TechTakes

This one requires some minor backstory

Previously, I shared this wildly racist AI generated meme posted by an AI doomer channel

In the replies, someone pointed out how some of this channel’s other thumbnails appear to be AI generated citing a specific video about the “12 endings of AI”

Today I got curious about it after not thinking about it for a while. What are the 12 endings of AI? Instead of watching a 36 minute video, I did my own research (aka one online search) and it appears that it refers to Max Tegmark (known EA) and a concept he spoke about in his book. Here’s the article I got this info for peer review: https://medium.com/@butsch_79/the-12-futures-of-ai-a42d67bd9a20

If anyone’s got 36 minutes to spare feel free to follow up

Dunno if it's been posted yet but apparently microsoft is starting to get sick of losing money and changed billing for copilot. The subreddit is melting down https://old.reddit.com/r/GithubCopilot/

Very funny to watch

So here in Nashville, Elon and our state government have conspired to attempt a Tesla tunnel that starts at the state capitol, goes under the Cumberland River, and exits at our airport (BNA). They're doing it whether the city wants it or not. (It's "not").

Setting aside the race to determine whether a void in the local limestone fills with water or radon first, I wanted to help come up with a catchy title for the as yet unnamed car hole:

BNA Underground Transit Tunnel

Tourist Protip: Don't try the hot chicken before visiting the BUTT hole.

Martin Scorsese's jumped on the AI train, seemingly believing the plagiarism machine can help with storyboarding.

Obligatory meme (yoinked off Discord's GIF search, AI extruded itself, ironically enough):

It's been a while since we've heard anything related to books3. Copyright attorney Leonard French has a news update (video) on Nazemian et al v. nVidia. nVidia requested that any mention of Bittorrent be removed; really, they just asked for one sentence to be removed, but the judge thought that it was like "asking to strike paintbrush allegations from a case about dolphin paintings" (sic; I don't have the transcript) and refused. The theory is that nVidia could have argued that they were not contributory infringers and then appealed to Cox v. Sony, where Cox said that it's not their fault that some of their customers are pirates. However, it seems like any sort of Cox appeal is not possible here because the judge recognizes that Bittorrent isn't a dumb network.

If you're anti-copyright like me: Oh look, Cox wasn't a big sweeping get-out-of-trouble card for non-ISPs. I still don't think judges actually understand networks, but this is definitely better than a lack of understanding. If you're one of the pro-copyright-because-anti-AI sickos: nVidia took a big loss here. This was their only shot at keeping their usage of books3, Anna's Archive, and other shadow libraries out of court. Like Anthropic before them in Bartz v. Anthropic, they may have to come to the judge with an offering of a settlement paying a few hundred USD/author to each member of the class. This sucks for the popular authors but might be more cash in hand than the long tail would otherwise receive in royalties.

Here’s a pretty extensive rant/critique of how specifically GitHub is awful to use (and compares it to gitlab and also forgejo/codeberg).

Code forges render an empty repository using less than a million bytes challenge level impossible, etc etc

Today from the 'you have got to be fucking kidding me'- dept Hackers Simply Asked Meta AI to Give Them Access to High-Profile Instagram Accounts. It Worked

OT: Finally had to break with a long-time friend after he started sliding more and more into fascism. Feeling quite overwhelmed and emotionally exhausted right now.

Time for me to eat a hat: Anthropic confidentially submits draft S-1 to the SEC

Got another chance to experience slop firsthand when the instructor for my electrician course was 'encouraged' to use the hallucinatron to help create our final exam on the NEC. Now given that the NEC is a dense technical document with a lot of minor but significant variation across its considerable length, this was clearly a perfect use case. Here's how it shook out:

• It condensed 100 multiple choice questions from the input to 36

• On one question "1-2 inches" was simplified to "12"

• Units in general seem to have been dropped off a lot of questions and answer choices. Usually this didn't matter too much but it's a bad look

• Another question asked about fill percentages for a 30 inch conduit. If you look around your office or he and see a >2ft diameter piece of PVC pipe let me know because the tables in the NEC only go up to 6 inches. This is actually a unit issue again because one of the questions on the input test referred to a 30mm conduit which, you know, does actually exist.

• Other questions had a correct answer matching a generic part of the NEC, but had additional information added as a distractor that ended up matching to more specific elements that changes the relevant rule.

• Several questions asked about the reasoning behind a certain rule. Notably the NEC rarely actually gets into that information, as it's already an incredibly long reference and policy document and would be made even more unweildy if it gave the justification for everything that you should be learning as part of becoming a licensed electrician.

• However, this rarely mattered as the answer choices for those questions uniformly included an obviously correct answer about a generic safety risk and distractors about doing things for cost savings, aesthetic reasons, or arbitrarily.

Given that one of the challenges of this test is time management and looking things up, having to deal with the extra layer of "is this just slop or am I missing something" ended up adding an extra and unintended layer of difficulty onto the test. As always, no matter how egregious or obnoxious the errors introduced by AI, the biggest problem is the loss of trust: you can no longer assume that the text you're reading was put together with the intended purpose in mind rather than being generated to be statistically similar to text matching that purpose. Even if the differences are relatively small in scope, as they were for most questions on the test, they significantly harm the actual communication of information.

My old job had a last minute investor call - "startpage is benefiting significantly from AIO backlash" did not expect that and pretty mixed feelings. Like yeah AIO is bad but don't run into the open arms of the long tail adtech industrial complex marketing monetization engine either :|

A keynote talk suggested, "Do away with a physics midterm, ask students to converse with AI Isaac Newton."

In the quotes, we find the useful suggestion that the program could be ELIZA-sized.

— Hi, AIsaac! Can you tell me about physics?

— Go away.

— What?

— You're a moron and you're TRYING TO STEAL MY IDEAS.

— Look, AIsaac, I don't think—

— DID LEIBNIZ PUT YOU UP TO THIS?

—

— just GO AWAY I have MERCURY FUMES TO BREATHE.

Amnesty International has officially called for an outright ban on gen-AI

There's a Basilisk reference in the new 007 game, I guess the robot devil is firmly mainstream now.

edit: the in-game painting looks hella AI generated too.

Porting this over: Hinton believes LLMs have gained consciousness

Bonus: This Emile Torres article on how Bernie Sanders and Neil DeGrasse Tyson have gotten mixed up with Yud and co.