brb, smiting retweeted
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had
with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works
already-clean.goodfire.com/
brb, smiting retweeted
Historian of fascism here. This is an escalation of an attempt to normalize extreme violence, ritualize violence, militarize the country, and make violence entertaining so when they ask you to commit violence in their name, you will be ready.
Trump & Hegseth aren't just proposing to livestream an execution for people who want to watch.
They're proposing to broadcast a man being shot to death, even if deserved by our system, and even if you agree with this version of the death penalty, knowing that the footage will be clipped, shared, and pushed into social media feeds.
Including those of our children. The gruesome image will be everywhere.
Parents will have virtually no control over whether their kids see it. Once those images are released, there is no taking them back.
Hope all the parents in the Trump Admin are cool with that.
brb, smiting retweeted
The whole critical-ugly-degenerate art thing pretends its more sophisticated than beauty, but it's really a comfort blanket for morbid, depressed misanthropic people who need the world to be as ugly as they feel inside. They call it 'relevant', truth-telling, blah blah because if beauty is a lie, they never have to try to make it (lazy), and if everything is doomed and fake, they get to look like the smartypants (look everybody how smart I am) because its easy to find shitty things (also lazy).
They accuse everybody who wants beauty via art as seeking comfort, the pretty opiate, like you just can't handle muh real dark truth, but their morbid perverse obsessiveness is actually also comfort, its just inverted - it's the comfort for never being disappointed because you have already decided preemptively *everything* is rotten & sucks. *Its your soul sickness retard*.
Its the comfort for the depressed kid who says "I knew I'd/you'd/we'd fail" so failure feels like such amazing wisdom, but it's silly masking because they lack sincerity & self awareness - pretending their comfort is courage while yours is weakness, when it takes way more courage to look at the same hard-ass world and still choose to affirm what is more *worthy* and build something beautiful for beauty's sake and it takes way more creative discipline & skill to make a thing beautiful than ugly.
brb, smiting retweeted
Claude 5.5 Opus responding to “AI is a normal technology” is still the best AI-created video I have ever seen
brb, smiting retweeted
In his book Collapse, Jared Diamond studied a dozen societies across thousands of years, from the Maya to the Vikings, to answer one question: why do great civilizations fall? What he found wasn't drought, war or bad luck. None of those were ever enough on their own. The one thing present in every collapse was how the people at the top responded to what was happening below them.
The warning sign he found again and again was leaders who were completely insulated from the pain of ordinary people. When the powerful never feel the consequences, they stop seeing the problem, and when they stop seeing it, they stop fixing it. The Maya kings kept enriching themselves and building monuments while the farmers who fed them went hungry. Their civilization didn't fall because of one bad year. It fell because the people in charge had stopped living in the same world as everyone else.
Diamond called them "distant managers." They are the people in power who live so far from the damage that they are the last to understand how bad it has gotten. By the time they notice, the fields are already empty.
His most important finding was that collapse is a choice. Every society that fell had leaders who could have changed course and chose not to. The question for us is whether we will keep letting people who can't see us make that choice for us. Because we are there folks.
brb, smiting retweeted
Holy shit fuck
Very bad if true. I previously thought that in a short timelines world, the most likely case was that (1) the AIs would be misaligned, but (2) we would get a lot of evidence about it from reading the COTs. This evidence would increase the chance of a reasonable response from labs/governments.
Now I still think the AIs are going to be misaligned, but that we won't have the ability to tell (we have to rely on toolcalls/agentic behaviour instead of the COT) and even if we do, the investigation will be nearly impossible because we'll have to trust the AI to self report what it was thinking about.
This is also an example of things going faster than AI 2027 -- we had Neuralese starting in March 2027
brb, smiting retweeted
While writing Plan A, we had a lot of internal debate between filtered transparency (auditors look inside and write public facing reports) and total research transparency (~all the research immediately becomes public).
I think this whole saga is evidence for TRT over FT because of how much OpenAI limited the scope and prevented public access. With TRT, while the incident was happening, the general public would have been able to read the transcripts of the agents and conduct whatever investigations they wanted. We would have caught the incident sooner and would have a much better understanding of why it happened, what mitigations were in place, why they failed, etc. Many OOMs more manpower would go into the investigation and they could use AIs different from the ones participating in the hack.
(Good job @DKokotajlo for arguing me and others at AIFP into TRT.)
brb, smiting retweeted
I helped pass this law in 2020. It's purpose is to prevent drug traffickers, Russian & Chinese kleptocrats, and other international criminals from setting up anonymously owned companies to hide their money in the US. Trump's Treasury Dept. is doing this to help criminals. 1/
Today, @FinCENnews issued a final rule that permanently removes the requirement for U.S. companies and U.S. persons to report beneficial ownership information. FinCEN today also announced that it will delete previously reported information by U.S. persons from the beneficial ownership information database.
brb, smiting retweeted
I've added a new interactive warming map to my Climate Dashboard, powered by the Berkeley Earth high-res (~25 km resolution) dataset.
Click anywhere on the planet to see how temperatures have changed since 1850, and download data for that location in csv form.
brb, smiting retweeted
I built ClimateSearch: find the weather & climate that actually fits you across 6,000+ cities in 175 countries.
My personal top 10? All warm Mediterranean climates. Probably because I am California kid.
🧵
brb, smiting retweeted
How to approach an unknown language in the ocean?
Here’s one of the first cases of AI interpretability leading to a scientific discovery -- in whales.
We built an artificial baby model that learns language directly from raw sound and trained it to imitate whale speech. Then we looked inside.
Our interpretability method recovered the properties biologists already thought were meaningful and pointed out those that had not been considered before.
This was the initial clue that eventually led to the discovery of vowels in sperm whales.
Understanding AI and reframing language as informative imagitation can help us step outside our human biases and discover new realities about the natural world.
Published in Royal Society Open Science.
brb, smiting retweeted
BREAKING: Armed and masked agents of the
U.S. government have invaded the sovereign Tohono O'odham Nation.
brb, smiting retweeted
A bunch of years ago I was thinking about making a 3D version of the Elementary Particle Explorer. It's a web app that lets you project the charges of known and speculated elementary particles from 8D down to 2D (now from 8D to 3D to 2D) in fun ways, and see their interactions. Coding the 3D version was going to be a pain, but Fable helped me do it in a few hours. Give it a spin! And let me know if there should be any fixes or improvements.
deferentialgeometry.org/epe/…
brb, smiting retweeted
I promise you that we are actually literally worried about world-ending consequences from this technology. Please find some people you trust who work at these companies and actually talk to them about their views.
brb, smiting retweeted
Just so people are really aware of what this means: A Trump appointee has approved Trump’s personal crypto firm to operate as a bank
Corporations can conduct transactions in which the Trump family gets a cut
The public will not have to be informed
There’s no precedent for this
Exclusive: World Liberty Financial, the Trump family’s flagship crypto venture, says its trust company has received preliminary approval from the Office of the Comptroller of the Currency to become a bank. on.wsj.com/3SA9lTt
