There are things an LLM can show you that are undeniably correct, like: this line of code here calls a “free” on a pointer which might be NULL, and in-fact will be NULL if you follow this path through the code: …

Think of it like “NP hard problems” - there are problems where the solution is hard to find, but easy to verify once you are given it.

When an LLM is giving you those hard to find, easy to veryify observations, that’s value. It doesn’t have to be perfect, it doesn’t have to be 100% complete.

Or, you can hire a team of engineers to burn their brains for months on end to maybe find the same things, maybe not.

There’s a problem with both human attention spans, and LLMs’ context window capacity - neither are up to the task of reviewing a full code base for something like a browser and “finding all the flaws” - but, if the LLM can give you flaws that humans haven’t been able to find… you should be taking those wins - before somebody else does and puts them to different uses.

It’s a different approach, you don’t abandon best practices, but this new tool does give information that was previously more difficult / costly to access - so use it too.

Nobody, and no LLM, knows everything. The LLMs know some things: https://arstechnica.com/information-technology/2026/05/mozilla-says-271-vulnerabilities-found-by-mythos-have-almost-no-false-positives/

Best to take the best information available from all sources. The attackers are also doing this.

You know what helps? After you’ve coded something that works - whether “vibe coding” or the old fashioned way, review it for security issues. “Vibe code reviews” performed by the same LLM tools that do “vibe coding” can be even more effective at finding issues than traditional methods.

But, just like real people, if you don’t bother to care about security, you’ll have holes.

It must be the Presidents of the United States…

Oversupply needs diversion of that supply to other uses - not destruction of established production.

It takes years to establish a productive fruit tree. If the land is really more valuable producing plums or apricots instead of peaches, then, sure, migrate it over slowly. Wholesale destruction with the assumption “the land will be put to better use” is the kind of bullshit that (all too rarely) gets laws and regulations passed to stop it.

If there are too many peaches for local markets, export. If the whole world is drowning in whole peaches, juice 'em.

Yeah, that phone in my shirt pocket set to record really gets noticed… by exactly nobody.

So, yeah, pulling the e-brake hard on the highway can be… exciting, which is generally not what you want in an emergency situation.

This was more of a case of: welp, I’m 10 miles from home and I have a choice: pull over and arrange for a tow truck, or proceed with all due caution on the safest possible routes and get it home without wasting many hours of my time and hundreds of my dollars on the tow.

Now, when the fuel line got chewed by squirrels and a gasoline spray-fountain was emerging from the wheel well… yeah, towtruck time. But bad brakes? Depends on the situation, many situations can be safely handled with the “performance level” you get from cable brakes on the rear wheels.

Oh, one tip should you ever try using the parking brake to stop while rolling: make sure you know how to release it and keep the ability to release it engaged whenever applying the brakes while moving. If you let it latch up, you’re gonna be a passenger not a driver.

700hp of stopping power per wheel isn’t regen braking, that’s dumping battery power into a stopping force.

I have driven home more than once using the cable brake backup after a hydraulic failure.

I also have owned vehicles where the heat-based pads and rotors system overheated and severely lost braking ability after a single stop from 70mph.

4GB is now a Nano model?

There’s plenty of negatives to any new tech, anything can be carelessly or ignorantly mis-applied.

The computer has been coming for our jobs since it was created. Bob Cratchit no longer works for Ebeneezer Scrooge, he’s been replaced with software.

People over-trusting software has been problematic since software became accessible to be over-trusted. A favorite (horrible) example from not-so-long ago, but pre-ChatGPT release I believe: https://www.amnesty.org/en/latest/news/2021/10/xenophobic-machines-dutch-child-benefit-scandal/

For the past year+ it has been popular sport to ask AI a question and poke fun at how wrong the answer is. I, too, get plenty of wrong answers from it - and anyone who trusts what it, or a Google search, or some post by some random troll with an axe to grind on some social media site, or even your high school whatever teacher, without verifying the results… gets what they deserve, in my opinion.

What changed for me within the last 12-16 months is: at least around questions in software development, the answers started being correct more than half the time. That was a critical watershed, because in essence that means that if you give your AI the tool to test its own work, it can work on hard problems that have easy methods to test for correctness (starting with compiler errors), and basically chip away at them - fixing problems until it has an answer that is correct enough to pass all the tests you have specified for it. Before that, an AI agent left to work on problems without guidance would more often get stuck in loops, or run off the rails altogether and never reach a viable solution.

In the past 6 months or so, tools like Claude have gotten much better - incorporating a lot of the kinds of things I (and many others) had to “tell them” manually 12 months ago to get good results into their normal response algorithms, anticipating and fixing problems in their work before presenting it as a solution for your consideration.

The language they present solutions in has been traditionally too over-confident, that’s a huge fault which I attribute to being trained on blog posts by know-it-all blowhard people who similarly present their ideas as gospel truth rather than their potentially flawed best efforts.

Clue for the clueless: even the best human experts in their fields are still only providing potentially flawed best effort answers. Once you leave self-defined fields like mathematics, all we have are our best guesses about how things really work.

On the “different rules for adults and students” thing… if the adults model responsible cell phone use, i.e. never in the classrooms or hallways during school hours, never “ducking out” to their car or the teachers’ lounge just for B.S. doom scrolling or un-necessary calls, IMO that would be much stronger than just banning phones on-prem for kids and adults alike.

The real key: you should control your cell phone, it should not control you - same thing as so many other addiction problems. And, there will be addicts who genuinely are incapable of controlling it, and cold turkey tee-total zero usage has been shown to be the most effective answer for them - just like alcoholism, not drinking is nothing to be ashamed of, having a problem and drinking anyway is much much worse.

As far as “cheating” goes, ever since I got out of the game of paying a bunch of academics to judge and label me, I have been actively encouraged to “cheat” by the people who pay me money… that’s real life.

If you’re using a Ginsu knife to knead dough, you might not have optimal results. Claude is pretty good at code, since about 4-6 months ago. Grok? last time I asked Grok for anything it was the fastest LLM on the market, and the most non-sensical - usless trash.

And you have calulators.

And Google search has been spotty since the beginning.

And Wikipedia article quality … varies.

Like people, if you give AI a sufficiently complex problem, it won’t get it 100% right on the first pass. But, if you give it enough detail to distinguish an acceptable solution from an unacceptable one, it might get 80% of what you’re looking for on the first pass, boost that to 96% on the 2nd pass, 99% on the 3rd pass, and eventually what’s left is simple enough that it finally does get it 100% right.

Anybody who accepts the first thing AI tells them with today’s tech, is using it wrong.