This article (regrettably substack) really explores all the nooks and crannies of cringe hiding in the musk chess post: https://davekarpf.substack.com/p/what-elon-musks-favorite-game-tells

You’re not just confident that asking chatGPT to explain it’s inner workings works exactly like a --verbose flag, you’re so sure that’s what happening that it apparently does not occur to you to explain why you think the output is not just more plausible text prediction based on its training weights with no particular insight into the chatGPT black box.

Is this confidence from an intimate knowledge of how LLMs work, or because the output you saw from doing this looks really really plausible? Try and give an explanation without projecting agency onto the LLM, as you did with “explain carefully why it rejects”

These videos are, of course, suspiciously cut to avoid showing all the times it completely fucked up, and still shows the engine completely fucking up.

• “This door requires a blue key” stays on screen forever
• the walls randomly get bullet damage for no reason
• the imp teleports around, getting lost in the warehouse brown
• the level geometry fucks up and morphs
• it has no idea how to apply damage floors
• enemies resurrect randomly because how do you train the model to know about arch-viles and/or Nightmare difficulty
• finally: it seems like they cannot die because I bet it was trained on demos of successful runs of levels and not the player dying.

The training data was definitely stolen from https://dsdarchive.com/, right?

it’s interesting that the only real “hallucination” I can see in the video pops up when the player shoots an enemy, which results in some blurry feedback animations

Well, good news for the author, it’s time for him to replay doom because it’s clearly been too long.

The Cybertruck guy using nostr for image hosting is the kind of brand consistency you can expect from this crowd.

I might be one of the few who had the reverse revelation, learning about her terrible politics, and then marveling at the amazing technical accomplishments when I read further. Is that a new different way to have a terrible time online? Discovering that Racist Duck also made milkshakes.

Urbit might be finished/usable in 100 years

In the comments someone asks the dev Jurij “What is the killer application or feature that I must have?” and he replies “No one knows.”

It probably does deserve its own post, I’m not sure I can do it justice. “Urbit is an attempt to make computers great again”, the phrase “from scratch” appears 7 times, “Urbit is doing God’s work to make the perfect computer”, [in a linked tweet] “Urbit fulfills prophecy”, “It’s [Urbit] a massive virtue signal”.

I missed the comments section at first. “This is one of the things Urbit got right. Especially useful to pique people’s interest and keep them invested by giving them a unique, cool identity.” Jurij claiming that Urbit is amazing because…usernames?

And then there’s the link to Plunder, which the NixOS-heads will have to investigate to figure out if it does what it says:

“Plunder is a new programming model where programs run forever. Hardware restarts are invisible to the software, as is moving a running program from one physical machine to another.”

Finally, Haywire has such lovely readers as *squints* “Eva Brawn”.

Rachel Haywire

Now that’s a rabbit-hole. As far as I can tell, she ran (runs?) a facebook group called “Humanity 2.0”. She “leaves neoreaction” every few years and is back within months. In 2019 she violated the by-laws of the Transhumanist Party elections by mentioning “Jeffrey Epstein or the history of Humanity+”. What they mean is that mentioning that Epstein donated $100,000 to Humanity+ in 2011 “hurts the public image of transhumanism”.

She apparently ran for Humanity+ President in 2013, lost(?), trash-talked Natasha Vita-More, who withdrew H+ support from Haywire since? It’s not clear, a lot of links are dead/scrubbed since then.

http://transhumanist-party.org/tag/rachel-haywire/
https://www.facebook.com/groups/transhumanistparty/posts/2389157701305002/
https://hpluspedia.org/wiki/Rachel_Haywire
https://snoo.habedieeh.re/r/Transhuman/comments/139x2s/natasha_vitamore_vs_rachel_haywire/ (alternative reddit frontend)

Bonus: Ye Gods, her interview with an Urbit developer is really, truly, amazingly awful: https://archive.is/5I7Td

Why would you restrict your criticism to within borders that you regard as illegitimate? I think the best way to not endorse territorial sovereignty would be to ignore those borders as much as possible.

Almost 900 words to say, poorly, “It was just a joke (but not really)”. The short tweets feel like they’re designed for traction, but of course EY falls back on ponderous beigeness when he gets pushback.

Holy shit, that shitshow in the Elon thread.

Possible, yes. Hard? Probably also yes. There’s a lot of addons for Firefox but I don’t use gestures so I have no idea.

Yes, not only is Vivaldi “just” patches on the Google chromium code, they are implicitly limited in what they can patch because they need to be able to port their patches to newer versions of Chromium in order to inherit all of the security features and continued compatibility with the chrome web store. The design and features of Vivaldi are necessarily heavily dictated by Google’s decisions.

Del declined to comment about whether the Carlin-sounding voice was generated by A.I.

I bet they used an AI trained on Carlin’s work to create this special, but Lugosi v. Universal Pictures, IMO (IANAL) means Carlin’s family will likely lose a suit based on imitating his likeness. Good thing I’m not a lawyer because apparently there’s several laws now, starting with the “California Celebrities Rights Act” meaning likeness rights are inherited and good for 70 years.

Did they even consider the possibility of a pedestrian being dragged / run over when designing their software?

From the report (pp 83 of the appendix), it seems like there’s no camera to monitor the undercarriage: it detected it was part of an accident and tried to find a side of the road to stop at, but then further detected something fucky (technical term) with one of the wheels so it just stopped. But at no point did it directly detect a whole human underneath it. It looks like it took ~4.5s to decide to stop after travelling ~20ft at 7.7 mph.

The report is a grim but fascinating read.

“Also talked to [a CHP official] for about 30 minutes this evening. He said they felt like they were ‘punked’ but also got him to a place where he said I am still your advocate, I believe in this technology, both of us could have done better here and we should pay more attention and @Matthew Wood’s verbal walk through as he did today is as helpful as it is to see the video. TLDR - we got to a good place - and he apolgized [sic] but felt like he needed to say what the team was thinking. It’s a good lesson to us that they rely on us a lot more than we anticipate and the trust that we’ve built us great but it’s fragile just as all relationships. We rehabilitated the dynamic I believe.”

CHP = California Highway Patrol. Complete farce of a system.

Yes, it’s normal to hire independent investigations, and I’m not sure why Verge worded it this way (sarcasm?), the actual 200 page report contains a lot of condemnation of Cruise…but inconsistently. The report does seem to massage the situation in favor of Cruise in several places.

I’ve only read about 30 pages of the report, but I don’t see it anywhere consider whether the connectivity issues were intentional, despite Director Matt Woods streaming the video from his home computer (!?) in three consecutive meetings without trying to fix the issue or have someone else play it (it was available to everyone via Slack, including Cruise execs who were in-person at the meetings).

In fact, on page 54 it says he just paused it before the dragging happened: “Cruise expected and intended its October 3 meeting with NHTSA to follow its past practices in which Cruise employees would show a video of an accident or incident and respond to regulators’ questions based upon what they observed in the video. But this did not happen for at least two reasons. There were internet connectivity issues and the Director of System Integrity Wood paused the video at the point of impact, and then never resumed playing it.”

Despite simply pausing the video at the moment of impact, the summary on page 13 says: “Virtual meeting with NHTSA representatives. Wood shows Full Video, again having internet connectivity issues causing video to freeze and/or black-out in key places including after initial impact.”

Page 61-63 shows that there’s debate whether they even showed the DMV the full 45s video or a 12s one that cuts before the dragging, with Cruise employees recalling both possibilities. Quinn Emanuel even hired an engineering consulting company to do forensics on Wood’s home PC, which concluded: “It is not possible to verify [which video he showed]”.

Yet, the summary of that meeting: “Cruise holds hybrid in-person and virtual meeting with DMV and CHP representatives. Wood shows Full Video”

These summaries seems extremely charitable to the point of inaccuracy, to me!

EDIT: extremely tangential but I find it funny that Quinn Emanuel founder John Quinn’s wikipedia proudly announces that he has a podcast where he has interviewed, among others, NYC Mayor Eric Adams and Depp’s lawyer Camille Vasquez.

Oh shit, I guess the rationalists are starting to figure out that sneerclub moved!

Every sneer is one transphobic act suppressed in fear.

https://nitter.net/koksvik/status/1741698756301791600 (original on twitter)

Mr PhD in philosophy here thinks it’s not Singer’s problem whether his partners tell their partners of affairs with Singer.

On this personal site, https://www.koksvik.net/, I find this:

Papers Under Review

‘Benefiting from Epistemic Injustice’

Hmm.