This class is about a fatal trap that lies in wait for the brilliant but unwary young person: success; other people’s definitions of success. I have seen your CVs. They are too long. You are in danger. What do we mean by success “Excellence”: Ordinary success. Pejorative. Default. extrinsic rewards. money, status, fame, Ivy League degree, fancy title, proximity to power “Greatness”. What do you…
Lesson plan Setup Start with philosophy of science to motivate the basic prob theory. Interactive: what do we want from a theory of science? coherence; answers the actual question \(P(H \mid E)\); generality; quantified uncertainty and graded confirmation; falsification = \(LR \to 0\); reproducibility; mechanisms for the intuitions (total evidence, order & stopping-rule invariance, surprise,…
Society is unlikely to fall suddenly under the spell of the transhumanist worldview. But it is very possible that we will nibble at biotechnology’s tempting offerings without realizing that they come at a frightful moral cost… Modifying any one of our key characteristics inevitably entails modifying a complex, interlinked package of traits, and we will never be able to anticipate the ultimate…
This is the editorial for this year’s “ Shallow Review of AI Safety ”. (It got long enough to stand alone.) Epistemic status: subjective impressions plus one new graph plus 300 links. Huge thanks to Jaeho Lee, Jaime Sevilla, and Lexin Zhou for running lots of tests pro bono and so greatly improving the main analysis. tl;dr Informed people disagree about the prospects for LLM AGI – or even just…
Politically, I am a tragedian: It is tragic that we need governments, that we can’t trust markets to deliver what everyone needs. It is tragic that we need markets, that we can’t just trust government to deliver what everyone needs (let alone what they want) and to not crush the Other. It is tragic that we need corporations, that economies of scale are so important that we must risk monopolies and…
The best Chinese LLMs offer frontier performance on some benchmarks; massive per-token discounts (~3x on input, ~6x on output); the weights . On-prem with fully free ~MIT licence, self-hosting, white-box access, customisation, with zero markup (and in fact zero revenue going to the Chinese companies); with a bit of work you can get much faster token speeds than the closed APIs; less overrefusal…
Bostrom (2014) defined the AI value loading problem as how could we get some value into an artificial agent, so as to make it pursue that value as its final goal? 1 JD Pressman thinks this is obviously solved in current LLM systems: The value loading problem outlined in Bostrom 2014 of {getting a general AI system to internalize and act on “human values” before it is superintelligent and therefore…
In what ways can we can fail to answer a question? (I mean necessarily fail: actual barriers to knowledge, rather than skill issue hurdles. But of course contingent failures are much more common: “We didn’t ask the question in the first place”, or “We didn’t have the particular insight that would have allowed for productive research”, or “We didn’t manage to remove every cognitive bias”, or…
He picks from uncountably many sets simultaneously without an algorithm. He preserves us from sets of measure zero and the non-measurables . In his loving arms infinity minus infinity equals whatever we need it to. He takes five loaves of spheres and two small spheres and feeds the host . He makes the inaccessible cardinals. He stops the inaccessible cardinals from accessing us He is risen, he is…
Time… burns without leaving ashes — Elsa Triolet Case study: whither Google Search? For now, Google is the premier information infrastructure on Earth. It is the most-visited website. It handles 14 billion search queries a day, nearly two for everyone on earth. 1 As of writing it’s more than the traffic of ChatGPT , Claude, Wikipedia, Bing, and Duckduckgo put together. 1 / 500 of your waking life…