John Gallagher

I think journals should have a template for reviewers.

Sensible chuckle

I think a really useful to interact with LLMs is to think of your audience as a vector space.

This is an excellent use case.

For those of you doing ML, do you find the model cards useful in any way? What would make model cards more effective?

I love the awkward in-progress prose of students writing out proposals for their final projects. it may be unvarnished prose but there are ideas ready for polishing.

On LLM agents. My critique of the industry is more in overshooting the tooling w.r.t. present capability. I live in what I view as an intermediate world where I want to collaborate with LLMs and where our pros/cons are matched up. The industry lives in a future where fully autonomous entities collaborate in parallel to write all the code and humans are useless. For example, I don't want an Agent that goes off for 20 minutes and comes back with 1,000 lines of code. I certainly don't feel ready to supervise a team of 10 of them. I'd like to go in chunks that I can keep in my head, where an LLM explains the code that it is writing. I'd like it to prove to me that what it did is correct, I want it to pull the API docs and show me that it used things correctly. I want it to make fewer assumptions and ask/collaborate with me when not sure about something. I want to learn along the way and become better as a programmer, not just get served mountains of code that I'm told works. I just think the tools should be more realistic w.r.t. their capability and how they fit into the industry today, and I fear that if this isn't done well we might end up with mountains of slop accumulating across software, and an increase in vulnerabilities, security breaches and etc. x.com/karpathy/statu…

Karpathy starts by saying human supervision is needed only because models aren't yet capable enough. But if you look closely at the obstacle ("I want to learn... and become better as a programmer, not just get served mountains of code that I'm told works"), it doesn't go away "when we reach AGI." +

GPT5 can already do like 500-800 lines of code by itself. I find it's coding skills quite useful but I still gotta implement them.

There is more plot and character development in the first 6 chapters of the Count of Monte Cristo than any *season* of a tv show I've seen in the past ten years.

The perplexity web browser is the end of online education.

Last Friday I gave a keynote at the University of Kentucky, titled, “A Spectrum of GenAI Use in the Classroom.”

I’d love to bring this discussion to other institutions as well as with anyone you think might be interested.

Yesterday, I spent an hour reading reaction from people I don't know about a podcast from journalists (Klein, Coates) who were talking about their writing that was a reaction to a Kirk. My kiddo came up to me, asking for a hug worried about school. I need to get offline. I need to touch more grass.

When you watch children learn, when they're screaming because it's so hard, or they shake because they don't understand, and you hug them and you talk and together you all stop for snack and then the goal finally happens, you realize that learning is not training a model.

So jealous!

substack.com/@meresophistry

I'm thinking of turning my blog essays into a book. Would you be interested in reading a collection of essays from it?

So I know there are lots of problems in the world but HADES 2 IS NOT ON PlayStation?!

I'm crestfallen.

I couldn't find the language for describing the training of the model because the attempts to dismiss DL

hahahah, the second one. but we did have a talk about the size.

I got into a heated 1 hour conversation with a colleague (and my best friend) about whether topic modeling was a LLM. I said it was not. But wow it took me longer than it should have to articulate why.

arxiv.org/pdf/2508.16040

"Disproportionate Voices: Participation Inequality and Hostile Engagement in News Comments"

"...the top 10% of frequent commenters account for nearly half of all comments in news comment sections (50.11%), while the least active half (bottom 50%) contributes only 14.99% of total comments..."

Lebron James is starting his 23rd season next month. He's been playing basketball since George W. Bush's *first* term. His son is in the NBA. An incredible feat.

My students and I talked about this today. I said it's like a crystal ball/fortune teller aura.

Seeking human traces in AI writing Introduction

My newest blog post is up: "Seeking human traces in AI writing" (2600 words)

I appreciate you taking the time to read it. Let me know what you think.

meresophistry.substack.com/p/seeking-hu...

“Time is a great teacher, but unfortunately it kills all its pupils.” ~Hector Berlioz

He only drinks it at work.

Everyone once a while I peek into my spam email report for a chuckle. This morning, I saw someone trying to sell me a "microgravity reactor." 🥴

Not enough people in the national media realize that the late night shows are actually youtube channels to most people