Ryan Butner

We're all "Kids left alone with mom's tablet who fall into an Elsa-gate style algorithm sink" now

This will sound dismissive - and it is - but I think the "AI is only hype" crowd have over-corrected on their skepticism. Currently reads like telling on one's self

Its not going to replace workers at the promised scales (yet?), but its certainly eating the busy-work/intro-level work for many.

I've felt this pretty acutely in my own work, esp. when put against growing pressure from my leadership to mentor junior/less domain-familiar peers.

I dont use LLMs for particularly exotic applications, but its actually making it *more difficult* to mentor than to just augment myself with an LLM

the computer knows where it is at all times. it knows this because of unregulated bulk resale of user data and metadata (2025)

The most misleading thing that you will get from people describing their process of solving a problem is how many things didn't work. Elite troubleshooting is constantly hitting your head against walls. You think the experts don't run into this, but they do. They just know all the walls to hit.

Statement about subversive hidden LLM prompts Submitting a paper with a "hidden" prompt is scientific misconduct if that prompt is intended to obtain a favorable review from an LLM. The inclusion of such a prompt is an attempt to subvert the peer-review process. Although ICML 2025 reviewers are forbidden from using LLMs to produce their reviews of paper submissions, this fact does not excuse the attempted subversion. (For an analogous example, consider that an author who tries to bribe a reviewer for a favorable review is engaging in misconduct even though the reviewer is not supposed to accept bribes.) Note that this use of hidden prompts is distinct from those intended to detect if LLMs are being used by reviewers; the latter is an acceptable use of hidden prompts.

ICML’s Statement about subversive hidden LLM prompts

We live in a weird timeline…

icml.cc/Conferences/...

Efforts to fight the war against misinfo on the basis of information lost it by overlooking attention.

Its pretty fitting in that regard that experts should find themselves flailing helplessly for the last decade or more against these trends: their only reaction is ever to introduce more information.

But that wont help them anymore, will it?

"In an information-rich world, the wealth of information means a dearth of something else: a scarcity of whatever it is that information consumes. What information consumes is rather obvious: it consumes the attention of its recipients. Hence a wealth of information creates a poverty of attention."

I suppose I cant contest the theory we're in the Eye of Terror too much.

After this week’s display, we’re going to see increasingly coherent and successful attempts at legislating AI knowledge products into mandatorily generating regime-approved “facts”, now that it has been shown that such control is and can be exerted

In news about people citing AI-fabricated nonexistent works, it sure seems like way too many professionals keyword search sources and drop them directly into a paper as citations without ever reading the articles or assessing the relevance to their argument; which is NOT how research works!

Tech companies only have five ideas: robot slave (actually just human slaves), hallucinatory counter-reality, untaxable money, The Everything App, and Clippy

If my computer wasnt offline during house painting I'd dig up the crazy glazing pro-Obama threads on 4chan /b/ lol.
He was cited as the "first meme President."

Eventually you need to realize this is less a condemnation of the powers that be than it is of the perennial opposition who is otherwise incapable of doing anything about it.

And yet if this war is so unpopular/illegal, why is it still happening anyways?

Check mate, nerds.

While everything is indeed horrible, you are not obligated to solely focus on the horror. The availability of a constant stream of bad news does not make that news valuable by itself. Paying constant attention is not "doing something." So if the information isn't actionable, filter it.

Facebook is great - it shows me all the recent AI trends.

But lately the vibe has been "body horror with a substantial carbon footprint"

The inflection point for me on LLM skepticism was in utilizing the models for something outside my wheelhouse/realm of expertise. The model can't outperform me 1:1 at things I am good at, but it can enable me to be passably performant at everything else I suck at.

I dont even recall the last time I attempted to look in on X for this very reason. Screenshots that leak out of the platform at, for the most part, the only reason the platform still has any ability to reach the outside world.

ai has already kind of materially ruined the basic act of looking at pictures

The European mind cannot fathom every American household has two gigabins

In my day to day work I'm increasingly seeing where thoughtful use of these models is actually a net good. SMEs without data science/SWE backgrounds can suddenly test and experiment with things at scales that were prohibitive before - but it remains a far cry from full-automation dystopia.

I once had a colleague tell me there used to be people whose entire role it was to be highly proficient at Microsoft Excel.

The skills to be truly good at Excel weren't often possessed other SMEs, but as it got easier to use, those roles went away.

I think about that a lot lately.

Its besides the point if any of them believe that or not - but it is entirely the point that you'll be spending more energy, time, and effort explaining why its wrong to an audience that doesnt know or care.

And you'll be busy doing that instead of literally anything else.

What's wrong babe? You've hardly touched your Spaghetti-baby

1. LLM-generated code tries to run code from online software packages. Which is normal but
2. The packages don’t exist. Which would normally cause an error but
3. Nefarious people have made malware under the package names that LLMs make up most often. So
4. Now the LLM code points to malware.