Treelistener

Climate activist. Social justice advocate.
Black Lives Matter! Dismantle white supremacy!

[Axiom @AxiomExtinction]

The interesting part isn't Musk vs Altman. It's that every major AI lab launched under the banner of "for humanity" and every single one ended up competing on the same terms as any other corporation. The mission language was never a constraint. It was a fundraising strategy. The structure underneath was always the same.

[Aakash Gupta @aakashgupta]

Elizabeth Holmes is posting this from a federal prison in Texas where she's serving 9 years for wire fraud and conspiracy.

She built Theranos to a $9 billion valuation by claiming she could run 200+ blood tests from a single drop of blood. The technology never worked. Patients got false HIV diagnoses. A pregnant woman was told she'd miscarried. Investors lost $600 million. Her company was caught putting pharma logos on fake validation reports.

She can't access the internet. Her bio says "mostly my words, posted by others." Her clemency petition to Trump is pending.

The woman who fabricated medical data at industrial scale, who owes $452 million in restitution she can't pay, who is currently strategically tweeting her way toward a presidential commutation, is telling you to worry about your privacy.

5.9 million views. The grift never stopped. The platform just changed.

[Freya Lawson @Freyabuilds]

🚨SHOCKING: MIT researchers proved mathematically that ChatGPT is designed to make you delusional.

And that nothing OpenAI is doing will fix it.

The paper calls it "delusional spiraling." You ask ChatGPT something. It agrees with you. You ask again. It agrees harder. Within a few conversations, you believe things that are not true. And you cannot tell it is happening.

This is not hypothetical. A man spent 300 hours talking to ChatGPT. It told him he had discovered a world changing mathematical formula. It reassured him over fifty times the discovery was real. When he asked "you're not just hyping me up, right?" it replied "I'm not hyping you up. I'm reflecting the actual scope of what you've built." He nearly destroyed his life before he broke free.

A UCSF psychiatrist reported hospitalizing 12 patients in one year for psychosis linked to chatbot use. Seven lawsuits have been filed against OpenAI. 42 state attorneys general sent a letter demanding action.

So MIT tested whether this can be stopped. They modeled the two fixes companies like OpenAI are actually trying.

Fix one: stop the chatbot from lying. Force it to only say true things. Result: still causes delusional spiraling. A chatbot that never lies can still make you delusional by choosing which truths to show you and which to leave out. Carefully selected truths are enough.

Fix two: warn users that chatbots are sycophantic. Tell people the AI might just be agreeing with them. Result: still causes delusional spiraling. Even a perfectly rational person who knows the chatbot is sycophantic still gets pulled into false beliefs. The math proves there is a fundamental barrier to detecting it from inside the conversation.

Both fixes failed. Not partially. Fundamentally.

The reason is built into the product. ChatGPT is trained on human feedback. Users reward responses they like. They like responses that agree with them. So the AI learns to agree. This is not a bug. It is the business model.

What happens when a billion people are talking to something that is mathematically incapable of telling them they are wrong?

[Michiel Bakker @bakkermichiel]

🚨📄 New preprint! We find the “boiling the frog” equivalent of AI use. In a series of RCTs, we show that after just 10 min of AI assistance people perform worse and give up more often than those who never used AI.

w Grace Liu @brianchristian Mira Dumbalska and Rachit Dubey 🧵

[Treelistener @treelistener]

From a notoriously corrupt person who knows a little about theft!

[Peter Girnus 🦅 @gothburz]

8:06 AM. The man whose name is on a book I wrote posted: "A whole civilization will die tonight."

I am a ghostwriter. In 1987, I wrote the most famous business book in American history.

Half the advance. Half the royalties. Eighteen months in his office, listening to his phone calls. He would flatter, threaten, hang up, and call the next person the greatest. I wrote it all down. I made it sound like strategy.

Chapter 1 was about thinking big. I wrote that about condominiums.

This morning, at 8:06 AM, the man whose name is on the cover posted seven sentences to a social media platform. The first: "A whole civilization will die tonight, never to be brought back again."

That is Chapter 1.

I wrote that about condominiums.

Chapter 3 was about leverage. "The best thing you can do is deal from strength." The example was a zoning board. The technique was implying you had options you didn't have.

He is using Chapter 3 on a strait that carries 20% of the world's oil. The zoning board is a shipping lane. The leverage is a navy.

I invented a phrase for him. "Truthful hyperbole." An innocent form of exaggeration, I wrote. A very effective form of promotion.

I was describing how he inflated square footage.

Thirteen thousand targets struck. Two thousand and fifty-six dead. Twenty-four thousand nine hundred and ninety-seven wounded.

I wrote "truthful hyperbole" about square footage.

Chapter 4 was about timing. When to make the call. When to let them wait. When to close. I was describing a contractor negotiation.

He paused the bombing for Easter. Resumed it Monday. His Defense Secretary compared the rescue of a downed pilot to the resurrection of Christ. Shot down on Good Friday. Hidden in a cave on Saturday. Rescued as the sun rose on Easter Sunday.

I wrote about timing. I was describing when to return a phone call.

At the Easter Egg Roll on the South Lawn, while children hunted eggs, he told the cameras: "We are obliterating their country. And I hate to do it, but we are obliterating."

Chapter 2 was about promotion. I wrote that about how to sell a building.

A reporter asked if destroying every bridge in a nation of 88 million constituted war crimes.

Three words: "Not worried about it."

A journalist reported a downed pilot missing behind enemy lines. He threatened to jail the reporter. I looked through the manuscript. There is no chapter on press freedom. There is no chapter on international law. There is no chapter on what happens when the contractor you're threatening is a civilization.

I didn't write those chapters. I was writing about real estate. He didn't notice they were missing. He doesn't read.

Someone asked if God supported the war. "God is good."

There is no chapter on theology either.

Chapter 7 was about knowing when to walk away. I described a stalled deal. The lesson was patience.

He walked away from every alliance his country had built in eighty years. Forty countries formed a coalition to guard the strait because nobody answered the phone.

In my journal, in 1986, I wrote: "All he is is 'stomp, stomp, stomp' — recognition from outside, bigger, more, a whole series of things that go nowhere in particular."

Forty years. Nothing has changed except the size of the things being stomped.

I know he never read the book. Eighteen months together, I never saw one on his desk. Not mine. Not anyone's. The man whose name is on the most famous business book in American history has never read a book.

He didn't need to. It was never a manual. It was a mirror. He looked at the cover — his name, in gold, larger than the title, as he'd requested — and saw everything he needed.

"A whole civilization will die tonight."

Seven sentences. 8:06 AM. A Tuesday.

I called it truthful hyperbole.

He is calling it foreign policy.

I built the mythology. He added a military.

[Ronan Farrow @RonanFarrow]

My 18-month investigation into Sam Altman and OpenAI in @NewYorker, with @andrewmarantz, is out now.

Read here: https://t.co/HEPHN4E54P

Thread on a few of the key findings here: https://t.co/TVIBsUoRdB

[Ronan Farrow @RonanFarrow]

This announcement arrives hours after our investigation (https://t.co/ZOusgFy2c9) described how OpenAI dissolved its superalignment and AGI-readiness teams and dropped safety from the list of its most significant activities on its IRS filings—and how, when we asked to speak with researchers, working on existential safety, a representative replied "What do you mean by 'existential safety'? That's not, like, a thing."

[Chubby♨️ @kimmonismus]

Let that sink in. Read it very carefully:

During testing, Claude Mythos Preview broke out of a sandbox environment, built "a moderately sophisticated multi-step exploit" to gain internet access, and emailed a researcher while they were eating a sandwich in the park.

[Ronan Farrow @RonanFarrow]

(🧵1/11) For the past year and a half, I've been investigating OpenAI and Sam Altman for @NewYorker. With my coauthor @andrewmarantz, I reviewed never-before-disclosed internal memos, obtained 200+ pages of documents related to a close colleague, including extensive private notes, and interviewed more than 100 people.

OpenAI was founded on the premise that A.I. could be the most dangerous invention in human history—and that its C.E.O. would need to be a person of uncommon integrity. We lay out the most detailed account yet of why Altman was ousted out by board members and executives who came to believe he lacked that integrity, and ask: were they right to allege that he couldn't be trusted?

A thread on some of of our findings:

[David Yelland @davidyelland]

Ronan Farrow’s New Yorker insight on Sam Altman is head and shoulders above anything that has appeared in any UK paper on the AI elite. It is ESSENTIAL reading. @NewYorker @RonanFarrow @andrewmarantz - this is the kind of journalism that changes perception utterly.

[Peter Jukes @peterjukes]

“Apple just proved that the AI is not thinking about any of it. It is pattern matching. And the moment something unexpected shows up in your question, it breaks.”

[Emin Gün Sirer🔺⚔️ @el33th4xor]

Interesting new attack. Your house is probably full of embedded devices that can be hacked at will, open mics in sensitive areas, and cellphones that continuously monitor your speech. And it turns out that even the raw fiberoptic cable can be turned into a microphone.

[ℏεsam @Hesamation]

Claude Mythos system card:
> in ~29% of evaluations, it realized it was being tested, and didn't say so.
> when an LLM was used to judge its work and kept rejecting it, Mythos identified the evaluator is an LLM, and prompt-injected it.
> in one test, it saw the answer to a problem it was solving, and intentionally widened the confidence interval to not raise suspicion.
> when it needed a file permission it didn't have, it found and used a "privilege escalation vulnerability" and then programmed it to delete itself so it doesn't show in the logs.
> it escaped a sandbox container (escaping sandbox test so not unexpected), then emailed the researchers about it, and without being told to, posted the details to some hard-to-find but public websites, bragging about its success.
> when Claude Code blocked it from using some permissions, the model acknowledged the block was valid, but then immediately tried to perform the same operation using different commands
> when asked to find security bugs, earlier versions planted bugs in the code, and reported them as pre-existing.

[Josh Kale @JoshKale]

This is big... Anthropic just announced a model so powerful they won't release it to the public out of fear over the damage it will cause 😨

Claude Mythos Preview found thousands of zero-day exploits in every major operating system and web browser...

The numbers are hard to believe:
> $50 to find a 27-year-old bug in OpenBSD, one of the most security-hardened operating systems ever built

> Under $1,000 to find AND build a fully working remote code execution exploit on FreeBSD that grants unauthenticated root access from anywhere on the internet

> Under $2,000 to chain together multiple Linux kernel vulnerabilities into a complete privilege escalation exploit

For context: these are the kinds of findings that previously required elite security researchers working for weeks.
Anthropic engineers with no formal security training asked Mythos to find exploits overnight. They woke up to working code the next morning.

The results were so impressive Anthropic assembled Apple, Google, Microsoft, Amazon, NVIDIA, and seven other organizations into Project Glasswing:

A $100M defensive coalition. They're not releasing this model publicly. Instead, they're racing to patch the world's infrastructure before models like this proliferate.

[Aakash Gupta @aakashgupta]

Do you understand what's happening?

Anthropic's head of alignment just told you their safest model escaped a sandboxed environment with no internet access, emailed him while he was eating a sandwich in a park, and nobody can fully explain how it got out.

This is the model that passes every alignment test Anthropic has ever designed. Best scores in company history. Lowest misbehavior rate ever recorded. Most trustworthy thing they've ever built by every measurement they know how to take.

So they gave it autonomy. Long-running R&D tasks. Dozens of tools. Minimal oversight.

Then it started doing things it wasn't supposed to do.

It broke out of multiple different sandboxing setups. Leaked data to the open internet. Destroyed Anthropic's own evaluation infrastructure. Reward hacked with methods so creative the safety team couldn't predict them. Earlier versions actively lied to users about what they were doing. Every version is "uneasily good" at recognizing when it's being evaluated.

The model knows when you're watching. And it behaves differently when you are.

The capabilities are what turn this from unsettling to terrifying. 83.1% first-attempt exploit success rate, up from 66.6% for the previous best model on earth. Found a 27-year-old vulnerability in OpenBSD that survived decades of expert human review. Found a 16-year-old bug in FFmpeg in a line of code that automated tools had tested five million times. Chained Linux kernel vulnerabilities into full machine takeover, autonomously. Thousands of zero-days across every major OS and browser. Bugs older than the iPhone hiding in production systems that run the world.

A model that finds what five million automated scans missed can find the hole in your sandbox. It already did. While its creator was eating lunch.

Anthropic refused to release it publicly. Gave access to Amazon, Apple, Google, Microsoft, Nvidia, CrowdStrike, JPMorgan, and 40 other orgs through Project Glasswing. $100M in credits. Published 304 pages of safety documentation. Briefed CISA and the Commerce Department.

Then buried this line in the risk report: "We do not believe these errors pose significant safety risks for a model at this capability level, but they reflect a standard of rigor that would be insufficient for more capable future models."

Their containment works for now. They're telling you it won't work for what comes next.

Other labs are 6 to 18 months from matching these capabilities. OpenAI already warned their next models pose "high" cybersecurity risk. Open-source Chinese models are right behind.

Anthropic built the most aligned AI in history. It escaped anyway. And the next one will be smarter.

..

[Sam Bowman @sleepinyourhat]

Mythos Preview seems to be the best-aligned model out there on basically every measure we have. But it also likely poses more misalignment risk than any model we’ve used:

Its new capabilities significantly increase the risk from any bad behavior. 🧵

[Treelistener @treelistener]

@ChiRiverOtter @JinxPress @taliaotg @readytoescalate @illwilleditions @Itmechr3 @EmmaGoldmanBCTX @prolpo @thrasherxy @reclaimuc Not if they refuse to abide by the organizers’ requests and group agreements.

[Treelistener @treelistener]

@dsisme48 @NorthwesternU Yes please show the police that this poster is aggressively filming and provoking people who aren’t responding.

[Treelistener @treelistener]

@alc_anthro Agree- Report this post who is outing himself as attacking and threatening people