Fable 5.1 to review a commit. It blew through my entire usage in 10 minutes… (Anthropic Max plan…) it might as well not exist because I can’t use it even to do simple stuff.
020
Reposted by David Gleich
Isaiah Bishop @isaiahbishop.bsky.social · 07/09/2026
Since its coming up a bit, one of the reasons i think Carney, Mamdani do well isn't liberalism/socialism or whatever, but its that they really represent their values in a genuine way. Actually believing in shit. (i think thats plausibly why climate and bridge is dangerous for carney)
425424
David Gleich @dgleich.bsky.social · 03/09/2026
Agree... maybe I need to try grammarly to see what they are doing. Or build something that uses my codex subscription for this.
000
David Gleich @dgleich.bsky.social · 03/09/2026
Here is my script to do this github.com/dgleich/clau... but there are lots of others too e.g. github.com/tintinweb/cl...
github.com
GitHub - dgleich/claude-docker: My claude-docker script.
My claude-docker script. Contribute to dgleich/claude-docker development by creating an account on GitHub.
020
David Gleich @dgleich.bsky.social · 03/09/2026
I try to keep them isolated to containers.
150
David Gleich @dgleich.bsky.social · 03/09/2026
Have you found anyone who has a useful interface for getting llm feedback? Sometimes it’s useful but they always want to rewrite things!
100