What are you talking about? Do you have any idea what the carbon emissions of fossil fuel extraction and refining are? Do you think the sludge that comes out of the ground goes right into a gas tank? I have 2 EVs, and our home is powered from a nuclear plant. I can absolutely guarantee that my pollution footprint is a tiny fraction of yours. I'll make you a deal: we'll both step into our respective garages, close the doors and seal them, put the cars on lifts, and run them at a leisurely 35MPH for 6 hours. I think you'll find your garage environment will be heavenly, while mine will be more earthly.

I know it's a typo, but I like ORC much more than OCR. Can we re-arrange the words Optical Character Recognition as Optical Recognition of Characters with a silent "of"? Wait a sec, hold that thought, with a silent "by" we can do ORCA, Optical Recognition of Characters by AI! No? Anyone? Bueller? Bueller? Bueller?

Your words, not mine.

I don't know the Sam personally, but I have no doubt he's a genius-level marketer, just not a genius-level technologist. The guy sells visions better than Nostradamus did. You don't raise $64B at a $500B valuation on $18B in revenue while losing a multiple of that unless you are a certified guru-level Jedi master storyteller. The guy definitely warps reality with his narratives, and I give him a well-earned Marketing Genius stamp for that. So extremely smart, but not in an Albert Einstein way, more in a P.T. Barnum way.

I agree with you, LLMs are not "intelligent", they mimic intelligence, and that's a fundamental limitation of autoregressive generation. The more interesting approach to me are diffusion models, particularly video diffusion models. I don't have a Nobel prize, but to me, those exhibit patterns that I'd be more willing to associate with AGI, at least from the perspective of understanding causality and the world it's operating in. The problem is diffusion models are very slow and don't play the quantization game nicely like text does, so nobody is pushing to commercialize them for text purposes.

Math-adept version:

Thank you for pointing out the obvious for the math-challenged. It's only free if electricity is free, and electricity is currently not free. Even if you are given the machine and maintenance for free, it still costs at least $7.50/gallon to produce. Using solar? OK, are they giving solar panels for free too? Nothing adds up, this is a party trick looking for a buyer.

Math-challenged version:

You can totally hook this up in a loop to perpetually fill a gasoline generator and generate endless free excess gasoline! All you need is scale, just like Sam Altman! So order a few dozen, and you can open your own gas station with zero fuel expenses, literally cannot fail. In an act of good faith, I'll buy your now useless solar panels off you for $0.05 per panel, since pennies are dead, and we need to record something for a sales price. You're welcome!

Given it’s well known that AI is like magic beans that you toss out the window and magic happens overnight, are they sure that they yelled "Use more AI" loud enough?

If they've yelled as loud as possible, then we need to apply some deductive reasoning. It's a fact that AI requires no investment in training, and we know it doesn't need specific integrations to leverage it, so the only possible conclusion is that it's the employees’ fault. The solution is to fire them all, and hire a single AI-native who demonstrates proficiency at discussing bondage with ChatGPT. That AI-native mastery will lead to outsized profits, virtually guaranteed.

No sane CTO would pick Windows as the foundation for autonomous agents. "Our outdated and insecure OS with declining market share that uses a non-UNIX file system and runs on non-ARM hardware is the best choice for running autonomous agents!" Where do I sign up?

"ProPublica investigation this year that exposed how Microsoft used China-based engineers to service the Defense Department's computer systems for nearly a decade"

MS let foreign employees service DoD systems? I can't even begin to fathom how this is even remotely possible. Is there a CCP mule leading services at MS? If not, there should be a congressional hearing on this, because this level of incompetence is really inexcusable.

> When I was doing Assembly on the 6809 in the 1980s, I wrote a framework that contained everything I used in most of my programs.

Come one now, no, you didn’t. You maybe wrote a "library", but not a framework. Assembly doesn't have frameworks. Heck, Assembly doesn't even have libraries in the modern sense. Even then, I doubt you linked binary object or archive files. You likely had some .s files you copied in, just like I do to this day.

You're right to call out that not ALL frameworks are bloated, leak, and vulnerability hell spawn, but it's probably safe to say that well under 5% of public frameworks deserve a lean and efficient label.

400kb, really? What are they doing with that post-strip "Hello World"? I completely agree that dependencies are a scourge, but then again I lump Cargo into that pile.

Healthcare professionals as a group, including nurses, comprise 4.2% of the H1B allocations. I don't disagree with you about the issue in healthcare, but nurses on H1Bs are a rounding error.

Since we're stepping back in time, I’m going there. Imagine a Beowulf cluster of tape readers! Ignoring if a cluster of tape readers even makes sense.

I don't have anything against JS, it's fine for manipulating webpages client side because it's the only standard option (fine, WASM, but realistically). Where I draw (drew?) the line is "let's run it on the server too!". That was a terrible idea. Then someone decided, hey, don't code that, use an npm that you have never read or validated. That was the terrible idea amongst terrible ideas. Now they've reached the apex of idea terribleness by letting LLMs write code that is never read, using npms that are never validated, on a server without a sandbox. I'm eagerly awaiting to see how they top this, might I suggest running the whole thing on Node using root?

> The program will primarily recruit early-career software engineers and data scientists.

What is the implied definition of "early" in "early career" here? Limited experience with a specific technology? Limited experience with a specific software stack? Limited experience working for the government? Is the "early" in "early career" subjective or objective? I'm going to go out on a limb and say it's objective, since admitting it's subjective would have literally all the age discrimination lawyers ordering new Porsches. Given "late-career" is generally defined as "the final phase of a person's working life", I would be more measured with my choice of words rather than try to argue that "early career", the polar opposite of "late career", does not implicitly include any age component.