Over time autonomous cars will not only be much safer than human-driven cars, but will also be networked to communicate and coordinate with each other in ways no human ever could, further increasing both safety and efficiency.

I can see how that could lead to human drivers being banned from public roads at some distant point in the future, but I'd expect that well before that, manually driving your car will become a hobby rather than a practical necessity, and this hobby will become more niche and expensive over time, kind of like vintage cars today.

I have no idea how far we are from all that, but it will be a while as at the moment autonomous cars are not able to drive everywhere, but only in well-mapped, geo-fenced areas under fairly good conditions, but already today in those areas they are safer than humans based on the data we have so far https://arstechnica.com/cars/2.... So maybe rather than human-driven cars being banned everywhere, they'd be banned in specific places, like highways, or in the densely populated and trafficked centers of specific cities.

At this point, we have data that shows that Waymo cars are safer than human drivers, based on comparison against human drivers driving in the same areas where Waymo drives: https://arstechnica.com/cars/2...

Probably Waymo can't handle Rome or Delhi right now, but it's already saving lives in the environments where it does operate.

That's an anecdote.

The statistics are that Waymo is about 5x safer than human drivers, and this is consistent whether you measure number of collisions with injuries, with property damage, the number of insurance claims against it, the number of airbag activations etc https://arstechnica.com/cars/2...

BTW, I rode in a couple waymo's in SF some months ago, and my own anecdotal data point is that it drove very well, somewhat defensively, and felt safer than many drivers I've ridden with.

In other words, ChatGPT and Grok demonstrated sycophantic behavior by trying to confirm your biases, as no doubt apparent in the way you formulated the question, while DeepSeek and Gemini tried to give a more balanced answer.

One reason I didn't buy a Tesla but a different electric car (this was before Musk's recent forays into politics) is that midway through the 2021 model year, Tesla removed radars from the cars, and said they'd add back the automated emergency breaking that radar enabled in a future software update. Yeah right. This was notionally because Musk believes that you shouldn't need Lidar or Radar for self-driving, because humans can drive without them. That's a moronic argument (is like saying that planes should flap their wings because that's what birds do), but anyhow the real reason is more likely that they were having trouble sourcing chips for the radar due to pandemic-era shortages, and decided to cut corners.

More broadly, I have zero confidence in Tesla's reckless "auto-pilot" and "full self driving" features that are actually level 2 autonomous driving and have led to a lot of deaths already. I'm not against self-driving cars, I've ridden in a Waymo, but the Ubers and Teslas of the world cannot be trusted to safely develop this kind of feature.

Actually, EVs have much lower maintenance costs than combustion engine cars. It stands to reason, because they are so much simpler! When we take in our EV for periodic maintenance (every 10000 miles), all they usually need to do is check the tires and put in the wiper fluid.

I own a Mach E, and confidently drive it to and from an airport 100+ miles away on a single charge to pick up visitors... total range is very close in practice to the advertised 300 miles. What kind of daily commute does your coworker have??

When you use a tool as part of your research that leads to a paper, it's good practice to include your tool in the citations for the paper. There's a gray area of course (you're probably not going to include a citation for latex and vim in every single paper you write with them), but if you use a specific piece of software written by other researchers in your field to do your calculations, you would generally include a citation.

Open source packages from academia often come with a request to cite a specific paper when using the package in research. Years ago, my dad even corresponded with some free software foundation folks, to see if that requirement could be included in a software license. The answer was that it was not legally possible (at least under US law), because licenses can only restrict things that are covered by copyright law, and this is not one of them.

Whatever the story behind this may be, it may be good context to know that Xiaofeng Wang is, by at least some metrics of academic success, the top researcher in his field in the world: https://nebelwelt.net/pubstats...

I've successfully used AI (Gemini, which is what my employer uses, integrated into the IDE) to:
- write a python script to heuristically parse a messy format that was output in some logs so I could troubleshoot a customer issue
- write some basic bash scripts to avoid doing some little boring task by hand
- help me write some python unit tests
- finish up some small refactoring of some code

It works sometimes more than others, but it will get better quickly. This is different from vibes coding, it's just another tool to make a software engineer a little bit faster at doing their job.

Then there is a use-case for vibes coding, of which I've seen examples on personal projects of some friends, which is using AI with something like Cursor to quickly build some small application (e.g., a phone app or website) that addresses a use-case you have, that you wouldn't be able to do or wouldn't want to invest the time to build otherwise. You end up with something you don't understand the details of, that will be hard to maintain in the future, but it lets you do something you just wouldn't otherwise do. This will be important because it empowers people without certain specialized skills to address some of their use-cases by themselves. This is not going to replace software engineering either, to me it feels more similar to how complicated Excel formulas and macros can be used by people who are not really programmers to solve certain classes of problems by themselves. Yes many people have excel spreadsheet that make a software engineer want to run away screaming in horror, but they get the job done.

Welcome to Enterprise fizzbuzz: https://github.com/EnterpriseQ...
This is what we call the "The Princess is in Another Castle" school of object oriented design!

IANAL, but in US courts of law at least, just copying files from disk to memory or from one part of a network to another counts as copying. So training is not copying but it definitely involves copying the data, because you can't get those bits into the model's training data without some copying.

What nonsense. Math is as essential a skill a reading: being innumerate is as bad in life as being illiterate, and a school system that isn't able to teach math to pretty much every student is failing horribly.

What is even this "traditional" approach that supposedly has "proven effective over the centuries?". Math teaching does not change fast but it certainly has not stayed the same "over the centuries", or across countries for that matter. So this "traditional approach" is that just whatever they happened to do in the good old days when you were in school: this is just another thinly disguised "why are they changing things?". Don't know what approach that was, but in many countries (including the US and, clearly from this paper, India, the "traditional" approach to math was not very successful, and left many students feeling they were "bad at math" instead of unblocking mathematical thinking for them.

I'm not a teacher, but I've spent a few hundred hours in recent years in elementary and middle school math classrooms, and my experience is that most children can do math at a much higher level than they're given credit for if they're given the right kind of teaching and attention. Being "good at math" isn't innate: I'm not denying there's a genetic component (as there is to almost everything in human nature), but to a large extent it's a skill that can be unlocked, when math starts to "click" for you.

So let's drop the condescension about most people being too dumb for math. Common Core may not be perfect but it's definitely a step in the right direction for math education because it focuses on having students understand why a solution is in a certain way instead of mindlessly memorizing procedures (is that your "traditional way"?), which doesn't work for any practical purpose (because those memorized procedures will be forgotten as soon as students have passed the test).

Don't give the supreme court ideas!

Exactly this. Most internal corporate communication that asks employees to do something somewhere online is done so poorly as to be indistinguishable from phishing.