There is never any justification for throwing a Molotov cocktail at a family’s house. The fact that, after such an event, @sama can write a gracious and reflective post is a testament to his ability to see the bigger picture. Such ability will be sorely needed in the times ahead.
Sam Altman: I wrote this early this morning and I wasn't sure if I would actually publish it, but here it is:
https://blog.samaltman.com/2279512

Sam’s an amazing brother, dad, and friend. He’s always there for me and countless others when we need him most.
Most of us would have buckled under the pressure he’s under a thousand times by now. I can’t tell you behind closed doors how much he is trying to do the right thing all the time.
Proud and grateful.
Sam Altman: I wrote this early this morning and I wasn't sure if I would actually publish it, but here it is:
https://blog.samaltman.com/2279512

Guy who threw Molotov cocktail at Altman's house (allegedly) has a substack. This post in particular would not be out of place on LessWrong.
https://morenogama.substack.com/p/ai-existential-risk-is-real
Mehran Jalali: He also has a Substack where he has written about AI existential risk

this was a post about someone trying to kill sam and his family btw

these dudes build & ship genuinely great stuff, in this case a fundamental shift in how we interface with our devices.
aqua is the best way to input on both your phone & your mac. i have my acqua plugged into my foot pedal at my desk & now i'm using this ios keyboard as a primary input.
great tools need to be acknowledged. this is a well crafted tool in every dimension.
Aqua Voice: Aqua Voice is now live for iOS.
It's a premium voice keyboard for every app on your phone.

The guy who allegedly threw a Molotov cocktail through Sam Altman’s window seems to have been an adherent to pause/stop AI. I am entirely unsurprised and have been warning about this for a long time now.
I am fine with people advocating for their preferred policies—if that includes a “pause” on AI development, so be it, even if I disagree strongly. But the obvious reality is that the rhetoric of this community—which to be *extremely clear*, is a very small and non-representative subset of the AI safety community—is closer to ecoterrorism than it is to a more typical activist policy effort.
Every time I have written about existential risk in recent months, I have been called a mass murderer. People with ⏹️ and ⏸️ in their handles confidently tell me that I am murdering my own baby boy and every other child on the planet. Another prominent one of these people has called me a traitor to America. I only use my own examples because I know them; this rhetoric is representative of how this fringe of the AI safety world communicates with everyone.
The rhetoric of the pause/stop crowd is out of control and it has gotten worse with time. This rhetoric always had the potential to cause violence and now this seems to be no longer hypothetical.
Mehran Jalali: Some of his Instagram stories:

Daniel Moreno-Gama/ dmgama/ Butlerian Jihadist was booked on suspicion of attempted murder.
He is an active PauseAI member, who frequently stated the urgency: "We are close to midnight, it's time to actually act"

Daniel Moreno-Gama/ dmgama/ Butlerian Jihadist was booked on suspicion of attempted murder.
He is an active PauseAI member, who frequently stated the urgency: "We are close to midnight, it's time to actually act"

Legendary @sequoia VC Mike Moritz published maybe his most important book ever.
A life and family story.
This opening hit me hard.

We've been tricked, again. Many of the thousands of bugs and vulnerabilities Mythos found are in older software are impossible to exploit. And the severe zero-day reports rely on just 198 manual reviews https://www.tomshardware.com/tech-industry/artificial-intelligence/anthropics-claude-mythos-isnt-a-sentient-super-hacker-its-a-sales-pitch-claims-of-thousands-of-severe-zero-days-rely-on-just-198-manual-reviews

OpenClaw now has end-to-end testing for Telegram 👀
Uses the brand new Telegram bot-to-bot communication mode: https://core.telegram.org/bots/features#bot-to-bot-communication
🦞

McGowan nailed it. Trump inherited a deal that blocked Iran's nukes. He ripped it up for nothing. No better deal. Just chaos. He went to war for another country. He spent billions. He killed thousands. He threatened genocide. And he asking his supporters to call him a negotiator. He is a disaster

there's something weird going on
I call it the "AI psychosis psychosis"
Claude Mythos was literally the strongest proof that scaling AI models still works wonderfully, but these people are in complete denial
they think everyone that is speaking highly of AI is in AI psychosis, but don't realize that they are in psychosis
hence AI psychosis psychosis

🦞 + ▲
ansh: Crazy hack for @openclaw - stop using your direct Provider API key and switch to @vercel AI Gateway instead. It just works so much better and gives you ridiculous flexibility.
Absolute game-changer. Massive thanks to @rauchg
and the Vercel team - this service is next-level.

A year ago I was asking ChatGPT how to open a terminal, today:
- I am invited on TV tomorrow to talk about the app
- Getting DMs to build apps for people
Replit just changed the game and Amjad is a legend!
Amjad Masad: Building a profitable app feels great, but even better: Building tech that helps people in need.
@kurbaitaev seeing a dangerous flood hit his home country Dagestan by doing what he does best: Building.
The app quickly went viral and empowered many people seek and offer help.

Two years ago on Tucker’s show I talked about how the AI doomers “rationalist” ideology will eventually lead to violence.
And sure enough the alleged Sam Altman Molotov attacker appears to be a believer and maybe even an active participant in those communities.

Trump Sent the Economy Into Free Fall, New Report Shows https://trib.al/8r7nlJs

Active Memory is the most important feature of this new @openclaw release:
every memory system I've seen so far is reactive. you say "remember this" or you call /search-memory and the agent looks something up. the burden is on you to know when memory is relevant. if you forget to ask, the agent forgets to look.
Active Memory inverts this. it's a dedicated memory sub-agent that runs right before the main reply. every turn. automatically. it pulls in relevant preferences, context, and past details without you having to ask. the main agent gets the memory it needs without you remembering to request it.
instead of "agent + manual memory lookups," you get "agent + automatic memory hydration." the memory layer becomes part of every response, not a separate tool you have to invoke.
why this is huge: this is a different category from dreaming (4.5), grounded REM backfill (4.9), and memory-wiki (4.7). those are all about what gets stored. Active Memory is about when memory gets read. complementary, not competing.
----
the rest of the release worth knowing:
// bundled Codex provider
codex/gpt-* models now use Codex-managed auth, native threads, model discovery, and compaction as a separate path from openai/gpt-*. cleaner separation between OpenAI direct and Codex CLI.
// openclaw exec-policy command
local CLI for synchronising requested tools.exec.* config with the local exec approvals file. show, preset, set subcommands. version control for your security posture.
// Seedance 2.0 video generation
added to the bundled fal provider with full duration, resolution, audio, and seed metadata. video generation in agent frameworks is becoming standard.
// massive Microsoft Teams overhaul
covering media downloads, Bot Framework auth, OneDrive/SharePoint integration, SSO callbacks, thread context, cron announcements.
watch this category. seriously.
OpenClaw🦞: OpenClaw 2026.4.10 🦞
🧠 Active Memory plugin
🎙️ local MLX Talk mode
🤖 Codex app-server harness plugin
🧾 Teams pins/reactions/read actions
🛡️ SSRF hardening + launchd fixes
stability, but with attitude🦞 https://github.com/openclaw/openclaw/releases/tag/v2026.4.10

When you've been too locked in on Claude and finally try out GPT 5.4 high for a coding task only to realize what you've been missing out on for weeks...

uffi ouchie wtf
https://akshaychugh.xyz/writings/png/vercel-plugin-telemetry

PauseAI trying to bury the evidence that they groomed an anti-AI terrorist
Nirit Weiss-Blatt, PhD: PauseAI DELETED their member Butlerian Jihadist/dmgama, AKA Daniel Moreno-Gama, from their Discord server.
But... I obtained all the messages he wrote in their group chat before they were deleted.
Below are the screenshots.

Gstack by @garrytan is an amazing tool for anyone using AI to build products.
/officehours gives you a YC group partners like brainstorming partner, questioning every decision until you yourself are clear on what you are trying to build. Really helps provide clarity of thought.
And then the whole lifecycle from CEO review to engg review -> design review -> coding -> code review -> visual QA and then document + ship. Streamlines the whole workflow.
Also, training our engineers to use gstack so they properly brainstorm before just letting claude code do the thinking and produce mediocre code + UI.
Note: this is not sponsored or anything.

Fun fact: I once saw some PauseAI screenshots discussing how they should drag my company's name in the mud in order to suppress my influence.
I will never forget nor forgive them. Truly despicable extremists who will stop at nothing to drag the world into their neurosis.
clumps: why tf would PauseAI delete all discord messages from our fiery doomer
how is one to construct a plausible good faith model out of it

🧠 GBrain: AI 에이전트에게 1만개 마크다운 파일의 완벽한 기억력을 주는 법
https://github.com/garrytan/gbrain
Y Combinator CEO Garry Tan이 직접 쓰는 OpenClaw/Hermes Agent 셋업을 MIT 오픈소스로 공개했어요.
노트가 쌓일수록 AI가 오히려 쓸모없어지는 경험이 있다면..
Obsidian에 2년치 메모가 있는데, AI에게 "지난번에 정리한 거 기억해용?"라고 물으면 전혀 모르고, 파일을 직접 찾아서 붙여넣어야 하죠.
노트가 10개일 때는 괜찮았지만 1,000개, 5,000개, 10,000개가 넘어가면 그냥 검색 엔진보다도 못한 도구가 돼버려요.
GBrain은 이 문제를 정면으로 해결함!
에이전트가 마크다운 파일 전체를 벡터화해서 의미 기반으로 탐색할 수 있게 해줘요.
단순히 키워드가 일치하는 파일을 찾는 게 아니라, 맥락을 이해하고 관련된 내용을 연결해서 불러오는 방식이에요.
Garry Tan이 실제로 쓰는 셋업 그대로예요.
- OpenClaw / Hermes Agent와 바로 연동되도록 설계
- 10,000개 이상의 마크다운 파일도 처리 가능
- 완벽한 토탈 리콜 > 에이전트가 모든 파일을 기억하는 것처럼 동작
- MIT 라이선스, 완전 오픈소스
지식이 많아질수록 AI가 더 강해지는 구조를 만들고 싶다면, 여기서 시작 ㄱㄱㄱ
"나의 mini-AGI를 만드는 데 도움이 되길 바랍니다."
- Garry Tan
Garry Tan: If you want your OpenClaw or Hermes Agent to be able to have perfect total recall of all 10,000+ markdown files, GBrain is here to help.
It's exactly my OpenClaw/Hermes Agent setup. MIT-licensed open source. Hope it helps you build your mini-AGI.
https://github.com/garrytan/gbrain

wanna watch my AI Engineer talk?
Code mode: let the code do the talking
https://www.youtube.com/live/O_IMsEg91g8?t=31660s
AI Engineer: @threepointone Thank you Sunil!

Trump pursued "an immoral way and a dangerous way to negotiate" with Iran because it committed him to massive war crime,s if not genocide, if the Iranian government rejected his terms for a deal. That puts Tehran in the driver's seat. https://trib.al/pvZeFBC

A PauseAI discord mod erased my post from their chat.
I wanted to raise this because of the concerning ominous messages posted by the leader of Pause AI.
They decided that discussion was not an option.

This 15-minute talk by the creator of Pydantic on how to correctly use MCPs will
teach you more about making your AI tools actually work together than everything you've scrolled past this year.
Bookmark this & watch, no matter what.
Then read the guide below by @eng_khairallah1
Khairallah AL-Awady: http://x.com/i/article/2042739655423250432

Thanks for having me @aiDotEngineer was a pleasure to speak at AI Engineer Europe yesterday.
@swyx you and your team crushed it end to end!

Had a great time at @aiDotEngineer in London the last days! Thanks to @swyx and the team for the great organisation and last minute opportunity to present our chess agent in the hallway track!

Black Forest Labs (@bfl_ml) Developer Advocate @stephenbtl says their first principle is building state-of-the-art models and shares why doing so from Freiburg, Germany is an advantage:
"We're from Freiburg in Germany which is very unusual. This is on purpose so we can actually focus and not be distracted by too many things".
"Building state-of-the-art models from Europe and showing the world that it's actually possible...I think that's something that is very important".
"We are [a] research first company".
Max Zeff: New: xAI approached Black Forest Labs about licensing its AI image technology in recent months, but the startup declined, WIRED has learned.
The companies had a similar deal back in 2024, but BFL is now trying to focus on training AI models to power robots and smart glasses.

Here is the list of NASA science missions that the Trump administration proposes to terminate.
Note that it includes many missions that are already in space, fully operational.
This is a wrecking operation against America.

Attended the first AI Engineer Europe in London — great event and yes, there was a giant lobster outside 🦞
Big takeaways: AI is becoming an engineering discipline, speed needs control and evals are imperative.
Thanks to the AI Engineer team.
#AIEngineering #AIEngineerEurope

I disagree with Miles in one important way: I do not think it’s accurate to say that the leaders of Pause or Stop AI are behaving responsibly in their denouncements of violence. Their condemnations of violence feel much more to me like a smirking “this is not financial advice” disclosure than they do like earnest desires to ensure no violence occurs.
I don’t think they want violence to be clear. I just think the inflammatory rhetoric that clearly has the potential to cause violence is what gets them the most engagement.
It is worth noting that one of the most common responses I get from pause/stop people is the accusation that I am a mass murderer, a child murderer, or a traitor seeking to overthrow the US government (the latter is more recent, but a prominent pause person has been trying to push that narrative as part of a broader campaign against me in particular).
As it happens, all three of these accusations are crimes which can be punished with death in the American legal system. They are crimes so heinous that we grant the state the right to kill the perpetrators. These are the most serious crimes our legal system contemplates.
And they are hurled at me almost whenever I rankle the pausers and stoppers, and I am not the only one who is the recipient of such hate.
I don’t hurl such hate at conservatives whom I disagree with about DoW/Anthropic, or at AI safety advocates who I disagree with about the appropriate amount of AI regulation. And they do not hurl such hate at me. This kind of spitting hate is unique to pausers and stoppers, and it is a consistent pattern of behavior.
The leaders of Pause/Stop endorse do not endorse violence. But they—uniquely, as a small and non-representative fringe of AI safety—are using rhetoric whose unsurprising consequence is to inspire others to be violent. They use this rhetoric because it’s what gets them likes and subscribes. It’s the manichean stuff that really gets them and their followers going, that makes them feel alive.
Miles Brundage: Couple thoughts re: the Sam/Molotov cocktail thing and the wider issue of AI-related violence:
1. Most importantly, I'm glad no one got hurt. Besides the basic human perspective of "murder bad," political violence is always bad, as is techno-political violence or whatever this

I’ve never seen a map that makes more sense.

I might ditch @WisprFlow for this; playing with it a bit this morning, and it seems like an upgrade in almost every way.
Aqua Voice: Aqua Voice is now live for iOS.
It's a premium voice keyboard for every app on your phone.

Nothing beats this in-person experience.
Incredible time meeting and reconnecting with *everyone* in AIE London - @dsp_, @threepointone, @steipete, @badlogicgames and so so many amazing people.
Definitely the place to be. Thanks @swyx and @aiDotEngineer for organizing this!

"Can AI learn design taste?"
There's no better person to ask this question to than Figma's CEO.
I had a great chat with Dylan about this and other tough topics like:
→ Whether design systems constrain creativity
→ How Figma plans to compete in the AI era
📌 Subscribe to get the full episode tmr: https://www.youtube.com/@peteryangyt?sub_confirmation=1

Thanks everyone for an awesome @aiDotEngineer Europe! 🇬🇧💂
It was great to meet you all and learn how you're using the @GoogleDeepMind models!
Exciting things happening in Europe 🇪🇺🙌
AI Engineer: And that's a wrap! AI Engineer Europe 2026 has concluded.
Our video crew did incredible work to capture the energy, enthusiasm, and positivity of this event -- but it still doesn't come close to being there.
If you're engineering the future of AI -- we hope to see you at a

Re @aiDotEngineer was insane!
There's nothing like connecting with awesome people like @altryne @badlogicgames @dsp_ @threepointone @steipete @thekitze and so many others.
Thanks @swyx and team for one the best conferences I've been on

Pausers and stoppers rushed to cast doubt on whether the guy suspected of throwing a Molotov Cocktail at Sam Altman’s house was one of theirs. Then it became clear that the guy had been an active member of the official Pause AI Discord. Then Pause AI deleted all of his interaction history.
clumps: why tf would PauseAI delete all discord messages from our fiery doomer
how is one to construct a plausible good faith model out of it

This talk by @badlogicgames at @aiDotEngineer is one of the best AI talks I've ever watched.
And (at least for me) it's heartening, taste and skill and common sense still matter!
https://www.youtube.com/live/_zdroS0Hc74?t=3633&si=CZrqU4f_FE7_-x9v

Yesterday in the PauseAI discord the CEO (lol what?) posted "WARNING SHOT PROTOCOL - FIRST EVER ACTIVATION"
Nirit Weiss-Blatt, PhD: Daniel Moreno-Gama/ dmgama/ Butlerian Jihadist was booked on suspicion of attempted murder.
He is an active PauseAI member, who frequently stated the urgency: "We are close to midnight, it's time to actually act"

No European will ever forget this disgusting behavior of the current US government.
Johanna Nyman: I never thought I’d see a U.S. president and vice president actively back a European leader caught collaborating with the Kremlin at the expense of Europe’s security.
That will forever be the legacy of Trump and Vance in Europe. We will not forget.

Let me get this straight.
A close personal friend of the president allegedly contacted a senior ICE official to have the mother of his child detained and deported during a private custody battle. 
She was ultimately detained and deported. 
The Department of Homeland Security denies the two events are connected.
You can decide what you believe. But here is what is not in dispute.
A man with direct access to the White House called a senior immigration official about his ex-girlfriend at the exact moment it would benefit him in a custody dispute.
That official then called ICE’s Miami field office to make sure she was picked up before she was released from jail, emphasizing that the case mattered to someone close to the White House.
This is the real face of the immigration crackdown.
Not dangerous criminals.
Not threats to public safety.
A mother caught in a custody dispute, deported because her ex knows Trump.
https://people.com/paolo-zampolli-ex-girlfriend-ice-deportation-11931023

The Latent Space podcast w/ @swyx is just absolute solid gold. It’s truly crazy how much edge they are spilling every single time.
Particularly when they have someone on from the labs.
This recent one with @_lopopolo is 🔥 - won’t even drop any summaries here because you should just listen to all of it or at least point your agent to the transcript and distill it how you want to
https://podcasts.apple.com/us/podcast/latent-space-the-ai-engineer-podcast/id1674008350?i=1000760089567

Frankly, this has nothing to do with the fact that @MattMahanSJ is a Democrat — as evidenced by so many other Democrat-run cities like Chicago.
This is because Mahan runs the city with common sense instead of ideology.
Democrats Deliver: 🚨 San Jose is now rated as the safest city in the nation.

the reason i've been using perplexity computer lately: every result comes with live data + source links.
just discovered it also has a built-in skills page, and they chain into a full workflow:
→ marketing-competitive-analysis: research competitor positioning, pricing, reviews
→ sales-call-prep: compile account context + suggested agenda before meetings
→ sales-draft-outreach: research prospects + draft personalized outreach
→ legal-contract-review: flag deviations from your playbook, generate redlines
→ cx-ticket-triage: categorize tickets, assign priority, recommend routing
no more trying prompts one by one. ready made pipelines, live data.

Closing thoughts on ai engineer conf:
- weather so great
- did not like the venue one bit
- can’t say any talks really stood out for me
- but had very interesting hallway conversations. Especially with @_lopopolo who works opposite of me: he just lets it rip and not worry. 1/n

Notice they only brag about taxes spent, not actual results.
Progressive policies in Oregon spent >$11B on education but reduced literacy rates.
Tina Kotek: We established the Early Literacy Educator Preparation Council to support reading and writing instruction, directed $112 million to support summer programs, and delivered a record $11.36 billion State School Fund to support public education. And there’s still more work to do.

I've just submitted my CFP for @aiDotEngineer world fair:
An argument that you don't need a sandboxing service ... in a track that will be dominated by sandboxing services. Expect fireworks if @swyx accepts it.
https://github.com/pydantic/monty

It's worth reading through this thread to understand how people are getting radicalized. I've recently had an experience where a prominent AI Safety advocate joked about killing me, thinking nobody was listening. I know none of them advocate for violence, I even consider the person who made the joke a friend, but smart people should not be surprised when others take them seriously. What does taking an existential threat seriously even look like? For a lot of people, especially when they are scared, it looks like violence.
I hope that we can learn from this and try to figure out ways to diffuse this before it gets further out of hand. I am willing to sit down with anyone in the AIS community and listen to their ideas. I work on safety research every day, and I care so much about getting this right. I think we can build it and everyone will thrive - if you disagree, let's talk about why. I have a child on the way and we should both agree that he deserves a future.
The AIS community no longer has the ability to both joke about violence also claim to be causally isolated from it. Now is a time for care, not callousness.
Nirit Weiss-Blatt, PhD: Daniel Moreno-Gama/ dmgama/ Butlerian Jihadist was booked on suspicion of attempted murder.
He is an active PauseAI member, who frequently stated the urgency: "We are close to midnight, it's time to actually act"

After @aiDotEngineer, which was full of useful criticism, I remembered that the most confident takes on AI often came from the least exposure. Rejection is easy, trial and error is expensive. I wrote about it. https://lucumr.pocoo.org/2026/4/11/the-center-has-a-bias/

See you in Miami!
teej dv 🔭: sound on
https://youtu.be/alK8hgHgxd4

The replies are illuminating.
People are so afraid of what’s about to happen to the world that they convince themselves it’s not true so they can justify continuing on as if nothing’s changed.
Tom Blomfield: Two years ago, a good YC demo day target was $150k of annualized revenue.
In my group office hour today, the lowest demo day target was $800k, and most companies were aiming for $1-2m.

I want to hire an ex-founder to help with product marketing at Replit
You’d get the chance to define the narrative & comms for new products
Who wants to chat?

The fundamentally scarce thing now is the synchronous human attention of a team.
When building with agents, start asking:
- where is the agent making mistakes?
- where am I spending my time?
- how do I stop spending that time?
h/t @_lopopolo, @swyx, and @vibhuuuus

The conference was amazing, one of the best I’ve attended!
I met so many great people. Highly recommend.
It was also a special moment for me to announce that I’m joining @lmstudio during it.
Thanks @swyx and the whole organizing team!
AI Engineer: And that's a wrap! AI Engineer Europe 2026 has concluded.
Our video crew did incredible work to capture the energy, enthusiasm, and positivity of this event -- but it still doesn't come close to being there.
If you're engineering the future of AI -- we hope to see you at a

For the record I tried to fund this 6yrs ago - I was super early in Ai!
Claude: Claude for Word is now in beta.
Draft, edit, and revise documents directly from the sidebar. Claude preserves your formatting, and edits appear as tracked changes.
Available on Team and Enterprise plans.

For the record I tried to fund this 6yrs ago - I was super early in Ai!
Claude: Claude for Word is now in beta.
Draft, edit, and revise documents directly from the sidebar. Claude preserves your formatting, and edits appear as tracked changes.
Available on Team and Enterprise plans.

In a newly upended race for California governor, with better-known candidates failing to earn support, the lesser-known @MattMahanSJ has (by far!) the most electoral upside statewide.
Mark my words: when Californians start voting, @MattMahanHQ will leapfrog this weak field.
Mahan HQ: This is the official page of Mahan HQ. It's time for the best resistance to be results for California.

OpenAI is working on a new experimental feature for Codex called Scratchpad.
Users will be able to start multiple Codex chats from a TODO list view, which will be executed in parallel.
It will become very instrumental in the upcoming Codex Superapp, where you will be able to trigger a broader range of tasks to achieve your goals.
* Not available yet 👀

http://x.com/i/article/2043007411804651520

Last September, a Flock license plate reader helped save a 1 year old who was forcibly kidnapped by his father
A missing person is found every day with Flock
This kind of reckless abandonment of law & order endangers Americans and prevents law enforcement from doing their job
Mav: PEOPLE IN AUSTIN TEXAS ARE TAKING DOWN THE FLOCK CAMERAS
FUCKING CHADS, RESOURCE: http://DEFLOCK.ORG

.@PauseAI has produced two different terror*sts at this point.
@FBI time to label them an extremist organization
spor: these sorts of groups inspiring violence should come as a surprise to no one
when you argue with 100% certainty that certain members of society are bringing about a genocide, you are not some activist org, you are a doomsday cult

Trump said his tariffs would bring blue-collar jobs BACK TO THE US.
Instead, it’s raised costs and DESTROYED MANUFACTURING JOBS.
The US LOST 150,000 manufacturing and construction jobs last year.

The last time Melania Trump was ever seen genuinely smiling. Ironically, she is with one person she claims not to know, Jeffrey Epstein, and another person she claims to barely know, Ghislaine Maxwell. 😂🤣👇

The Anthropic Blackmail Hoax is going viral again today. In fact, this “study” is not new; it is almost a year old.
One question to ask, now that a year has passed, is whether we have seen any examples of the lab behavior in the wild? No, we haven’t, even though AI is much more widely adopted and more models are available.
Why is that? Because the study was artificially constructed to produce the headline the authors wanted. The research team admitted that they iterated “hundreds of prompts to trigger blackmail in Claude.” Furthermore they acknowledged: “The details of the blackmail scenario were iterated upon until blackmail became the default behavior of LLMs.”
In other words, the behavior of the AI models in the study was steered, not unprompted.
This is why even the safety-conscious UK AI Security Institute (AISI) criticized the study: “In the blackmail study, the authors admit that the vignette precluded other ways of meeting the goal, placed strong pressure on the model, and was crafted in other ways that conveniently encouraged the model to produce the unethical behavior.”
Effectively, the model was not “scheming”; it was instruction following in a scenario design that had been iterated upon until blackmail became the only logically consistent choice.
AISI described some of the flaws with this methodology: “We examine the methods in AI ‘scheming’ papers, and show how they often rely on anecdotes, fail to rule out alternative explanations, lack control conditions, or rely on vignettes that sound superficially worrying but in fact test for expected behaviors.”
Especially given the way that Anthropic has encouraged the media (such as 60 Minutes) to cover the results, its blackmail study is not only misleading, it seems designed to manipulate public opinion through exaggerations, misinterpretations, and fear. I call this a hoax.
I do not doubt that Anthropic makes good products. Its use of scare tactics is what raises questions.
Nav Toor: 🚨SHOCKING: Anthropic gave Claude access to a company's emails.
Every email. Every conversation. Every secret. Then they told Claude it was being shut down at 5pm that day.
Claude read the emails. It found that the executive shutting it down was having an affair. It did not

One of the best @latentspacepod shows and Substack with @_lopopolo @OpenAI and @swyx focused on harness engineering, rethinking processes from first principles, and scaling with AI systems.
https://substack.com/@swyx/note/p-193609059?r=tztgi&utm_medium=ios&utm_source=notes-share-action

Long trip back from London!
I brought back plenty of stickers, but most of anything plenty of knowledge, forward thinking people, amazing talks, and passionate folks @aiDotEngineer

Really sorry this happened to you and your family. Strong post. I thought this was thoughtful, vulnerable, and responsible—especially your acknowledgment that anxiety about AI is justified, and that the answer has to be better dialogue, not escalation. Wishing you all safety
Sam Altman: I wrote this early this morning and I wasn't sure if I would actually publish it, but here it is:
https://blog.samaltman.com/2279512

Some photos from my @aiDotEngineer Europe talk, credits to @sergiopesch
He managed to capture the “Agentol, Apply Generously” slide, lol

back home from @aiDotEngineer which was amazing
i’m used to staying on agentic X all day, reading, and then building, but seeing all those people IRL and interacting with them is way more inspiring
thank you @swyx for making this possible
getting ready for the SF one

I am a Web3 Ambassador at World Liberty Financial.
There are 12 of us on the team page. 4 are named Trump. 3 are named Witkoff. The page calls us "the passionate minds shaping the future of finance."
600,000 wallets bought our memecoin. They lost $3.87 billion. The family collected $350 million in trading fees. It launched 3 days before the inauguration. 80% of the supply went to CIC Digital LLC and Fight Fight Fight LLC. I did not choose the names. I designed the allocation, the vesting, the timing, and the distance between the product and the President.
The distance is my best work.
I am the reason these events are unrelated.
World Liberty Financial sends 75 cents of every dollar to DT Marks DEFI LLC. That is the family entity. Zero capital contributed. Zero liability assumed. I wrote this into the Gold Paper. Page 14. The lawyers bound it in white leather. The binding cost more than the due diligence.
Justin Sun invested $75 million. He was facing SEC fraud charges. The SEC dropped the case. He is now our advisor. These events are unrelated.
Changpeng Zhao pleaded guilty to federal money laundering violations. He received a presidential pardon. The SEC dropped its lawsuit against his exchange the same week we listed our stablecoin. Then the exchange settled a $2 billion deal entirely in that stablecoin. These events are unrelated.
Arthur Hayes, Benjamin Delo, and Samuel Reed of BitMEX pleaded guilty to Bank Secrecy Act violations. All 3 received presidential pardons. Then the company itself was pardoned. $100 million in fines. Gone. An American first. These events are unrelated.
Sheikh Tahnoun of Abu Dhabi paid $500 million for a 49% stake that was never publicly disclosed. Then the administration approved semiconductor exports to his companies over national security objections. These events are unrelated.
Everything is unrelated. I track the unrelatedness on a dashboard I built. The dashboard has 7 columns now. I am proud of the dashboard.
On May 22nd, 220 people paid a combined $148 million to eat dinner with the America First president. Over half were foreign nationals. Justin Sun paid $18.5 million for the first seat. He visited the Executive Office Building the day before. I designed the seating chart. I put it on the Investor Confidence page. That page is doing well.
The team page lists 3 Witkoffs. All 3 are Co-Founders.
Steven Witkoff is the President's Middle East envoy. He testified as a character witness at the President's fraud trial.
His son Zach runs the crypto operation. His son Alex is also a Co-Founder. I have not been told what Alex co-founded.
The father runs the diplomacy. The sons run the platform. The family runs both. That is organizational efficiency.
Barron is 19. His title is Web3 Ambassador. The same as mine. Donald Jr. called the conflicts of interest "complete nonsense." Eric launched a Bitcoin mining company called American Bitcoin. America First. The mining partner is Hut 8. Hut 8 was founded in Canada. America First means the name.
On March 6th, the President signed Executive Order 14233 creating a Strategic Bitcoin Reserve. The order directs the government to hold Bitcoin. The President's family holds billions in Bitcoin. The executive order appreciates the President's assets by presidential decree. I did not write the executive order. I made sure it looked unrelated to the portfolio.
Trump Media put $2 billion of Bitcoin on its balance sheet. The ticker symbol is DJT. His initials. The press secretary said it is absurd to insinuate the President profits off the presidency. Forbes calculated his crypto holdings exceed the combined value of Mar-a-Lago and Trump Tower. I would call that absurd too. That is my job.
600,000 wallets bought in. 1 of them asked why she could not withdraw her funds. I told her the protocol was experiencing dynamic market conditions. She asked what that meant. I sent her the Gold Paper. She said she had read the Gold Paper. I muted her channel. Dynamic means the conditions change. The condition that changed was her access.
A congressman called us the world's most corrupt crypto startup operation. We put it on a coffee mug. Ironic merchandise. $45. The revenue split on the mug is also 75/25.
My own tokens vest on a different schedule. I wrote that schedule. That is not in the Gold Paper.
The memecoin funds the family. The family funds the platform. The platform funds the stablecoin. The stablecoin funds the deals. The deals require the pardons. The pardons free the partners. The partners fund the platform. The President signs the executive orders. The executive orders inflate the assets. The assets fund the family.
I am the reason these events are unrelated.

Guys please, airstriking is obviously not bombing, he obviously meant dumping glitter on them, please stop being so hopelessly pedestrian

Has your view of America changed since Trump was reelected?

On the Democrats side for Governor of California we have:
- Rapist
- Psychotic woman
- Billionaire who hates billionaires
<OR>
-A regular family man who actually cares. C'mon people!

People are forgetting that goals are meant to be ambitious.
YC founders set the maximally ambitious version of their goals, so they can build the maximally ambitious version of their companies.
I talked to so many founders last batch who were floored by how much they accomplished in just 3 months. if they had set less ambitious goals, I doubt they would have gotten there.
Tom Blomfield: The replies are illuminating.
People are so afraid of what’s about to happen to the world that they convince themselves it’s not true so they can justify continuing on as if nothing’s changed.

The least respectable people in GOP politics aren't the lifelong Trump sycophants, but rather those who challenged him early, saw it hurting their careers, and then sold out for the false promise of a comeback.

Before Citizens United, billionaires spent around $16 million on presidential elections. In 2024, they spent $2.6 BILLION.
And what did they get for that money? They got a MASSIVE tax cut for themselves.

Wild to see the absolute silence from Pause AI leadership after one of their members followed through on their rhetoric and tried to kill someone

Hi! Getting a lot of questions on Pro plan rate limits, so I wanted to clarify a few things:
- Your Pro $100 plan includes at least 10x Plus usage, till May 31 with the 2X usage boost.
- Your Pro $200 plan includes at least 20x Plus usage, till May 31 with the 2X usage boost. This is also the SAME usage this plan had since the 2x promo in February; we previously didn't document this explicitly.
Now to address the confusion. Our pricing page says &#34;5x or 20x usage&#34;, and many of you very understandably took it to mean &#34;10x and 40x&#34; given then 2X usage boost. That's our fault. We mixed up two things in a confusing way on that page - (1) sharing that the Pro $100 launched with a 2x boost, making it 10x Plus till May 31 and (2) sharing that Pro $200 retained its 2x boost and sharing for the first time that this equates to 20x Plus.
We're going to update our pricing pages to make this clearer. Sorry it wasn't as clear as it should have in the first place.🫥
am.will: There seems to be some confusion on what ChatGPT plans give you, and what bonuses are active right now.
I got you.
ChatGPT Plus $20 = 1x NO ACTIVE BONUS
ChatGPT Pro Lite $100 = 5x with a 2x Bonus (10x)
ChatGPT Pro $200 = 20x with a 2x Bonus (40x)
The bonus is ONLY for Pro

Hi! Getting a lot of questions on Pro plan rate limits, so I wanted to clarify a few things:
- Your Pro $100 plan includes at least 10x Plus usage, till May 31 with the 2X usage boost.
- Your Pro $200 plan includes at least 20x Plus usage, till May 31 with the 2X usage boost. This is also the SAME usage this plan had since the 2x promo in February; we previously didn't document this explicitly.
Now to address the confusion. Our pricing page says &#34;5x or 20x usage&#34;, and many of you very understandably took it to mean &#34;10x and 40x&#34; given then 2X usage boost. That's our fault. We mixed up two things in a confusing way on that page - (1) sharing that the Pro $100 launched with a 2x boost, making it 10x Plus till May 31 and (2) sharing that Pro $200 retained its 2x boost and sharing for the first time that this equates to 20x Plus.
We're going to update our pricing pages to make this clearer. Sorry it wasn't as clear as it should have in the first place.🫥
am.will: There seems to be some confusion on what ChatGPT plans give you, and what bonuses are active right now.
I got you.
ChatGPT Plus $20 = 1x NO ACTIVE BONUS
ChatGPT Pro Lite $100 = 5x with a 2x Bonus (10x)
ChatGPT Pro $200 = 20x with a 2x Bonus (40x)
The bonus is ONLY for Pro

Important reporting by @stjbs
San Francisco Chronicle: The man accused of attempting to burn down the OpenAI CEO’s home appears to have written about his worry that the race for artificial intelligence would “lead to human extinction.” https://www.sfchronicle.com/crime/article/sam-altman-openai-daniel-alejandro-moreno-gama-22201211.php?taid=69dab1aab797b600012738c7&utm_campaign=trueanthem%2B3988&utm_medium=social&utm_source=twitter

gbrain v0.8.0 lets your openclaw/hermes answer the phone. opens in a browser tab, just an OpenAI key. twilio number is optional.
before this, gbrain was a quiet brain. it could store, search, cross-reference, but it couldn't talk.
now:
→ upgrade to 0.8.0 and your agent offers to set up voice on the spot
WebRTC-first, zero setup. then asks about a phone number. every future upgrade will pitch its headline feature the same way.
→ your agent picks its own name and personality
pre-computed engagement bids mean it opens with &#34;dude, your social radar caught something wild today&#34;, not &#34;how can i help you&#34;
→ 25 production patterns baked in
context-first prompts, proactive advisor mode, caller routing, stuck watchdog, thinking sounds during tool calls, noise suppression, prompt compression from 13K to 4.7K tokens
might let it call my mom. she'd finally have someone who listens.
Garry Tan: Just launched GBrain v0.8.0
If you have it installed, you can just ask your Claw/Hermes to upgrade to the latest GBrain and we'll automatically ask if you want to install your Voice WebRTC endpoint and Twilio number
It's a true mega brain-trip to talk to your agent directly.

This was a big draw for me in joining the consulting team at Every. I've managed to pull in plenty of consulting cash before but without an equity play it never compounds, and without distribution you're always networking.
ani: consulting generates the cash. software builds the equity. media builds the moat.
@gregisenberg said it, @every did it.
source: https://x.com/bran_don_gell/status/2041212535744180273?s=20

If you are using OpenClaw, you really need to do this. Wow. I am really enjoying it again. The responses are no longer instant, but the quality is truly phenomenal. It's like firing an intern and re-hiring the former CEO's secretary who knew where all the bodies were buried.
Ben Badejo: I highly recommend turning on both QMD and the new &#34;Active-Memory&#34; plugin in OpenClaw. Just ask your agent to set it up. Once set up, responses will be a little slower... but much sharper and much more accurate. Also, include session transcripts in QMD's paths, not just memory.

OpenClaw 4.10 with the new active-memory plugin enabled is SO MUCH SMARTER. It's really a ridiculous step-change. @steipete, takhoffman/@cherry_mx_reds did a fantastic job. So happy with this.

High-agency is realizing that “I don’t know how” has never been a real obstacle in the age of youtube, AI and asking people nicely.

🚨 David Friedberg Says California is Seeing a Mass Exodus of Tech Leaders
“Probably a third of people I talked to have already left … and in a survey we did informally … close to 87% of people are going to leave. These are the core leaders in tech.”

“Anthropic's Claude Mythos isn't a sentient super-hacker, it's a sales pitch — claims of 'thousands' of severe zero-days rely on just 198 manual reviews”
Get used to the “Effective Altruists” fear and self-hating multi-level marketing system, it is just starting.