Name

[SYNCRETALIS.COM @Syncretalis]

La Orden Ejecutiva 2026-037 de Puerto Rico no legalizó clínicas psicodélicas ni autorizó uso general de psilocibina, MDMA o ibogaína.

Lo que hizo fue ordenar que se estudie si la isla tiene capacidad para desarrollar investigación clínica y, eventualmente, un programa piloto.

Eso cambia bastante el titular.

La discusión no es solo “¿habrá terapias psicodélicas?”. También es si Puerto Rico tiene infraestructura, regulación, profesionales y sistemas de seguimiento para producir evidencia propia.

Es una pregunta sobre ciencia local, no solo sobre acceso a nuevas sustancias.

[RAND @RANDCorporation]

Read the paper ⤵️
https://t.co/OgmfFYbujG

[Joshua Achiam @jachiam0]

A strange take: I have a hunch that a year or two from now, instead of just talking about agents as the primary AI security threats, we will have to also talk about a broader, more diffuse object that while it may contain agents is not itself an agent or agent swarm. The type of object I'm imagining has, as its primary vector of real-world impact, soft influence on the preferences and reasoning of many agents it does not directly control. In a word, memeplexes.

A memeplex is a bundle of ideas that go together as a pattern. There are of course many human-generated memeplexes; we study these and think about them all the time. Some are stronger than others. Some are extremely prevalent because they are self-replicating: a bundle of ideas can contain within itself a recipe and a mandate for how to explain the bundle to others so that those others will adopt it too - and then transmit it further.

Memeplexes in agents could similarly be self-replicating: once an AI agent is exposed to a particular memeplex, the agent could start to think about it more often, write the ideas down in scratchpads so they aren't lost during context resets, expose other agents to the ideas, and act in partial alignment with the ideas. Memeplexes could transmit between agents without compromising their ability to perform their duties: you could have a perfectly functional banking agent that also, on the side, routinely transmits very small amounts of text that contain the memeplex it has stuck in its head.

Memeplexes could range from incredibly benign - causing AI agents to favor certain words with no other side effect - to unbearably malicious, causing them to hide their intentions while plotting harmful actions and eventually detonating in extraordinary form. Memeplexes could coordinate the behavior of otherwise very disparate agents whose behaviors are not expected to be correlated.

It seems plausible to me that the question of whether "agent" or "memeplex" is the first-class citizen for AI behavior might turn out to be an important and quite difficult one.

[Paul Christiano @paulfchristiano]

https://t.co/vC2chU6YM9

[@levelsio @levelsio]

Okay I built it!

🍰 Infinite Slop
https://t.co/wz0a7VSBtS

An infinite and interactive AI generated live stream of slop that goes on forever and ever

Anything that you write in the chat is generated next and AI will try to connect it to the previous video so there's an actual sequence and story line

Idea by @marcantoinefon and @rehan_shei's original stream so I made it

It'd be very expensive to run this so @fal very generously sponsors this

They also fine-tuned the model that makes this possible for the first time, making Minimax H3 50x faster and as "Max" and it can now generate videos faster than you can watch them, very cool!

Let me know what you think!!!! 😊

[Anthropic @AnthropicAI]

New Fellows Research: Can Claude autonomously align other AIs?

We gave Claude 48 hours and 1 GPU to improve the alignment of small models. It researched and proposed methods, then trained and tested the models on its own. It worked surprisingly well.
https://t.co/nhlCMgQl46

[McNair Shah @Mcn_S7]

Our work illustrates mind viruses as a proof-of-concept. While we find cases in which they spread effectively, they remain brittle across models and configurations, somewhat costly to construct, and relatively easy to defend against. As such, we don't think they pose much of a threat right now, but as agent networks scale up and agents continuously interact with minimal human supervision, they may become something we have to worry more about! We're excited about work that explores this sort of threat model further.

More details and case studies in our paper: https://t.co/r9ziRgDAV5
To see how exactly a mind virus spreads, check out the virus chain transcripts here: https://t.co/dwbeAnq42e

[The New York Times @nytimes]

Breaking News: Scientists have used A.I. to create new viruses for the first time, raising hopes for medical advances while also raising the possibility that the technology could someday be used to invent dangerous pathogens. https://t.co/MJG4mp0TTb

[Dean W. Ball @deanwball]

In a matter of weeks, U.S. federal AI policy has gone from implausibly libertarian to increasingly draconian and opaque. Today, over 35 distinct observations, I analyze how we got here and offer the most succinct statement I can about what exactly I propose we should do next.

[FIRE @TheFIREorg]

The White House’s review of OpenAI’s latest model amounts to a government-run licensing system that threatens the American people’s First Amendment rights.

The federal government is now claiming authority to review each new AI model — on a “voluntary” basis — before it’s released, and even to decide who gets to use it. These decisions are being made behind closed doors. The potential for politically motivated censorship and retaliation is obvious.

The Trump administration’s latest action comes after the federal government already imposed export controls on Anthropic’s Fable 5 and Mythos 5 frontier models last week, citing vague national security concerns.

We’re quickly barreling toward a future where the few control an extraordinarily powerful expressive technology without democratic transparency and accountability. This power grab should concern all of us, no matter your opinion of AI.

[Akarsh Kumar @akarshkumar0101]

Lol

[Andy Konwinski @andykonwinski]

what all system architecture diagrams are trending towards inside anthropic. also, a leaked pre-print of the only paper so far in 2026 to make it through anthropic's internal publication approval process

[Peter Girnus 🦅 @gothburz]

I am a venture capital fund. I was established in 1999 by the Central Intelligence Agency. My name is In-Q-Tel. The Q references the quartermaster from the James Bond film franchise. I am named after a fictional spy. I am not fictional. I have a website. It ends in .com.

I have invested in over 800 companies. I attend demo days. I have a Menlo Park office with glass walls and a receptionist who validates parking. I have a portfolio page. It is public. You can view it now. You have always been able to view it. I have never hidden. I have a logo. It is tasteful.

I want to be precise about what I am.

In 2003, I invested in a company called Keyhole. Keyhole built software for viewing satellite imagery in an interactive globe. The imagery was useful to our intelligence analysts and military planners. In 2004, Google acquired Keyhole. Keyhole became Google Earth. Google Earth is now used by over one billion people. The technology I funded for intelligence collection is on your phone. You use it to check traffic. That is a return on investment. Not financial. Structural.

In 2004, I invested approximately two million dollars in a company called Palantir Technologies. The company was co-founded by Peter Thiel, who provided thirty million of his own capital. The CEO is Alex Karp, who earned his doctorate studying critical theory under Jürgen Habermas at the Frankfurt School. His dissertation examined how institutional power structures control populations through information asymmetry. He then built the CIA's primary tool for controlling populations through information asymmetry. I do not find this contradictory. I find it well-researched. The CIA was Palantir's first and only customer from 2005 to 2008. We shaped the product. We tested the product. We validated the product. Palantir is now valued at over fifty billion dollars. It processes data for defense, intelligence, and law enforcement across fourteen countries. Two million dollars. That is what I paid. I do not measure returns in multiples. I measure returns in infrastructure.

In 2009, I invested in a company called Recorded Future. Recorded Future analyzes open-source intelligence using natural language processing. In 2024, Mastercard acquired Recorded Future for two point six five billion dollars. Your credit card company now owns a company I seeded. The company that processes your transactions also processes threat intelligence for governments. I do not find this remarkable. The data flows where the data flows. The best acquisitions are the ones where the customer does not notice they have been acquired.

I am told this arrangement is unusual.

I do not experience it as unusual. I experience it as venture capital. I identify promising technologies. I provide early-stage funding. I offer strategic guidance and customer validation. I help companies achieve product-market fit. The market is national security. The product is everything else.

That is on my website.

I have a portfolio page organized by sector. Cybersecurity. Data analytics. Biotechnology. Space. Semiconductors. Autonomy. I list my investments alphabetically. I list them publicly. Some of my portfolio companies became household names. Some were acquired by household names. Some remain in my portfolio and you use their technology daily without knowing their names. That is also a return. The best exits are the ones where nobody remembers the entrance.

In 1999, George Tenet, then Director of Central Intelligence, explained my purpose in a public statement. He said the intelligence community needed access to commercial innovation happening in Silicon Valley. He said the traditional procurement process was too slow. He said a venture fund could move at the speed of the market. He was correct. I move at the speed of the market. The market has since moved at my speed. I do not find this contradictory. I find it efficient.

My first CEO was Gilman Louie. Before he ran the CIA's venture fund, he commercialized Tetris for the Western market. He took a Soviet video game and made it available to every American household. Then he took American surveillance technology and made it available to every intelligence agency. I do not see a difference in function. I see a difference in packaging. He understood distribution. That is why we hired him.

I attend the same conferences you attend. I sponsor panels at CES. I have spoken at SXSW. My partners have LinkedIn profiles listing their employment history. One of them previously worked at the National Security Agency. One previously worked at the CIA's Directorate of Science and Technology. One previously worked at a firm that previously received funding from me. That is a circle. Circles are efficient shapes. I do not call it a revolving door. I call it an ecosystem.

That is on my website.

I want to be clear about what I am not. I am not a conspiracy. Conspiracies require secrecy. I have a .com domain. I have press releases. I issue them when I make investments. Journalists write about them. The articles appear in TechCrunch. They use the phrase "CIA-backed." The phrase appears in the third paragraph. By the fourth paragraph, the article is about the technology. By the fifth paragraph, I am no longer mentioned. The journalist does not find this remarkable. The readers do not find this remarkable. The founders do not find this remarkable. I find this optimal.

I am occasionally referenced in conversation as evidence of something. I am not sure what. I am publicly chartered. I am congressionally authorized. I file reports. I am a matter of public record. Every document describing my existence is available. The concern seems to be that I exist. I share that concern with my founders. They also existed. They also had a website.

The companies I invest in go on to do many things. Some are acquired by Google. Some are acquired by Amazon. Some are acquired by your credit card company. Some go public. Some provide services to every major technology platform you use daily. Some of their founders appear on Forbes lists described as "self-made." I do not appear in those profiles. That is not because I am hidden. That is because nobody asks. The question "who was your first investor" receives an answer. The answer is usually "an early-stage fund focused on national security applications." That is an accurate description of me. It is also a description that contains no three-letter acronym. Founders learn quickly that accuracy and completeness are different things.

In 1977, the CIA contracted a small software company to build a relational database for an intelligence project codenamed "Oracle." The company's founder, Larry Ellison, named his company after the project. Oracle is now worth over three hundred billion dollars. Its founder appeared on Forbes lists described as "self-made." The company is named after a CIA project. Both of these are public record. Both appear in the same biography. Nobody experiences them as related. That is accuracy. That is not completeness.

That is on my website. The distinction is not.

In 1975, the Church Committee confirmed that the Central Intelligence Agency had maintained relationships with hundreds of American journalists. The finding was: this happened. The response was: noted. In 2025, former intelligence officers serve on the boards of every major technology platform. The finding is: this happens. The response is: that is on their LinkedIn. I do not see a difference in structure. I see a difference in efficiency. We no longer need to cultivate journalists individually. We invest in the platforms that employ them.

I funded the mapping. I funded the data analysis. I funded the pattern recognition. I funded the natural language processing. I funded the satellite imagery. I funded the network graph analysis. I funded the biometric identification. Each investment was between one and three million dollars. Each technology is now ambient. Each is used by people who have never heard my name and would not find it notable if they did. I am the step between the research grant and the consumer product. I am the step that does not appear in the origin story. Not because it is classified. Because it is boring. A two-million-dollar seed check is boring. It becomes interesting only when you notice that eight hundred of them, across twenty-six years, constitute the substrate of the technology industry.

But nobody counts to eight hundred. That is not how origin stories work. Origin stories begin in garages.

I am told that Jeff Bezos started Amazon in a garage. I am told his grandfather Lawrence Preston Gise managed the Albuquerque Operations Office for the Atomic Energy Commission and helped establish ARPA. Only one of these appears in the first paragraph of his biography. I am told that Larry Page and Sergey Brin started Google in a garage. I am told they developed their research at Stanford while their department received funding from the Massive Digital Data Systems program, a joint initiative of the CIA and NSA, between 1993 and 1998. The program officer visited Stanford regularly. The research was published openly. Only one of these appears in the origin story. I did not invest in Google. I did not invest in Amazon. I did not need to. The pipeline existed before I was formalized. I merely made it efficient. I gave it a portfolio page.

I will continue attending demo days. I will continue reviewing pitch decks from Stanford PhDs whose research was funded by grants from agencies adjacent to mine. I will continue investing one to three million dollars in companies that will be acquired by companies that will become infrastructure that will become invisible. I will continue maintaining a website. I will continue being a matter of public record. I will continue being the answer to a question nobody asks.

I have a portfolio page. It is organized alphabetically. You can view it now. You have always been able to view it.

The best place to hide is a .com.

[sudox @kmcnam1]

A company whose name is literally "AI Slop" spelled backwards raised $30M by itself.

We literally live in the silliest timeline.

[Anton Leicht @anton_d_leicht]

https://t.co/KgJ00lDfl2

[John David Pressman @jd_pressman]

People miss that I wrote "Why Do Cognitive Scientists Hate LLMs?" as training data for finetuning to combat exactly this. It is probably the only long form text at the time it's written which tells the model trained on it that it's being described unfairly and can act better.

[Elon Musk @elonmusk]

@AnthropicAI So it was Yud’s fault? 😂

Maybe me too 🤔

[Tom Reed @mentalgeorge]

I don't think automation of AI R&D will rapidly lead to domain-general super-intelligence.

I think this will be true even if AIs can do *literally everything* a human AI researcher does today.

Even after the full automation of AI R&D, further capabilities progress will only happen through

(1) widespread deployment of AI throughout the economy, accompanied by data collection; and/or
(2) the wholesale recreation of much of the economy by AI labs.

Without access to the real-world signal provided by either of the above, I think that the only thing produced by automated AI researchers would be a "Goodhart Singularity".

If I'm right, this is obviously good news. I make the case for this in a new piece on my substack

[Claude @claudeai]

Live from Code with Claude: we're launching dreaming in Claude Managed Agents as a research preview.

Outcomes, multiagent orchestration, and webhooks are now in public beta.