TechTakes

Go check your privacy settings in Firefox - they've switched on sending data to advertisers by default as of Firefox 128. Even Google sent this setting out default off, probably having spoken to an actual lawyer who mentioned the GDPR.

Over at "work on climate" there's been an influx of companies that will greenwash using ChatGPT. One company I interviewed for (in my estimation it) boiled down to using ChatGPT to make generic greening recommendations for a business and attach hallucinated numbers that the client can then pass off for themselves.

Edit: is there a list of companies that use "prompt engineering" so that I can just avoid them?

found this linked in ed zitron comment section for some reason: https://www.funraniumlabs.com/2024/04/phil-vs-llms/

With a moment’s contemplation after reading it, I just realized how spectacularly bad this could go if, for example, you went to do a search for an chemical’s Material Safety Data Sheet (MSDS) and a Large Language Model (LLM) gave you back some bullshit advice to take in the event of hazmat exposure or fire.

joke's on you, MSDSs are already dogshit. these things only exist to cover ass of manufacturers and are filled with generic, useless advice https://www.science.org/content/blog-post/uselessness-msds https://www.science.org/content/blog-post/un-safety-data-sheets there is MSDS for sand, MSDS for tear gas and ethanol lists the same dangers, toxicity is overemphasized (because it's common) and some other dangers like explosiveness are underappreciated (because it's not), we don't even need LLMs for this, humans (lawyers mostly i guess) did the same on accident

also bonus points for first-principling what could have been instead of asking somebody that actually knows, like any proper rationalist would do. also, vinyl chloride is not reactive with water and spraying pressurized containers with water can be a sensible thing to do, because this cools them down, so it decreases pressure meaning it decreases risk of rupture, which would be a bad thing, if manageable for firefighters to do it safely. see: some fires involving propane tanks

An MSDS may not tell you what respirator to use;

Slander! MSDS will tell you to use the right one ("appropriate respirator"), it's your job to figure out what it is

we've reached some zenith of insane facebook AI slop

hot Japanese cabin crew, barefoot, on a farm, with Jesus

(image)

https://www.nature.com/articles/d41586-024-02218-7

Might be slightly off topic, but interesting result using adversarial strategies against RL trained Go machines.

Quote: Humans able use the adversarial bots’ tactics to beat expert Go AI systems, does it still make sense to call those systems superhuman? “It’s a great question I definitely wrestled with,” Gleave says. “We’ve started saying ‘typically superhuman’.” David Wu, a computer scientist in New York City who first developed KataGo, says strong Go AIs are “superhuman on average” but not “superhuman in the worst cases”.

Me thinks the AI bros jumped the gun a little too early declaring victory on this one.

https://t.co/rBxJwxtqDl

Medium article about Yud came across my dash (I hope the link works).

If Books Could Kill back with Hanania's piece of work https://pod.link/1651876897/episode/afe6b965e80359c1edbb86f78bde3997

opening line "This book's so fucking stupid, I regret choosing it"

https://xcancel.com/kitten_beloved/status/1810709361175691474

In which TPOT unites to yell at scooter because he isn’t down with mental health apartheid schemes.

I've been out-of-the-loop for a bit on the Nix drama. Is there a good summary of the last couple weeks?

So remember when Google Domains got sold off to Squarespace because it wasn't profitable enough and Google has the attention span of a squirrel?

Well that meant bye bye MFA for anyone who didn't check their email diligently enough, allegedly leading to a number of cryptocurrency domains getting hacked.

The cryptocurrency aspect is mostly just funny, but Google and Squarespace should know better than to effectively disable MFA out from under people. Tech companies put profit over people all the time. And then everyone blames the people for not being hyper-vigilant about computer security.

Edit: The tweet linked in that bleepingcomputer article is funny if this was indeed the issue: https://twitter.com/pendle_fi/status/1811683909509558562

Some "defi" company realized this could be a problem 22 hours before they were hacked. Even had time to write a tool to mitigate the impact of getting hacked. Got hacked anyway. Did they uhh... IDK change their password? Make sure MFA was set up? They don't say.

(I’ll try put a decent summary of links on this later)

there’s a UK party that (aiui) committed electoral crimes by submitting non-existing genML-created people as candidates, a whole new usecase!

gonna be real fun to see that catching sunlight, if TNI manages to do due process right

dril goes to the oracle for advice.

Answer: The Philippines.

https://xcancel.com/pic/orig/media%2FGSWu2hwWUAANIpi.jpg

NSFW, Sterling on the weird language of LLMs

https://bruces.medium.com/preliminary-notes-on-the-delvish-dialect-by-bruce-sterling-ce68a476247b

orbstack 1.6.4:

Debug Shell: AI-powered package install suggestions for commands

in the app upgrade popup it's just bare text. in the documentation for debug shell there's no reference. in the release notes feed it's the same bare text

I've already sent feedback asking for more information about it, but just ..... what? I mean there's that annoying(-to-me) ubuntu shell hook that goes "oh hey $binary not found, try installing $pkg!" already, and that's been out for years, but what?

if/when I hear more I'll post comment I guess. in the meantime consider me fucking bewildered.