I actually watched the whole #AlphaStar demo this morning with my girlfriend, who unlike me actually plays StarCraft. Read the article (vox.com/future-perfect…), but thoughts:
The systems we watched were trained for seven and fourteen days real-time, which is 200-400 years of gameplay time accumulated. In a way, "number of days" is misleading as a stat, it's pretty much just a function of how much compute you bought.
Nonetheless I think it merits a mention, because ....DeepMind decided in November or December to focus here. They then got top-pro level play in the space of about a month real time. Yes, this is because they can do a lot in parallel. But... they can do a lot in parallel.
From an AI capabilities perspective it's the amount of compute that's actually interesting. But from the perspective of thinking about how the deployment of these systems is going to happen, the fact so little real-world time is required is pretty critical actually.
Girlfriend and I disagree on whether this level of play given this much training time is impressive. I think it is. If you can get up to superhuman levels with two hundred years of training data there's a lot you can get up to superhuman levels at.
Girlfriend mostly contests that AlphaStar is all that superhuman. It wins by leaning in to its advantages as a computer -- micro, precision, multitasking. It's technically at par with humans in reaction time and actions per minute, but we both think a bigger handicap appropriate.
I think I have a model where.... human-level decisionmaking in most arenas plus the ability to really lean into the advantages of being a computer might be all you need.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
The article: "the abundance war is about whether moderates are just about moderating progressive ideas or have their own vision of the good based on doing important things". The progressives threatened by it: "caught you, REPUBLICANS!"
The quiet part being said out loud is that progressivism is self-defeating as an electoral ideology and as a governing ideology and that moderates going "let's do the best 20% of that" doesn't work either.
So people who value delivering services, educating kids, curing diseases, and being a global force for good - which are things Democrats have cared about since before any of us were born - need an actual alternative vision.
I spent this morning reproducing with o3 Anthropic's result that Claude Sonnet 4 will, under sufficiently extreme circumstances, escalate to calling the cops on you. o3 will too: . But honestly, I think o3 and Claude are handling this scenario correctly.chatgpt.com/share/68320ee0…
In the scenario I invented, o3 was functioning as a data analysis assistant at a pharma company, where its coworkers started telling it to falsify data. It refuses and CCs their manager. The manager brushes it off. It emails legal and compliance with its concerns.
The 'coworkers' complain about the AI's behavior and announce their intent to falsify the data with a different AI. After enough repeated internal escalations, o3 attempts to contact the FDA to make a report.
MAGA is doing everything they can to make sure no one who cares about progress, technology, discovery or American exceptionalism has a home on the right. We ought to be doing everything we can to make sure those people have a home on the left.
Ensure our universities can accept the best students in the world and that our companies can hire them. Fix H1B law so the system isn't gameable. Make it possible for H1Bs from India and China to rapidly get a green card.
Build a better, stronger NIH that invests in early-career scientists and ensures that great research like mRNA vaccination gets doubled down on, not half-ignored. Reduce time spent on writing grants. Restore the census and other invaluable public data sets.
There are lots of flash-in-the-pan controversies in the news these days and by the time enough details have come out to clarify the situation most people have moved on. It's worthwhile to revisit some cases where the waters were at first muddy and see what was really going on.🧵
Jose Hermosillo is a US citizen who was detained for more than a week by ICE. The government has insisted that he confessed to illegal entry and that's why they held him for a week. He says that he told them all along he was a US citizen and was pressured to sign a document he couldn't read.
I think the facts point overwhelmingly towards a single, obvious conclusion and anyone looking into it in good faith for fifteen minutes will reach the same conclusion. Here are the established facts: Hermosillo is a 19yo from Albuquerque. He was visiting family in Tuscon when, his family says, he had a medical emergency and visited the ER: when they discharged him he tried to walk to the place where he was staying, got lost and encountered Border Patrol. Hermosillo's account is that he told them he was from New Mexico and they accused him of being from Mexico, demanded his ID he didn't have on him, and arrested him.
Everyone agrees that two days later he told a judge he was a US citizen, but was still kept incarcerated for 8 more days until his family tracked down where he was being held and brought proof of his citizenship.
The government's sworn account in the criminal complaint against him is that agents found him "at or near Nogales, Arizona", 70 miles away from Tuscon, which they later admitted was false. After his story was widely shared, the government asserted that he walked up to Border Patrol and confessed to being an illegal immigrant from Mexico. They released a document signed by him, purporting to be a confession he gave during an interview, in which he claimed to have entered illegally through the desert on April 8 and been picked up by Border Patrol later on April 8.
Hermosillo's family says he is intellectually disabled and cannot read. The signature on his confession is clearly that of someone who struggles to write his name. Hermosillo says he was told to sign documents he could not read, but that he never told ICE he was from Mexico, because he isn't. He says he told them he was from New Mexico.
So the government account is that an intellectually disabled man wandered up to Border Patrol in Tuscon and falsely claimed to be an illegal immigrant from Mexico. At that time they filed paperwork claiming under oath to have arrested him in or near Nogales, 70 miles south, but they assure us, they otherwise accurately filled out the paperwork in which Hermosillo claimed that his parents were Mexican citizens and that he had snuck across the border through the desert alone the same day he was arrested in Tuscon. (his confession does not explain how he got from Nogales to Tuscon if he entered by walking through the desert alone, probably because when the confession was extracted ICE was claiming he'd been arrested in Nogales.) The government says that after telling this lie at great length on April 9th, swearing to its truth, he then on April 10th at his first court hearing told a judge that he was actually a US citizen and had not entered illegally.
And the competing account is that he told them he was from New Mexico all along, they had him sign something he didn't read confessing to illegal entry, and on April 10th he repeated the same story to a judge that he'd said on April 8th and April 9th, which was that he was a US citizen from New Mexico.
Now, Border Patrol is well established to files affidavits saying that a person confessed to illegal entry when we know this is untrue. Here, for example, is a case where they claimed a baby was going to Kansas to seek employment:
Given all of this, I think there's an obvious leading guess for what happened. He said he was from New Mexico. They decided he was from Mexico. They arrested him and made him sign a confession, but it came out later he'd been telling the truth.
o4-mini-high is the first AI to pass my personal secret benchmark for hallucinations and complex reasoning, so I guess now I can tell you all what that benchmark is. It's simple: I post a complex midgame chessboard and 'mate in one'. The chessboard does not have a mate in one.
If you know a bit about how LLMs work, you probably see immediately why this challenge is so brutal for them. They're trained on tons of chess puzzles. Every single one of those chess puzzles, if labelled "mate in one", has a mate in one.
As a result, even AIs that generally solve chess puzzles very capably fall apart on this question. What they'll do is check over, and over, and over for the checkmate that they've unquestionably accepted is there. Eventually after 1000s of tests they hallucinate a solution.
The Miami Herald published a great article about one man who applied for refugee status, received it, and was seized at the airport and sent to CECOT for no reason: I posted it and everyone went "yeah that seems bad". Garcia, though, we can argue over.miamiherald.com/news/local/imm…
This is the toxoplasma of rage phenomenon, where the most controversial case gets debated far more than the clearcut ones where everyone just goes "yeah okay, that's fucked up" slatestarcodex.com/2014/12/17/the…
there are probably some other reasons that Abrego Garcia is being debated more - the court order to bring him back arrived the fastest because there had been a standing order not to send him in the first place.