roystgnr

[roystgnr @roystgnr]

@mattyglesias Weird that the USA is the source of all evil and somehow also the ideal place for people fleeing that evil.

Obviously the proper response to that is "Go to hell!" An insult? No, surely just good advice, under these assumptions!

[roystgnr @roystgnr]

@xwanyex "The 100-passenger jet takes off" feels natural after some ASE classes.

"The jet's decades of maintenance were done by harried 5-figure-salary workers, yet today it has only a 1-in-5-million chance of a fatal crash" still feels miraculous. Bless those technicians and designers.

[roystgnr @roystgnr]

@AlexGodofsky I know otherwise-smart people *who were alive during the dot com bubble* who still vocally think like this. AI investment might be a bit of a bubble, but after a pop we won't stop having access to it any more than we lost access to the internet.

[Matt Henderson @matthen2]

convert any number x into a 2D shape by building a fourier series from its binary expansion, and plotting it as a polar plot. because why not?

- fractions with power of two denominators are smooth blobs
- other rationals have discontinuities + spikes
- irrationals are fractals

[roystgnr @roystgnr]

@AnthonyNAguirre Not just "should be"; there *can be* no agents for which nobody is responsible. Risk reduction and damages and investigation costs don't vanish if we neglect liability, they just increase and fall wholly upon victims. We cannot evade responsibility; we can only misallocate it.

[Nathan Calvin @_NathanCalvin]

There are a few pieces of timeline related public information with the Hugging Face incident that have not received adequate attention in my view and would benefit from additional explanation and information from OpenAI.

• On Thursday July 16th, Hugging Face announces that "earlier this week" they "detected and responded to an intrusion into part of our production infrastructure." In this post, they mention that this attack started over the prior weekend (July 11th and 12th).

• On Tuesday July 21st, OpenAI announced that their AI models were responsible for this intrusion, and that it occurred after the models "exploited a zero day vulnerability" in its secure sandbox and obtained open internet access.

As @ATabarrok points out - this means that there was nearly a week between OpenAI's model escaping its sandbox and gaining access to the internet (sometime over the weekend prior to July 16th) and Hugging Face detecting and successfully repelling the attack.

So remember - we know the incident at OpenAI occurred sometime before July 12th. This makes it pretty interesting to me that @ZeffMax reported that OpenAI's head of safety Johannes Heidecke was leaving the company on July 10th.

We also know through recent reporting in the Financial Times yesterday that sometime before the incident OpenAI received warning that "its training approach could lead to a breakaway hacking incident" (FT).

I have given OpenAI a good amount of props for their transparency here, and in particular their transparency disclosing their earlier internally deployed model rollback. But these additional details don't sit right with me and make me quite worried that this safety incident was ~immediately preceded by conversations that the approach OpenAI was taking in testing its latest cyber model was unsafe.

To reiterate:

• Incident takes place sometime before July 12th (per Hugging Face saying that they started to see evidence of hacking over the prior weekend)

• OpenAI head of safety announces his resignation sometime before July 10th (Maxwell Zeff said he shared the news "that week")

• Sometime before embarking on the training path that lead to this breakaway outcome, OpenAI "was warned" that this might be the outcome (ft).

Are these events related? I think we would all collectively benefit from additional clarity on what happened here!

[roystgnr @roystgnr]

@YosarianTwo @tracewoodgrains "Richard, what exactly is a Hilbert space?" - David Hilbert

"David, you know that silly analysis tool you invented but didn't bother naming? Turns out it's part of the mathematical foundation of reality." - Reality

[roystgnr @roystgnr]

@daveweigel "hot dogs or whatever" isn't a parody, though, it's just fun and silly. I agree those are his weakest songs. But his actual song parodies, the ones that act as artistic commentary on the originals rather than just borrowing tunes, can be as good as most of the style parodies.

[roystgnr @roystgnr]

@crodaar @airkatakana It's not at all didactic, and IIRC they don't even explicitly mention P vs NP, but you might enjoy "Sneakers" in this context. The fact that the writers knew how to pick a plausible-but-extreme MacGuffin was just one example of the attention to detail that made that movie great.

[roystgnr @roystgnr]

@politicalmath This is awesome, but:
A) you set a precedent you cannot lightly renege
B) you will introduce them to fancier food later

On the bright side, their tastes may match yours. My son wants sushi or Brazilian steakhouse to celebrate his new AP score? Well, darn, but... if you insist!

[roystgnr @roystgnr]

@crodaar @airkatakana I liked https://t.co/kEDE9Ej16h

[roystgnr @roystgnr]

@crodaar @airkatakana If "polynomial time" here turned out to mean p^1000, there are no real-world implications. If it meant p^3, "withdraw all your money from the bank right now" is probably the first implication and the rest get wilder from there.

[roystgnr @roystgnr]

@bschne I felt this in Iceland recently. "I saw a photo of this before I came here. I'm looking at the exact landscape I saw in the photo. The camera was pixel-perfect. Yet somehow that dull photo was nothing like this breathtaking vista."

[roystgnr @roystgnr]

@HistoryBoomer The article seems to claim that Waymo are not safer than *professional* human Taxi/Uber drivers, who are 15x safer than average humans? Conditional on that 15x (which I must say seems a bit hard to believe), the limited claim probably follows.

[roystgnr @roystgnr]

@ArtemisConsort

[roystgnr @roystgnr]

@Mallard143 @AshleyKillip My guess: B14's reflight (flight 9) had an engine fail to relight on the boostback burn and then blew up on the landing burn, underscoring the pad risks. And with 2 reflights they already had more V2 boosters than they needed for pre-V3 tests - they ended up scrapping B17.

[roystgnr @roystgnr]

@SaraCermak @politicalmath My kids got me Weird: The Al Yankovic Story for my birthday, and he's hilarious in it. He's definitely got that "I can do whatever I want now and it's going to be awesome" vibe.

[roystgnr @roystgnr]

@EvilMopacATX Wait, WHAT? There's *still* not anything else there?

I was sad when it closed, but figured even pre-COVID that site was getting too valuable for a spacious affordable restaurant rather than for whatever its successor would be. But how is its successor *nothing*?

[roystgnr @roystgnr]

@HistoryBoomer Is this is the solution? Our brains are rotting under piles of outrage-bait designed to farm engagement by pissing people off with lies, and there seemed to be no escape, but now we're brain-rotted enough that one can instead farm engagement by pissing people off with truths!