Hell, they don’t even have radar anymore, despite even a lot of low end cars having that.

Technically cost savings, but it seems mostly about stubborn insistence on cameras being enough.

And because a friend insisted that it writes code just fine.

It’s so weird, I feel like I’m being gaslit from all over the place. People talking about “vibe coding” to generate thousands of lines of code without ever having to actually read any of it and swearing it can work fine.

I’ve repeatedly given LLMs a shot and always the experience is very similar. If I don’t know how to do it, neither does it, but it will spit out code confidently, hallucinating function names or REST urls as needed to fit the narrative that would have been convenient. If I can’t spot the logic issue with some code that isn’t acting correct, it will also fail to generate useful text that would describe the problem.

If the query is within reach of copy/paste of the top stack overflow answer, then it can generate the code. The nature of LLM integration with IDEs makes the workflow easier to pull in than stack overflow answers, but you need to be vigilant as it’s impossible to tell a viable result from junk, as both are presented with equal confidence and certainty. It can also do a better job of spotting issues within things like key values that are strings with typo than traditional code analysis, and by extension errors in less structured languages like Javascript and Python (where ‘everything is a hash/dictionary’ design prevails).

So far I can’t say I’ve seen improvements, I see how it could be seen as valuable, but the resulting babysitting carries a cost that has been more annoying than the theoretical time saves. Maybe for more boilerplate tasks, but generally speaking those are highly wrapped by libraries already, and when I have to create significant volume of code, it’s because there’s no library and if there’s no library, it’s niche enough that the LLMs can’t generate either.

I think the most credible time save was a report of refreshing an old codebase that used a lot of deprecated function and changing most of the calls to the new method without explicit human intervention. Better than tools like ‘2to3’ for python, but still not magical either.

Yeah, it does some tricks, some of them even useful, but the investment is not for the demonstrated capability or realistic extrapolation of that, it is for the sort of product like OpenAI is promising equivalent to a full time research assistant for 20k a month. Which is way more expensive than an actual research assistant, but that’s not stopping them from making the pitch.

Already happened in my work. People swearing an API call exists because an LLM hallucinated it. Even as the people who wrote the backend tells them it does not exist

True, though if we are talking about tax bracket going over 30 percent, that would be at nearly 200k, so well above those thresholds too. Of course the numbers aren’t 28 and 33, but that is the closest threshold to the example.

If getting specific, there’s no 28 percent or 33 percent bracket, so these are all examples rather than real figures. I did make a comment using real numbers, same general magnitude but just more specific about the brackets.

But your tax bill doesn’t go up 5%.

Ok, let’s get this close to real numbers. The cited tax brackets don’t exist, so I’ll go with the 24% to 32%. So if your earnings are 1 dollar into the 32% tax bracket, you are going from AGI $191,950 to $191,951. Your tax bill at $191,950 would be:

$11,600 * 0.10 +
$35,550 * 0.12 +
$53,375 * 0.22 +
$91,425 * 0.24 
---------------------------------
$39,110.74

And your tax bill at $191,951 would be:

$11,600 * 0.10 +
$35,550 * 0.12 +
$53,375 * 0.22 +
$91,425 * 0.24 +
$1 * 0.32
--------------------------------------
$39,111.06

Your tax bill goes up by a whopping $0.32 or 0.01% by earning that extra dollar, meaning you still got to keep $0.68 of that dollar. When they say that dollar would cause their tax bill to go up a lot, that’s pretty much exclusively owing to the misconception that people assume their tax bill would have gone to $61,424, so in the misconception that dollar would have cost them $22,313.

Would have to be mandated by workplace regulations, no company is going to voluntarily educate their employees that more money has no downside.

I’ll also say this doesn’t help, it strangely avoids the actual numbers. It should state explicitly that his total taxes would be $1,600+$4,266+$2,827=$8692, and not $13200. Needs to include the scenarios specific results and contrasted with what the viewer would have assumed otherwise.

The whole notion of “kicked up a tax bracket” is also a misleading thing. Only a piece of your income goes into the “new bracket”, all pay under the new bracket is taxed as they would have been used to.

Most of the likely credits tend to phase out gracefully. So it’s true that we can’t be certain, based on my experience of when people are afraid of making too much money, it’s almost always because they think a higher tax bracket applies flatly across their income not due to nuanced understanding of tax credit and welfare benefits.

This all boils down to a common misconception about ‘tax brackets’.

To simplify, pretend there’s a 28% tax bracket up to 100,000 dollars, and a 33% tax bracket when you hit 100k. The first 100k is always taxed at 28%, no matter what you make, and it’s only the incremental amount that gets taxed heavier. So here in this example, that would mean tax burden would be 28,000.33 instead of 28,000.28. These are not the exact brackets or percentages, but it’s at least showing the right magnitude of increase versus total amount.

However, many people are “afraid” of bumping a higher tax bracket. They think the tax bill would go from 28,000.28 to 33,000.33. That the tax bracket bumps up all your liability. I remember growing up people saying “I have to watch out and not hit the bigger tax bracket, if I’m close then I need a big raise to make it worth it, or else the raise is going to cost me more than it would make me”. This a big driver of antipathy toward democrat tax policies, a belief that mild success will punish them, despite it only increasing on the incremental amount.

The LLM doesn’t have to innately implement filtering. You can use a more traditional and concrete filtering strategy on top. So you sneak something problematic by in the prompt and it’s too clever to be caught by the input filter, but then on the output the filter can catch that the prompt tricked the LLM into generating something undesired. Another comment specified they tried this and it started to work but then suddenly it seemingly shut out the reply in the middle, presumably the minute the LLM spit something at a more traditional filter and that shut it down.

I think I’ve seen this sort of approach has been applied to largely mask embarassing answers that become memes, or to detect input known not to work, and to shut it down or redirect it to a better facility (e.g. redirecting math to wolfram alpha).

Hate to break it to you, but the 3.5" form factor would absolutely not be cheaper than an equivalent bunch of E1.S or M.2 drives. The price is not inflated due to the form factor, it’s driven primarily by the cost of the NAND chips, and you’d just need more of them to take advantage of bigger area. To take advantage of the thickness of the form factor, it would need to be a multi-board solution. Also, there’d be a thermal problem, since thermal characteristics of a 3.5" application are not designed with the thermal load of that much SSD.

Add to that that 3.5" are currently maybe 24gb SAS connectors at best, which means that such a hypothetical product would be severely crippled by the interconnect. Throughput wise, talking about over 30 fold slower in theory than an equivalent volume of E1.S drives. Which is bad enough, but SAS has a single relatively shallow queue while an NVME target has thousands of deep queues befitting NAND randam access behavior. So a product has to redesign to vaguely handle that sort of product, and if you do that, you might as well do EDSFF. No one would buy something more expensive than the equivalent capacity in E1.S drives that performs only as well as the SAS connector allows,

The EDSFF defined 4 general form factors, the E1.S which is roughly M.2 sized, and then E1.L, which is over a foot long and would be the absolute most data per cubic volume. And E3.S and E3.L, which wants to be more 2.5"-like. As far as I’ve seen, the market only really wants E1.S despite the bigger form factors, so I tihnk the market has shown that 3.5" wouldn’t have takers.

Another interesting thing to consider.

To be clear, he is rich. But he’s not crazy crazy rich, like nowhere near billionaire status.

With that in mind, his kernel is a key component of RedHat’s, SuSE’s and Canonical whole business, with at least two of those being multi billion dollar businesses.

His kernel is a key component of Android phones, which represent over 50 billion a year in hardware spend, and a bunch of software money on top of that.

His kernel is foundational to most hosting/cloud services with just mind blowing billions of revenue quarterly.

It’s used in almost every embedded device on the planet, networking gear, set top boxes, thermostats, televisions, just nearly everything.

People with a fraction of that sort of relevance are billionaires several times over. A number of billionaires owe much of their success to him. Yet he is not among their numbers.

Now there’s more to things than just a kernel to be sure, but across the hundreds of billions of dollars made while running Linux, there was probably plenty of room for him to carve out a few billion for himself were he that sort of person, but he cares about the work more than gaming the dollars. I have a great deal of respect for that.

Means that while he may not always be right, but I at least believe his assessments are sincere and not trying to drive some grift or cover some insecurity about being left behind.