Oh man, I remember marveling at BeOS in the day and for a brief moment in time when SSDs first hit the scene you could have a credibly fast Windows boot… Nowadays it’s worse than ever despite super fast storage, fastest CPUs, and gobs of RAM…

There was a while back some Windows developer externally lamenting how ass-backwards they were and as a result their NT kernel was woefully under-featured compared to other contemporary OSes…

Then I think they forced him to take it back and say ‘um actually our kernel is actually super awesome, my mistake’.

They did not, they had some touch screen button.

They basically needlessly increased risk for the sake of avoiding optics of a safety driver with direct controls possible.

That’s a good point, also if you can compare like to like conditions and what the data does if you exclude teen drivers. Also if you can identify incidents related to bald tires and brake failures that wouldn’t apply.

Also would be interesting to compare human augmented driving miles to full autonomous miles. With the automated emergency braking/collision alert/lane centering assist. Anecdotally was teaching my teen to drive. Suddenly a car pulls out right in front of us, zero warning. If that happened to me, with experience on a formerly normal car, I’m pretty sure I would’ve wrecked. However my kids reflex to swerve triggered the cars “evasive steering assist” and did an action movie worthy maneuver, avoiding going off into the ditch and returning just right into the lane after getting around the other car.

Thing about autonomous driving is that it seems to get the stupid easy stuff wrong in dangerous ways, but if you have a demanding precise maneuver to make, it has a better chance once that maneuver is needed.

The challenge is one approach only needs to modify the transit infrastructure. The other means having to tear down and build new commercial and residential properties and force people and businesses to relocate in order to have a vaguely sane transit system. My area desperately wanted to do transit but even with rather significant hypothetical funding, they could only service about 10-15% of typical trips. They’ve settled on a plan that is much less money, but only serves like 5% of trips. To go with that plan, they are making restrictions around zoning to force mid density mixed use construction only, favoring one of the two chosen transit corridors.

They are trying but just people are distributed very awkwardly for mass transit.

Let’s put it this way. If you knew a person, and that person just had their fourth crash in 8 years having driven 160k miles, would you think “this person is a bad driver” or would you think “they only crashed 4 times, let’s see where this goes”.

Especially if you’ve seen this driver drive in the wrong lane, go straight in a turn only lane, and other dodgy maneuvers regularly.

A small sample size would just make the prediction highly uncertain, could be way better or way worse.

However others have made the observation that it’s reasonable to consider the miles driven as the sample size, and at over 300,000 miles, it is a bit more credible sounding.

Funny part was that tesla taxis also had a human attendant, but for the sake of appearance made them sit on the passenger side. They deliberately limited staff from being able to interact with steering and pedals.

They eventually moved those to the driver seat.

Going by inflation adjusted market cap values, it certainly looks like the financial facet of the AI companies alone are bigger than both those events… This is going to be beyond messy…

The issue is that as dumb as it is, SATA ssds are still a big part of the consumer market.

Even though nvme isn’t appreciably more expensive to make, it’s still used as a “premium” product., and SATA is a product tier to capture budget market whole protecting their more premium market.

This move is a clear symptom of the real issue. Manufacturers shifting as much capacity as possible towards big datacenter buildouts at the expense of starving every other market for these products. Trillions of dollars that will pay whatever it takes competing with a more rational market

So our utility came out and said they have to raise residential rates by a rather large amount, largely because so many data centers are demanding so much power they need to upgrade, so residential rates have to fund that…

So assuming 10 lbs of force, as measured 1 meter away from the hinge, you have about 44.5 Nm of torque. Assuming each door opening was about 90 degrees, then you have about 70 Joules per door operating event.

Each door opening would have a physical theoretical max of 0.02 watt-hours.

Assuming you spent 8 hours opening a door every 10 seconds constantly, then you have 58 watt-hours of energy at the end of the day if you had 100% efficient generators. One typical solar panel would hit that in under 15 minutes in real-world energy collection, not theoretical.

So yes, the law says there is some unavoidable, unusable waste heat, the question is how much of that heat is really unusable?

For example, you have lava at around 1,000 degrees. You certainly can harvest energy from that, hit some water with it and spin a turbine.

For the most part, once we get under 100C we run out of ideas on how to realistically harvest energy out of it, but there’s still a pretty good delta between that an ambient. The claim of this article is he has an approach to harvest energy at an even lower temperature delta.

If it got to harvesting all of the temperature delta of a system, then we can say “not at all possible based on current understanding of physics”, but if the process leaves some waste heat unharvested, then it’s not yet violating that law. The law just says it gets less and less likely as the amount of heat in question diminishes.

Well then you need to handle backfeeding all sorts of circuits, which is generally a pain to the extent it works. But it also would barely do anything.

It is scalding hot, but I think the key takeaway is that it’s not hot enough to boil into steam, which is our current go-to for harvesting energy from heat.

So after you do your steam turbine and you are left with not-quite boiling water, by today’s standards it is useless for further harvesting for electricity. If this article is as-advertised (a big if), then we can harvest more, adding efficiency to any process that boils water to turn a turbine.

Hypothetically, any energy harvested from a zero-emission strategy might at least displace combusting some hydrocarbons.

If the code actually works and is vaguely important, I think you are right.

If anyone ever has to fix it because it’s also broken on top of being a mess, well they aren’t quite so safe. Maybe if you are always available to fix it same day, but if you ever go on vacation and it hits the fan while you are unreachable…

I can’t speak to this situation, but broadly speaking I am familiar with general messed up stuff like this as well as perhaps adjusting some fine details to make the scenario relatable to an audience unfamiliar with the specifics of the real situation and/or obfuscating the details so that the person doesn’t out themselves to someone else familiar with the specifics enough to recognize.

The broad strokes seem plausible and any oddities in details I consider to be less important and/or understandable if it was tweaked for an internet audience.

The type of problem in my experience is the biggest source of different results

Ask for something that is consistent with very well trodden territory, and it has a good shot. However if you go off the beaten path, and it really can’t credibly generate code, it generates anyway, making up function names, file paths, rest urls and attributes, and whatever else that would sound good and consistent with the prompt, but no connection to real stuff.

It’s usually not that that it does the wrong thing because it “misunderstood”, it is usually that it producea very appropriate looking code consistent with the request that does not have a link to reality, and there’s no recognition of when it invented non existent thing.

If it’s a fairly milquetoast web UI manipulating a SQL backend, it tends to chew through that more reasonably (though in various results that I’ve tried it screwed up a fundamental security principle, like once I saw it suggest a weird custom certificate validation and disable default validation while transmitting sensitive data before trying to meaningfully execute the custom valiidation.

I’ve been using Claude to mediocre results, so this time I used Gemini 3 because everyone in my company is screaming “this time it works, trust us bro”. Claude has not been working so great for me for my day job either.