wrs

I’m a CMU SCS alum (actually I predate SCS being a full school). Please, someone make me feel better about this. They’re renaming the school after this person.

This conundrum (and many others) is why insurance is a highly regulated business. The commission working in the public interest decides how the compromise is drawn, not the insurers.

Mermaid is open source, no? Can we just add these placement constraints to Mermaid instead of making a whole new language?

They already bought the product, and it came with a license, then they wanted to unilaterally change the license. When they couldn’t, they declared the product a supply chain risk. The idea is you declare it a risk before you buy it.

Not to mention, they simultaneously said it was a supply chain risk and that it would continue to be used by agencies that were already using it in a critical capacity. Which makes absolutely no sense.

Or they could just not buy the product in the first place. The supply chain risk part is pure retribution.

But if they don’t jump at the chance to retroactively rewrite the already-signed license, you’re OK with designating them a supply-chain risk?

That’s going to make it rather risky for anyone to sign licenses with the DoD. In fact, it renders the licenses pointless since you just have to do anything the licensee says.

So in your mind there’s absolutely no subtextual message here in choosing wolves. It would have meant exactly the same thing if he’d chosen raccoons, or pigeons, or deer, or rabbits?

I'm talking about data not able to leave the building, not even data getting lost on the way.

One example was a site that had fiber and coax, from different companies. They might have shared a pipe at some point along their length, hard to say. But both connections went down at the same time for digital reasons, not physical. The providers had simultaneous unrelated backend router problems and the site lost contact to both gateways.

This was with a nice SD-WAN system that routed everything dynamically across both pipes to address latency and errors, but if the packets get dropped at the first hop in both networks, there's not much you can do!

The saving grace in that case was the third redundant connection, a 4G cell modem, with a lot less bandwidth but able to keep the critical transactions going. These days I'd certainly want Starlink on the roof as well.

> What has to go is the paper as the unit of achievement, the thing that gets counted and compared. If the ecosystem is the unit of research, then building something that other people can build on has to count for more than the next increment.

The focus on building stuff as the measure of accomplishment is one reason I so enjoyed being at CMU a few decades ago, so I'm very happy to see this sentiment is still expressed by the new generation of faculty.

What is the giant blinkenlight panel on the right?

I’ve had enough double-redundant links fail - and I mean pretty thoroughly double-redundant, different provider, different media, different physical path - that I’m quite surprised something like the air traffic control system only has two links.

Based on the examples, It’s like the thing took a writing class from Claude, but isn’t quite as smart — kind of worst of both worlds. The “Interactive Reporting” one is particularly terrible. A huge amount of waffly padding around a thesis that may or may not exist. So if the goal is to make long fancy reports that nobody will read, then, yes, very cost-effective.

“the question it raises matters more than the answer” ???

“So ‘the river drifts from cool to warm’ is not a figure of speech.” Oh really?

As someone who’s had to refresh electronics a few times over the decades, I remember this course as being particularly good, for a couple of reasons.

The whole thing is organized around building explicit abstractions. You’re always making models, and choosing what to pay attention to and what to ignore. By making the real circuit match the model closely enough, you get to ignore the circuit and use it as building block for bigger things. This same approach is used from resistors to amplifiers to logic gates to CPUs.

Of course this how engineering actually works, but I haven’t seen an intro course presented this way so rigorously. If you’re a software person who deals with black box abstractions all day, it feels like home.

The course introduces concepts in order of simplicity, meaning you get MOSFETs and digital gates before you get BJTs. Many courses are arranged chronologically, so after you do Ohm’s law and capacitors you get dumped into BJTs, which are one of the most complicated things in the discrete toolkit, and hardly ever found in common use nowadays (except as cheap switches). Instead, you’re happily making amplifiers, switches, and logic with FETs, a much gentler introduction.

I also remember somewhere in the course lectures the professor says (paraphrased) “You may be thinking, why do I need to learn how a logic family works? Won’t I just be using some existing chip? This is MIT — you’re supposed to be the one making the chip.”

>[Nadella said] if he “had been made aware that OpenAI had scraped and trained on information that was behind a paywall,” he would have “invoked [Microsoft’s right to] require OpenAI to retrain its models.”

Whew, good thing it’s too late to be accountable for that now, huh? Water under the bridge. Mistakes were made. Eggs, omelets.

Isn’t the “error in the equation x = 0” just…x?

It says they can manually search them. According to this, the ruling explicitly doesn’t say if they can digitally search them.

This would be more convincing if there were a simple stat on the home page of the dispositions of the reported bugs (fixed/no-repro/won't-fix).

As a random sample of one, I looked at one of the bugs this reported on Tailscale (first thing on the homepage) [0], and the pull request ends with "Apologies for the lack of due diligence here. I'll go ahead and close this out."

[0] https://github.com/tailscale/tailscale/pull/17843

That’s what I thought too! And now I kind of want to see that.

Oh, well, after rotating the board, sure, I get that.

Do we really know that? Or have we only seen the sequences from people searching for Numberwang? Has a number been both Numberwang (or Wangernumb) and not Numberwang (or Wangernumb)?

> If you know proposition X is true, it’s easier to figure out why.

This misses the point entirely. If you know proposition X is true or false, you won’t bother spending years trying to prove or disprove it, developing deep understanding and potentially even developing entire new fields of math in the process. (See FLT.)

That’s why it’s so destructive to the discovery process to have an LLM just generate a proof or counterexample without the side effect of generating useful explanatory or generative structures that we can build on.

That said, there are examples like Ramanujan where someone did an info dump of unexplained theorems that people try to mine useful structures from, but that’s not at all the mainstream of mathematical progress.

The author argues for evaluating Ph.D. candidates based more on the oral thesis defense than on the actual thesis.

By essentially the same reasoning, I’ve been arguing for prioritizing in-person design/code reviews over code-only async PR comments.

The important thing is to verify that the human has a coherent design in mind and can demonstrate that it got implemented, regardless of who or what was at the keyboard. “I dunno, I guess Claude thought this was a good idea” is not a coherent design.

I keep the studio computer OS a full version behind.

My impression from audio developers is that the audio infrastructure is very badly documented (like so much of MacOS nowadays) and using it properly requires a combination of luck, WWDC videos, and devrel threads. Also, Apple introduces changes that are genuine improvements in theory, but break undocumented assumptions.

The lack of rigor is a bit surprising given that audio is probably the only mission-critical hard-real-time thing MacOS does. People routinely run an entire live concert on a Mac and an audio glitch is not OK! (Yes, if it’s Taylor Swift budget then there are at least two machines in sync with a big red button to switch over…)

But given what they’re dealing with, I understand why it’s more involved than just “the developers didn’t try the betas”.

>They took actions that would be considered as crimes if a human took them

Um, hang on, if you meant that to be taken literally then we have a major problem. If you want to do something criminal, you just need to ask ChatGPT to do it for you?

I’m still not at all clear on why OpenAI shouldn’t be facing CFAA charges over this.

I’m a bit baffled by how one could learn math without doing the homework. In my experience the homework is how you learn the math! So the exams would essentially already be testing whether you did the homework.

I don’t quite understand why the author says Context doesn’t count here. I’ve certainly experienced a top-down version where because a function starts accepting a Context, and therefore needs to be cancelable, lots of functions it calls now need to take a Context too, and check for cancellation. Perhaps the color here is cancelability, not the Context per se. (Which may be even worse, as a kind of “invisible color”.)

We’re talking about the UI here, not the binary executable. Finder was extensively reworked in 2001 from the file manager in NeXT to be much closer to the original Finder, and most of the complaints here are about UI decisions taken from the original Finder, released in 1984. That predates Windows (not just Windows Explorer, which is from ten years later), Nautilus (created by Andy Hertzfeld of the original Mac team to be like Finder!), and the Unix variants, even the venerable xfm, which dates from the early 1990s.

The PARC interface (by which I guess you mean the Xerox Star) had significant differences, and I think would confuse any modern computer user far worse than the Finder. The Lisa was closer to the Star but the Mac used a simplified model that just equated files to icons.

Source: I lived through this process and used all of these for actual work at some point. The Finder’s influence on GUI file management was rather like the iPhone’s influence on phone management.

“Weird for the sake of being weird”…you know Finder existed first, right?

Aaargh! OK, you made me look it up. :) The word is "chordophone" (an instrument that makes sound from vibrating strings). A "fretted" instrument is a chordphone whose pitches are fixed by frets. But I can't find a word for the class I'm talking about, "a chordophone with a fingerboard", as opposed to a piano, which is a chordophone with one string per pitch (and is also a percussion instrument!).