Polo M (slow/steady)

Research Scientist. AI, machine learning and tech policy. after hours ai takes. passionate about data security, integrity, and cogsec

[Jessie @jessie_thinker]

>Intelligence may be the navigation of structured spaces

[Bloomberg @business]

The US government should consider the possibility that AI becomes so advanced that it can improve itself without human intervention, and agencies should factor that risk into any federal testing of cutting-edge models, a GOP senator said https://t.co/zWEMWb96Qo

[Greg De Temmerman @Gregdt1]

Cool graph from @Bloomberg showing the evolution of data centre capacity and the planned/announced projects.
Data centres globally now consume close to 500TWh- basically as much as Germany.

https://t.co/coj5s7pyD1

[Reuters @Reuters]

Donald Trump told reporters that his team might buy US stakes in artificial-intelligence companies and said he would host a meeting with AI executives as soon as next week https://t.co/PEEBpte0jN

[Polo M (slow/steady) @traintest_split]

its alreadybhere

[Ben Goldhaber @BenGoldhaber]

Gwern describes current AI training and architecture leading to a slop-pocalypse

[Polo M (slow/steady) @traintest_split]

It may become necessary to move to annual or biannual release schedules, with mandatory prep and assessments occurring in the intervening months, both societally and security wise. Further, energy and build out constraints will start to hit hard next spring

[Yoshua Bengio @Yoshua_Bengio]

If leading AI companies are indeed approaching the point of recursive self-improvement, a coordinated, verifiable, and universally applied pause is probably the only responsible solution to mitigate several major AI risks; at least until safety guarantees are developed and demonstrated. Ensuring that such a moratorium is respected would require sincere collaboration between various countries and companies, but I definitely believe it is achievable if others follow in @AnthropicAI's footsteps.

[Matthew Petti 👨‍🎓 @matthew_petti]

The report seems to imply that Israel was bugging the phones of U.S. troops sent to defend Israel.

[nxthompson @nxthompson]

The biggest leap in AI use cases in 2026: agents, astrology, and fan fic.

[Congressman Bill Foster @RepBillFoster]

We need guardrails now, before these materials end up in the wrong hands.

https://t.co/wJ2RbSvkqS

[Sriram Krishnan @sriramk]

🇺🇸🚀 SOME NEWS: I'll be leaving my role at the White House at the end of this month. After a break I’ll be working on helping tackle some of the large challenges facing America on AI (more on that later).

It is hard to express how big a privilege it has been to serve the American people and how grateful I am to have had the opportunity to do so.

First and foremost, it has been an honor to serve under President @realDonaldTrump . Without his leadership, we would not be leading in the AI race.

Second, I owe a lot to the person I’ve worked mostly closely with over the last 18 months - @DavidSacks . His continuing advocacy for America winning on AI has been and continues to be crucial.

Some key public accomplishments from last year I’m proud of

1. Architecting and publishing the American AI Action Plan - charting the course for America to win on AI and helping execute on that for the last year.

2. The AI acceleration partnerships to help American AI stack win globally.

3. The National AI Policy Framework for Artificial Intelligence executive order (forming the basis for working with the Hill this year)

4. Advocating for the American AI stack with our allies globally (the AI summits in France and India, state visits to the UK, the Middle East and more)

So what’s next?

The past 18 months have given me a front row seat to this critical moment on AI facing America and our allies. Whether it is energy, data centers or a clear path for Americans to experience the benefits of AI, there are many tough issues we all need to navigate together. I plan on building institutions that help tackle some of those challenges for America and its allies.

I want to thank many others who have helped along the way in the administration : Kevin Hassett, @mkratsios47 , CoS @SusieWiles47 , VP @JDVance , @StevenCheung47 , Sec Bessent, Sec Lutnick, Sec Rubio and @jacobhelberg , @USWREMichael , Josh Gruenbaum, Watson Fagan, Ryan Baasch, Jeff Kessler, Alexei Bulazel, DepSec Landau, DepSec Dabar, Will Scharf, Taylor Budowich, @JamesBlairUSA , @elonmusk and many, many others. You know who you are and I know I’ll continue to see you a lot more.

Most of all, I want to thank @aarthir on supporting everything and being part of this unexpected but amazing journey from last January. None of this would be possible without her.

This journey has been the privilege of a lifetime and shown me how special this country is and how it needs all of us to contribute in anyway we can - and I plan on continuing to do just that.

🇺🇸

[Polo M (slow/steady) @traintest_split]

My hot take is that it no longer matters, LLMs make it trivial to reformat legal documents however you want to digest them

[dave kasten @David_Kasten]

One of my most-desired reforms for the federal government is that all EOs, regs, and laws should be required to output their tasks in a list like this, and the status of them should be publicly trackable (barring classification concerns etc.)

[Dan Hendrycks @hendrycks]

Four papers out recently:
1. https://t.co/p1fG3egjYS: Measures and reduces political bias in LLMs; Claude is especially biased
2. https://t.co/MYmRECqJtF: The public can insert backdoors into AIs, creating supply-chain risks; this deters forms of recursive improvement and military use
3. https://t.co/W1uhcPfw18: ASIs can have rational reasons to preserve humans, even when we aren't economically useful
4. https://t.co/vvXxbQWGQ4: AIs increasingly act like they have functional pleasure and pain

[Governor Kathy Hochul @GovKathyHochul]

Offering a bounty on the first bill introduced to ban this dystopian nightmare.

[hardmaru @hardmaru]

Today, we are officially launching the Sakana AI RSI Lab in Tokyo to build open-ended, adaptive AI systems that collectively self-improve. I am incredibly proud of our team’s work over the past 2 years, shipping the breakthrough research that laid the foundations for this moment.

Building in Japan provides us with the ultimate design constraint. Just like Japan’s historical dominance in manufacturing was achieved by fundamentally redesigning the factory floor to do more with less, we are focused on compute-efficiency.

We are not building the most compute-hungry self-improvement engine. We are building the most sample-efficient one.

If you are entirely unsatisfied with the brute-force status quo and ready to build the self-improving future in Japan, come join us.

[Polymarket @Polymarket]

JUST IN: Trump announces he will meet with AI leaders next week to discuss possible U.S. government investment in their companies.

[MTS @MTSlive]

SITUATION DETECTED: President Trump has signed a National Security Memorandum directing the U.S. military and intelligence community to rapidly adopt advanced commercial AI, build high-security compute facilities, and onboard models from multiple vendors onto classified networks.

[Director Michael Kratsios @mkratsios47]

Today, @POTUS signed a National Security Presidential Memorandum on AI in the national security enterprise. The men and women who defend our nation deserve the best, most secure, and most reliable AI in the world, and our citizens deserve to know it is handled responsibly with the care and seriousness they expect.

The NSPM accelerates AI adoption from multiple vendors to prevent single points of failure, updates @DeptofWar’s guidance on autonomy in weapons systems to keep pace with the frontier, and ensures no entity can disable or degrade an AI system our warfighters depend on without prior approval.

Under @POTUS’s leadership, we are putting our AI dominance to work to defend the American people.

[Alex Albert @alexalbert__]

We just published internal data on how much of Claude's development is already being done by Claude:
- Over 80% of all code merged into our codebase is now written by Claude
- It's been months since many researchers at Anthropic hand-wrote code
- The typical Anthropic engineer ships 8x as much code as they did in 2024
- On the most open-ended engineering tasks, Claude's success rate jumped from ~26% to 76% in 6 months
- When research sessions went off-track, Claude proposed a better next step than the human took 64% of the time

We're not at recursive self-improvement yet, but it could come sooner than most expect. I highly recommend reading the full blog post.

[Polo M (slow/steady) @traintest_split]

This pressure will grow in the 2028 primaries. Gamble is do you get ahead of it or concede a little later when pressure is higher.

[Andrew Curran @AndrewCurran_]

NOTUS is reporting that senior administration officials have held discussions with major AI labs - including OpenAI - about the possibility of those labs giving the government equity stakes by voluntarily ceding shares. The returns from those shares could eventually be directly distributed to American citizens through an AI dividend.

[Benjamin Todd @ben_j_todd]

8h at 80% on METR in 2030 seems slow, will be hit within a year at current trends