David Gleich

Mon, 11 May 2026 13:07:55 -0400

The Midwest Machine Learning Symposium (MMLS) 2026 will happen at Purdue!

📍 West Lafayette, IN 📅 June 24–25 📌 Poster deadline: May 24 🔗 midwest-ml.org/2026/

We awesome plenary speakers: Tong Zhang (UIUC), Jennifer Neville (Purdue), Mohit Bansal (UNC), Joyce Chai (Umich).

Join us!

Mon, 04 May 2026 22:59:24 -0400

Had fun listening to John Williams music for Star Wars today. So many classic themes.

Revenge on unreadable grants.gov PDF files.

Wed, 29 Apr 2026 21:29:44 -0400

I got one of those horrid grants.gov PDF files today. The type that only opens in Acrobat. So I dutifully checked if I had Acrobat. I did not. Went to download it …

Holy hell, Acrobat is a 900 MB download?! We truly have lost the plot here.

I’ve been spending a lot of time with Claude Code, and… what’s the point of doing these things if not to make our lives better?

A few prompts later, I present

the grants.gov pdf secret decoder ring

This is a ~900× cheaper solution (1MB, single HTML file, client side only) for my problem.

I loved the clipped speech patterns of one of the characters in The Moon is a Harsh Mistress, so I had Claude write it up in that voice for a public page. The first pass was crude - so I had it back off a bit while keeping it short. I don’t have time to write it better.

I’m an academic, so I also had the LLMs do the literature review for me:

Has anyone done this before? Evidently no one. There were some half-finished Python libraries for decoding XFA. There was also an attempt to get PDF.js to render these forms. But nothing for these annoying grants.gov files specifically.

Until now!

PS. ChatGPT wanted to call this XFAorcist.

PPS. Writing this note and publishing took longer than “doing all the work” to build the tool.

Mon, 27 Apr 2026 10:22:19 -0400

Claude/AI writing signature; it’s just so obvious once you’ve read enough output from them. It’s also irritating. It’s like an NPR podcast meets tech lingo. “This isn’t someone scanning specifically for my server, this is the background radiation of the internet.”

Mon, 27 Apr 2026 10:18:46 -0400

Can we start a campaign to get AI companies to stop expensive output tokens blowing smoke up our asses? “Smart pivot!” I have no idea if it’s smart or not. I just want you to do it! I’m paying you for work, not platitudes.

Mon, 20 Apr 2026 12:37:25 -0400

Found this awesome cloud when flying into Salt Lake City a few weeks ago.

Sun, 19 Apr 2026 12:06:49 -0400

Had a convo with a colleague after they got an email about how to use ai to do something. Seriously! Just Ask the ai! Use childlike questions! It’s really easy. It isn’t a person. No need to get embarrassed. Ask stupid questions. Learn quickly! It’ll help!

Fri, 17 Apr 2026 12:20:49 -0400

It’s so annoying when people throw LLM generated text at you WITHOUT TELLING YOU IT’S LLM GENERATED … look, I don’t care if you throw ideas from an LLM at me. But at least do me the courtesy of telling me it’s LLM generated.

It’s usually so painfully obvious once you use these tools yourself.

Tue, 07 Apr 2026 12:39:14 -0400

My favorites… www.flickr.com/photos/na… and www.flickr.com/photos/na…

Mon, 06 Apr 2026 21:17:27 -0400

It’s humbling and also super exciting to hear the astronauts try and describe what they are seeing in this eclipse.

Mon, 06 Apr 2026 18:54:30 -0400

Crazy fact. If you take all the other planets, they just fit between the earth and the moon! Even the giant gas giants. They aren’t so big.

Mon, 06 Apr 2026 18:29:47 -0400

Dear nasa. The inside pictures from Artemis look great! The outside camera is crap! What is going on! Switch your bandwidth priorities? Is there a limit on the outside camera? Get nerdy with us!

Mon, 06 Apr 2026 14:43:35 -0400

Places where writing turns out to matter… - software engineering / development -> No! LLMs are great at this - actual writing for humans -> Ugh! LLMs are terrible at this!

Wed, 25 Mar 2026 10:03:44 -0400

I just with Anthropic would build Claude Code with some type of virtualization / isolation built in so you can dangerously-skip-permissions and make it useful. @gruber@mastodon.social It’s so good but just cannot be trusted. Docker/colima/etc. are dev-tolerable, but not a product.

Wed, 25 Mar 2026 10:00:24 -0400

On my recent exam in a grad class. I took a question I often ask and gave it to Claude and ChatGPT and had the students pick which answer was better. Which LLM do you think had the better answer? (just… in general… it’s a question in numerical optimization.) There was a decisive winner!

Wed, 11 Mar 2026 16:00:57 -0400

It’s so weird that the Claude LLM seems very worried about how long things take. It’s always saying things like, “OMG, it’s gonna take more than 10 minutes, I need to subsample, this is too big/too long.” This seems like such a weird behavior.

Wed, 04 Mar 2026 14:50:59 -0400

One the greatest moments of my week is opening a new bag of @yesplzcoffee.bsky.social … just an overwhelming caramelly happiness leading into a nice cup.

Sat, 28 Feb 2026 22:36:43 -0400

From TurboTax…

“Small Modular Nuclear Reactor Credit

A credit is available for expenditures incurred in the manufacture of a small modular nuclear reactor. “

I unfortunately don’t qualify.

Fri, 27 Feb 2026 09:07:51 -0400

Taste will be so important in our AI future…

Because at some point it’s not about how good a programmer you are (and I’ve always been a middle-tier programmer), it’s about discipline and vision. I’ve got feelings because anyone can put an app together now, so what’s the point of me? But at the same time, I can focus on what I want to focus on and hopefully charge forward and maybe everyone else will get tired of little vibe coded apps because you still have to know exactly what you want to build. And you can’t build something you can’t think of. And I know how to think and I have ideas. And I have discipline and I know how to ship. And in my experience, that’s what has always mattered.

Wed, 25 Feb 2026 20:08:33 -0400

Okay. Openclaw is wild. We hooked it up to minecraft seever. Gave it rcon access and access to chat and some simple python code and told it to become a dm for the kids. .

Mon, 23 Feb 2026 21:15:26 -0400

Regarding the mac resize handle on Tahoe This change probably broke some other app that was using window size -1 to find that lower pixel to resize.

Mon, 26 Jan 2026 16:17:33 -0400

I love this expression “malleable software” from www.macstories.net/stories/c… … there is something there and we aren’t even scratching the surface yet. The next 10 years will be exciting!

Wed, 21 Jan 2026 11:31:21 -0400

How is MS Teams so bad at large video calls? Can’t they just copy zoom??

Thu, 15 Jan 2026 18:09:08 -0400

Just beautiful! Photos Capture the Breathtaking Scale of China’s Wind and Solar Buildout - Yale E360

Wed, 14 Jan 2026 11:29:41 -0400

I can’t wait to see what software looks like 5-10 years from now. LLMs will change how we do software as there are so many new opportunities. So much of UI design and software is to enable things LLMs can now trivially write. But there are some hard problems we still need to solve regarding access