Keep in mind this is a system with millions of miles under it’s belt and it still doesn’t understand what to do with a forced left turn lane in a very short trip in a fairly controlled environment with supremely good visual, road, and traffic conditions. LIDAR wouldn’t have helped the car here, there was no “whoops, confusining visibility”, it just completely screwed up and ignored the road markings.

It’s been in this state for years now, of being surprisingly capable, yet horrible screw ups being noted frequently. They seem to be like 95% of the way there and stuck, with no progress in reality just some willfull denial convincing them to move forward anyway.

That safety driver did not give a single fuck about driving on the wrong side of the road…

Which brings up an interesting question, when is a driverless car ‘parked’ vs. ‘stopped’?

Navigation issue / hesitation

The video really understates the level of fuck up that the car did there…

And the guy sitting there just casually being ok with the car ignoring the forced left going straight into oncoming lanes and flipping the steering wheel all over the place because it has no idea what the hell just happened… I would not be just chilling there…

Of course, I wouldn’t have gotten in this car in the first place, and I know they cherry picked some hard core Tesla fans to be allowed to ride at all…

The thing that strikes me about both this story and the thing you posted is that the people in the Tesla seem to be like “this is fine” as the car does some pretty terrible stuff.

In that one, Tesla failing to honor a forced left turn instead opting to go straight into oncoming lanes and waggle about causing things to honk at them, the human just sits there without trying to intervene. Meanwhile they describe it as “navigation issue/hesitation” which really understates what happened there.

The train one didn’t come with video, but I can’t imagine just letting my car turn itself onto tracks and going 40 feet without thinking.

My Ford even thinks about going too close to another lane and I’m intervening even if it was really going to be no big deal. I can’t imagine this level of “oh well”.

Tesla drivers/riders are really nuts…

Time to install to OneDrive.

I’m making a note here: Huge Success

One thing left unclear is how the determination is made about emergency versus non emergency.

If it’s a separate number, ok, seems clear cut enough.

If it’s human always answers and if it’s some bullshit they just click a button to punt to AI instead of just hanging up, ok.

If they are saying the AI answers and does the triage and hands off immediately to a human when “emergency detected”, then I could see how that promise could fail.

To be fair they made a lot of strides to the point where config file wrangling went from mandatory to almost never done.

But yes, Nvidia would have quirks driving people back to wrangling config file, but they got better too.

Though I’m not particularly interested in X11. The biggest thing they had was trivial application forwarding, but the architecture didn’t scale well to modern resolutions and UI design that was largely bitmaps being pushed, as well as not handling higher latency networks too well.

I’d say that those details that vary tend not to vary within a language and ecosystem, so a fairly dumb correlative relationship is enough to generally be fine. There’s no way to use logic to infer that it’s obvious that in language X you need to do mylist.join(string) but in language Y you need to do string.join(mylist), but it’s super easy to recognize tokens that suggest those things and a correlation to the vocabulary that matches the context.

Rinse and repeat for things like do I need to specify type and what is the vocabulary for the best type for a numeric value, This variable that makes sense is missing a declaration, does this look to actually be a new distinct variable or just a typo of one that was declared.

But again, I’m thinking mostly in what kind of sort of can work, my experience personally is that it’s wrong so often as to be annoying and get in the way of more traditional completion behaviors that play it safe, though with less help particularly for languages like python or javascript.

Fine, a chess engine that is capable of running with affordable even for the time 1970s electronics will best what marketing folks would have you think is an arbitrarily capable “reasoning” model running on top of the line 2025 hardware.

You can split hairs about “well actually, the 2600 is hardware and a chess engine is the software” but everyone gets the point.

As to assertions that no one should expect an LLM to be a chess engine, well tell that to the industry that is asserting the LLMs are now “reasoning” and provides a basis to replace most of the labor pool. We need stories like this to calibrate expectations in a way common people can understand…

Oh man, I feel this. A couple of times I’ve had to field questions about some REST API I support and they ask why they get errors when they supply a specific attribute. Now that attribute never existed, not in our code, not in our documentation, we never thought of it. So I say “Well, that attribute is invalid, I’m not sure where you saw to do that”. They get insistent that the code is generated by a very good LLM, so we must be missing something…

To be fair, a decent chunk of coding is stupid boilerplate/minutia that varies environment to environment, language to language, library to library.

So LLM can do some code completion, filling out a bunch of boilerplate that is blatantly obvious, generating the redundant text mandated by certain patterns, and keeping straight details between languages like “does this language want join as a method on a list with a string argument, or vice versa?”

Problem is this can be sometimes more annoying than it’s worth, as miscompletions are annoying.

GPTs which claim to use a stockfish API

Then the actual chess isn’t LLM. If you are going stockfish, then the LLM doesn’t add anything, stockfish is doing everything.

The whole point is the marketing rage is that LLMs can do all kinds of stuff, doubling down on this with the branding of some approaches as “reasoning” models, which are roughly “similar to ‘pre-reasoning’, but forcing use of more tokens on disposable intermediate generation steps”. With this facet of LLM marketing, the promise would be that the LLM can “reason” itself through a chess game without particular enablement. In practice, people trying to feed in gobs of chess data to an LLM end up with an LLM that doesn’t even comply to the rules of the game, let alone provide reasonable competitive responses to an oppone.

There’s not an express purpose of the emissions, the emissions are a side effect of combusting to propel the aircraft.

The injection, release, or dispersion, by any means, of a chemical, a chemical compound, a substance, or an apparatus into the atmosphere within the borders of this state for the express purpose of affecting the temperature, weather, climate, or intensity of sunlight is prohibited.

Key phrase being “for the express purpose…”

My understanding is that these bills would make that illegal.

Pretty sure that the actual components of airline exhaust would be considered legal. They wouldn’t just ban all fossil fuel powered flight, they just want to ban the non-existant mind-control/climate manipulation they think the airplanes are doing.

Without being explicit with well researched material, then the marketing presentation gets to stand largely unopposed.

So this is good even if most experts in the field consider it an obvious result.

Particularly to counter some more baseless marketing assertions about the nature of the technology.

And that’s pretty damn useful, but obnoxious to have expectations wildly set incorrectly.