drpixie

Or more to the point, it's surprising that humans can make sense of LLM text, quite as much as they do?

Because the sense and the semantics is entirely within the human.

It's disappointing that this article won't rate higher here, as it's spot on and most points very difficult to refute. Disappointing, but given the audience, unsurprising.

Do the non-connected TVs bitch about lack of connection and desperately prompt you to upgrade?

>> because semantics is not involved

And there, in five words, is the problem/feature/definition of AI.

And NEVER use an email address on some domain/service for administration of that domain/service.

Availability of an OpenWrt port is the first thing I check when looking for network hardware :)

I've been self-hosting (using a server at home) using MailInABox ( https://mailinabox.email ) for around 5 years.

I've been very happy with it. I have a perfectly normal email service, over which I have full control. MailInABox handles all the necessary DKIM etc. I had minor hassles, when I first started, with sending to google and microsoft, but these were reasonably easily managed by following microsoft's advice. (And I have a good ISP.)

The only hassle is that, because MailInABox uses greylisting, initial account-validation emails are sometimes delayed. But because I have control of the service, I can whitelist anything important :)

Must be the shortest ever NYT article ... "No."

I don't suppose it includes mindful discussion of how easily humans are fooled... and how desperately some people want to fool themselves.

The really interesting thing about "AI" is it has shown how easily we are fooled, and how willing we are to be fooled.

There's some great research areas opened up by "AI". But it's research into people, not "AI". We should be looking into how human vision works, given that we take obviously generated images to be real. And how we grant intention to text which clearly has no intention.

On Bing!!!! Sorry, in the back of my mind, bing is that very sad MS search that was terrible at everything :) And it gave you something useful - astonishing.

Or, it's impressive that, amongst the millions of hacks it has copied from various chat sites, were some that worked in this case.

The scale of these things is impressive, but the mechanism is not much better than brute force.

Hi author, where did the aircraft and engine data come from?

That data looks to be rather arbitrary. Real aircraft performance is hugely dependent on aircraft weight, altitude, and ambient (air) temperature. And engine performance is very dependent on air temperature and airspeed. So your table of cruise/climb speeds etc might produce interesting results but the results are not going to be at all realistic.

To give you an idea of data complexity, all airliners come with hundreds or thousands of pages of closely printed tables giving performance at specified weight, temperature, altitude, etc.

> if the value is in the prompt, give me that prompt – I have access to a model too.

Absolutely. I don't want to waste my time reading pages of (literally) mindless slop.

If you can't be bothered writing well, just write points. Don't use "AI" to expand your points into slop. I might read your points, and will judge you on your lack of time and/or care. But I will despise your slop.

That is how the PR reads, but is not at all what happened.

A team of highly trained and skilled people used an AI tool, through many many instructions (prompts), to produce a specific mathematical theorem. The tool is impressive, the result (possibly/probably) interesting, but the PR skips the vital role of the humans (for the usual PR reasons).

And be your own critic. Aim for great, not just ok. Find out what dry-joints look like. And too much solder. Then aim for a really nice job :)

Yeah - and if they wanted some cheap PR, that's one way to get it.

Their basic point is very valid, and the term "AI" is now so overused that it's become essentially meaningless.

Recently I've heard "AI" used to mean:

* using large language models for code/text/image generation

* goal-driven search of a problem space

* pattern extraction from images

* algorithmic decision making

* audio pattern recognition

* something (unspecified) using lots of computers

* and more trivially, "doing it on a computer"!

LLMs have shown us how easily humans are fooled. Just provide a similar answer to which was seen in similar circumstances, and we start misleading ourselves into it's method and "understanding".

Though perhaps I'm wrong that people use a model of the world to think through answers to queries. Maybe many people do just repeat something like whatever they heard before. In which case, LLMs make as much sense as the average ignorant human.

It's been amusing to watch the points bounce up and down on my comment.

I guess equally half the readers agree with my sentiment, and half down-vote, being upset by my attitude to the AI industry :)

PS. I'm quietly waiting for the bubble to pop - the main interest being will it pop with a bang and cause grief to many, or will it just go with a long drawn-out fart that can be ignored by most.

> GPT-5.6 Sol Ultra produces proof of the Cycle Double Cover Conjecture

Very misleading article title.

Title should be "Un-named humans produce unverified proof of CDC Conjecture using GPT-5.6" ... but I expect only advertising copy when it comes from the AI industry.

Is there anything on the website that isn't "AI" generated? Oh dear, oh dear, oh dear :(

Can't imagine it would survive the first descent storm. Unlike conventiolnal turbines, it doesn't seem to have any way of protecting itself.

Few thousand? Really? Larousse Gastronomique is 3600 pages, averaging maybe 1 recipe and 1 ingredient per page - and that's just classic French cooking.

Me too. The number of "revolutionary" designs that are announced but disappear makes me cynical. Looking wings on real aircraft, unless freshly painted, they're pretty close to finely sanded :) If the airlines and engineers saw a significant performance degradation with wear, they'd be out there polishing and repainting wings.

On a similar note - How many times have you seen announcements about someones blended wing that is going to save 50% fuel? But there are very few blended wings in nature (eg. rays), and those are in a very slow-speed regime.

I'm inclined towards keeping an ancient android for those apps that require it, and maybe something open for actual use. Or perhaps a crappy old android for android and a small non-android tablet/laptop for daily-driver stuff, which always works better as a computer anyway!

I'm also becoming open to using software that lies to google about what it is :) Google will treat us like sh*t, why shouldn't we reciprocate.

Same reason as "make another (better) windows" is very difficult - almost everyone wants to be able to run existing apps and drivers, so you're forever playing compatibility catchup with android (or windows).

That's the reason companies are desperate to be first/biggest - once you're it, you're it until you finally fall on your face and dwindle to a nobody.

I've always felt that to treat infinity as number is to commit a category error (aka type conflict), to confuse the process with the outcome of the process. Infinity has proven to be very useful, but usefulness doesn't make it always valid.

It's a great pity but I won't be buy more Ubiquiti hardware unless their direction improves. I loved the Ubiquiti hardware until a year or so ago. It was well spec'd, well priced, reliable, and it was easy to upgrade to OpenWrt. (Changing to OpenWrt was definitely an upgrade - get away from their you-must-do-it-our-way software. OpenWrt isn't perfect but mostly just works.)

But Ubiquiti have changed. The new hardware is silly - little screens and do-everything-on-an-app-with-our-cloud firmware. It's my infrastructure, I don't want to rely on your cloud! I only hope they go back to making good hardware and letting us use it for what we want.

>> However, the tech exists for a reason and is not inherently bad, the issue is the lock-in, the lack of choice and interoperability.

The marketing excuse for the tech might be features or efficiency, but the reason for the tech is lock-in and minimising product lifetime.

The days when manufacturers had friendly, cooperative relationships with their customers are long gone :( Can we bring them back? I hope so, but am not hopeful.

>> It has barely hit 50% and it's already plateauing.

Well, the curve has got to level-out at 100%.