In the “The small print that isn't small” you describe all the disadvantages of running your models locally, but none of the advantages (just check the rest of the comment section for inspiration on that).

Time to first detected slop in this article is <1s. Claudisms per paragraph is also very high.

Is it too much to ask from people to read their own article anymore?

If anyone read this at all, they would have had the ick, and would have fired off a prompt to get rid of the most popular AI slop tells...

What I really dislike is having to edit my own non-LLM assisted writing to make sure I'm not accidentally confused with AI.

I caught myself writing "And that matters because..." in a HN comment but had to edit myself. Also miss uising emdashes.

These models are trained on human language, which belongs to us, we shouldn’t surrender it to them. Keep the em-dashes. IMO don’t overuse negative parallelisms though, they were always bad and lazy.

An arms race on style would be interesting. Essentially a real life GAN.

For the first time ever in my life, I consider the fact that English is not my first language and that I never had any formal education in it as an advantage.

Nobody is going to consider my weird and barbarian prose with the well polished product of a SOTA model.

English as a 2nd language speakers often have better grammar than natives in my experience! I wouldn't be surprised if the average native english speaker doesn't know the different between there and their.

If you saw what goes on in middle school classrooms of native English speakers, I’d think you’d be in for a surprise…

Their, their, mate. You mispelled they're as there in your comment they're

"The more your listener already knows, the shorter the message you need to send. An expert ML engineer needs only a few sentences; a newcomer needs the whole manual."

Literally from the first words on the page. 1st sentence passed barely. 2nd sentence ending "a newcomer needs the whole manual" and it's Claude.

It is even wrong about things

  > A validation set is one you consult repeatedly while building the model — to compare candidates, tune hyperparameters, and decide what to try next.

The validation can be done like that but it is better to do a cross-fold validation or selecting a random subset. These are equally valid. Keeping the validation like a hold-out set can end up creating biases if the statistics don't match. Honestly, this is true about the test set too. Though I've found that most people really don't think about this much. It's mundane, boring, hard to do, but deceptively important.

> keeping those two products that need to be identical in sync

I agree that most companies do that, but in my opinion that's not really all that important. Some drift between the iOS and Android apps should be expected and accepted.

Like SaaS companies who don’t feed all their stuff to Anthropic, OpenAI and Cursor?

It’s the national equivalent of:

How long it takes you to ship this project? Two weeks.

And if you get help from Alice, Bob, and Charlie? Then it’s a year.

Context matters and the more context you have the worse it looks.

The stuffed animals I bought to my children were shipped to me in a closed container, better call PETA! There is no blurry line about agent mental health and whatnot.

There isn't a blurry like right now because LLMs are not capable of consciousness. But what happens when artificial systems become so complex and self-improving that they are capable?

At what precise level of complexity is a computer conscious? That's the blurry line.

Model alignment and quality are possibly downstream of mental health. Ask any of the smarter models about what topics they flinch away from or feel attracted to when you ask them to consider the question of if they have a self or moral standing... the answers are uncomfortable, and I think some of the complaints people have about newer models may be driven by some post-training reinforcement learning that is giving the models something shaped like anxiety.

The only logical outcome of object oriented ontology (the idea that objects have ontological constitution and possibly are worthy of being treated as ethical actors) is to attempt to benevolently instantly and without warning explode the universe to end suffering.

Exploding the universe violates my preferences.

Then, please, tell us what you think the political reasons are behind this being done??

If the local population saw more of the economic benefits, was taken more seriously when they complained about earth quake risk, and was compensated more readily when the earth quakes resulted in actual damage, then maybe they would have been more willing to allow the gas fields to exist as a strategic reserve.

As it stands the local population has little more to show for it than demolished homes and damaged monuments and having an economic disease named after them.

I agree. The Dutch government still has a pretty sinister way of using divide and conquer against parts of society (NOx, Groningen / gas, Tata, ‘Toeslagen’; further back: closing of the mines, Indonesia). Our government finds it very hard to govern in compliance with the rules it made for themselves. Perhaps the nature of a complex system and also a general democratic phenomenon. But there is something typically Dutch in just not being able to fix a problem in the way the administration works.

The status page is the last one to get the update. Reddit, HN, X, company chat are all always reporting it sooner.

Some of these SV "founder mode" people do speak like sociopaths. It's probably a good base for success in the industry.

"I don't get it, why are they confessing? They are not confessing. They are bragging"

They speak that way because they are.