prompt injection phish by email

so glad these things have a solid security model and this totally won’t result in a scrambled half-assed fix

Elon makes Grok developers install intrusive surveillance software on their laptops. They're being told to enable screen captures and URL tracking.

So random thought, you now how LW is worried about an seed AI going foom right, how it bootstraps itself into a new AI with better capabilities which bootstraps etc all the way till the singularity.

Why isn't it happening? Gpt-2 was released 2019, 3 2022, 4 2023, so where is 5 and 6? Where is our double event?

This isn't an original thought, but a better matrix for comparing the ideology (such as it is) of the current USG is not Nazi Germany but pre-war US right wing obsessions - anti-FDR and anti-New Deal.

This appears in weird ways, like this throwaway comment regarding the Niihau incident, where two ethnic Japanese inhabitants of Niihau helped a downed Japanese airman immediately after Pearl Harbor.

Imagine if you will, one of the 9/11 hijackers parachuting from the plane before it crashed, asking a random muslim for help, then having that muslim be willing to immediately get himself into a shootouts, commit arson, kidnappings, and misc mayhem.

Then imagine that it was covered in a media environment where the executive branch had been advocating for war for over a decade, and voices which spoke against it were systematically silenced.

(src)

Dude also credits LessOnline with saving his life due to unidentified <<>> shooting up his 'hood when he was there. Charming.

Edit nah he's a neo-Nazi (or at least very concerned about the fate of German PoWs after WW2):

https://www.lesswrong.com/posts/6BBRtduhH3q4kpmAD/against-that-one-rationalist-mashal-about-japanese-fifth?commentId=YMRcfJvcPWbGwRfkJ

I'm going to put a token down and make a prediction: when the bubble pops, the prompt fondlers will go all in on a "stabbed in the back" myth and will repeatedly try to re-inflate the bubble, because we were that close to building robot god and they can't fathom a world where they were wrong.

The only question is who will get the blame.

Penny Arcade chimes in on corporate AI mandates:

https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/

When developers are allowed to use AI tools, they take 19% longer to complete issues—a significant slowdown that goes against developer beliefs and expert forecasts. This gap between perception and reality is striking: developers expected AI to speed them up by 24%, and even after experiencing the slowdown, they still believed AI had sped them up by 20%.

womp, hold on let me finish, womp

Another day, another jailbreak method - a new method called InfoFlood has just been revealed, which involves taking a regular prompt and making it thesaurus-exhaustingly verbose.

In simpler terms, it jailbreaks LLMs by speaking in Business Bro.

https://www.lesswrong.com/posts/JspxcjkvBmye4cW4v/asking-for-a-friend-ai-research-protocols

Multiple people are quietly wondering if their AI systems might be conscious. What's the standard advice to give them?

Touch grass. Touch all the grass.

In the recent days there's been a bunch of posts on LW about how consuming honey is bad because it makes bees sad, and LWers getting all hot and bothered about it. I don't have a stinger in this fight, not least because investigations proved that basically all honey exported from outside the EU is actually just flavored sugar syrup, but I found this complaint kinda funny:

The argument deployed by individuals such as Bentham's Bulldog boils down to: "Yes, the welfare of a single bee is worth 7-15% as much as that of a human. Oh, you wish to disagree with me? You must first read this 4500-word blogpost, and possibly one or two 3000-word follow-up blogposts".

"Of course such underhanded tactics are not present here, in the august forum promoting 10,000 word posts called Sequences!"

https://www.lesswrong.com/posts/tsygLcj3stCk5NniK/you-can-t-objectively-compare-seven-bees-to-one-human

In the morning: we are thrilled to announce this new opportunity for AI in the classroom

In the afternoon:

Someone finally flipped a switch. As of a few minutes ago, Grok is now posting far less often on Hitler, and condemning the Nazis when it does, while claiming that the screenshots people show it of what it's been saying all afternoon are fakes.

Today's bullshit that annoys me: Wikiwand. From what I can tell their grift is that it's just a shitty UI wrapper for Wikipedia that sells your data to who the fuck knows to make money for some Israeli shop. Also they SEO the fuck out of their stupid site so that every time I search for something that has a Finnish wikipedia page, the search results also contain a pointless shittier duplicate result from wikiwand dot com. Has anyone done a deeper investigation into what their deal is or at least some kind of rant I could indulge in for catharsis?

The Gentle Singularity - Sam Altman

This entire blog post is sneerable so I encourage reading it, but the TL;DR is:

We're already in the singularity. Chat-GPT is more powerful than anyone on earth (if you squint). Anyone who uses it has their productivity multiplied drastically, and anyone who doesn't will be out of a job. 10 years from now we'll be in a society where ideas and the execution of those ideas are no longer scarce thanks to LLMs doing most of the work. This will bring about all manner of sci-fi wonders.

Sure makes you wonder why Mr. Altman is so concerned about coddling billionaires if he thinks capitalism as we know it won't exist 10 years from now but hey what do I know.

"Another thing I expect is audiences becoming a lot less receptive towards AI in general - any notion that AI behaves like a human, let alone thinks like one, has been thoroughly undermined by the hallucination-ridden LLMs powering this bubble, and thanks to said bubble’s wide-spread harms […] any notion of AI being value-neutral as a tech/concept has been equally undermined. [As such], I expect any positive depiction of AI is gonna face some backlash, at least for a good while."

Me, two months ago

Well, it appears I've fucking called it - I've recently stumbled across some particularly bizarre discourse on Tumblr recently, reportedly over a highly unsubtle allegory for transmisogynistic violence:

You want my opinion on this small-scale debacle, I've got two thoughts about this:

First, any questions about the line between man and machine have likely been put to bed for a good while. Between AI art's uniquely AI-like sloppiness, and chatbots' uniquely AI-like hallucinations, the LLM bubble has done plenty to delineate the line between man and machine, chiefly to AI's detriment. In particular, creativity has come to be increasingly viewed as exclusively a human trait, with machines capable only of copying what came before.

Second, using robots or AI to allegorise a marginalised group is off the table until at least the next AI spring. As I've already noted, the LLM bubble's undermined any notion that AI systems can act or think like us, and double-tapped any notion of AI being a value-neutral concept. Add in the heavy backlash that's built up against AI, and you've got a cultural zeitgeist that will readily other or villainise whatever robotic characters you put on screen - a zeitgeist that will ensure your AI-based allegory will fail to land without some serious effort on your part.

Love how the most recent post in the AI2027 blog starts with an admonition to please don't do terrorism:

We may only have 2 years left before humanity’s fate is sealed!

Despite the urgency, please do not pursue extreme uncooperative actions. If something seems very bad on common-sense ethical views, don’t do it.

Most of the rest is run of the mill EA type fluff such as here's a list of influential professions and positions you should insinuate yourself in, but failing that you can help immanentize the eschaton by spreading the word and giving us money.

A hackernews muses about vibe coding a chatbot to provide therapy for people in crisis. Soon, an actual health care professional shows up to butcher the offender and defile the corpse. This causes much tut-tutting and consternation among the locals.

https://news.ycombinator.com/item?id=44535197

Edit: a shower thought: have any of yall noticed the way that prompt enjoyers describe using Cursor, tab completions, and such are a repackaging of the psychology of loot boxes? In particular, they share the variable-interval reward schedule that serves as the hook in your typical recreational gambling machines.

A Supabase employee pleads with his software to not leak its SQL database like a parent pleads with a cranky toddler in a toy store.

https://news.ycombinator.com/item?id=44502318

trying to explain why a philosophy background is especially useful for computer scientists now, so i googled "physiognomy ai" and now i hate myself

https://www.physiognomy.ai/

Discover Yourself with Physiognomy.ai

Explore personal insights and self-awareness through the art of face reading, powered by cutting-edge AI technology.

At Physiognomy.ai, we bring together the ancient wisdom of face reading with the power of artificial intelligence to offer personalized insights into your character, strengths, and areas for growth. Our mission is to help you explore the deeper aspects of yourself through a modern lens, combining tradition with cutting-edge technology.

Whether you're seeking personal reflection, self-awareness, or simply curious about the art of physiognomy, our AI-driven analysis provides a unique, objective perspective that helps you better understand your personality and life journey.

Bummer, I wasn't on the invite list to the hottest SF wedding of 2025.

Update your mental models of Claude lads.

Because if the wife stuff isn't true, what else could Claude be lying about? The vending machine business?? The blackmail??? Being bad at Pokemon????

HN commenters are slobbering all over the new Grok. Virtually every commenter bringing up Grok's recent full-tilt Nazism gets flagged into oblivion.

LessWrong's descent into right-wing tradwife territory continues

https://www.lesswrong.com/posts/tdQuoXsbW6LnxYqHx/annapurna-s-shortform?commentId=ueRbTvnB2DJ5fJcdH

Annapurna (member for 5 years, 946 karma):

Why is there so little discussion about the loss of status of stay at home parenting?

First comment is from user Shankar Sivarajan, member for 6 years, 1227 karma

https://www.lesswrong.com/posts/tdQuoXsbW6LnxYqHx/annapurna-s-shortform?commentId=opzGgbqGxHrr8gvxT

Well, you could make it so the only plausible path to career advancement for women beyond, say, receptionist, is the provision of sexual favors. I expect that will lower the status of women in high-level positions sufficiently to elevate stay-at-home motherhood.

[...]

EDIT: From the downvotes, I gather people want magical thinking instead of actual implementable solutions.

Granted, this got a strong disagree from the others and a tut-tut from Habryka, but it's still there as of now and not yeeted into the sun. And rats wonder why people don't want to date them.

Musk objects to the "stochastic parrot" labelling of LLMs. Mostly just the stochastic part.

LW:

Please consider minimizing direct use of AI chatbots (and other text-based AI) in the near-term future, if you can. The reason is very simple: your sanity may be at stake.

Perfect. No notes.

Can’t find the angle to spin this out into a grown-up buttcoin post, but if I did, the title would be “Horse_ebutts”.

Anyway: recently I’ve been burdened with the knowledge that there’s a bunch of horse racing related crypto companies. They’re all obviously terrible.

• Zed Run: a play to earn (P2E) virtual horse NFT racing game. Defunct as of February, probably due to rug pulling, they are pivoting to “Zed Champions”, which is… pretty much the exact same thing, with likely the same fate.
• EquineChain: a blockchain platform for tracking horse care history, because apparently people don’t trust horse caregivers and need GPUs to remember how much ivermectin and ketamine their show-ponies have mainlined.
• BTX Racing: a blockchain platform for buying stake in horses. Not sure if you get to choose which cut of the horse you own. Also, not sure if when you liquidate your equine tranche you get cash or a bucket of glue.

Also, insert obligatory stablecoin reference here.

I have become aware that there is a very right wing Catholic podcaster who has a Catholicism AI chatbot app. It's called Truthly.

Your Trusted Catholic AI Conversation Companion Deepen your understanding, explore ideas, and engage in meaningful dialogue—anytime, anywhere.

If someone could call up Pope Leo and get him to excommunicate the guys who invented this, that would be great.