David Valerio

I wrote about the Epistle to Diognetus. discern.earth/p/in-the-world…

@AndyMasley How do you mean planting trees a fake climate intervention? The fact that most projects fail or the principle of planting them at all?

Separate from the water stuff I thought we all got the message that planting trees is a fake climate intervention

Interesting that you didn’t listen to him when he told you to your face not to do something.

Finance people always try to scare you into letting them gamble with your money instead of you doing it yourself because "you're going to lose all your money day-trading against Wall Street wizards and geniuses" when in reality they already outsourced trading to ChatGPT.

@UnintendedCons5 Amazing

lol

OpenAI released ChatGPT as a research preview on November 30th 2022 (1230 days ago). On that day: • NVIDIA had a market cap of under $400 billion - today it is over $5 trillion. • Anthropic had been founded (by OpenAI alumni) for about a year but did not yet have any revenue. Today it has an annual revenue run rate of $30 billion. • ChatGPT 3.5 could read simple code and occasionally provide helpful comments. Today, Anthropic Mythos (and likely soon OpenAI Spud) can complete end to end cybersecurity tasks that would take an expert human dozens of hours. • Sam Altman and his family were not being viciously attacked by maniacs with guns and molotov cocktails. Now they horrifically are. • AI Policy was mostly a niche technical issue in Washington DC. Today, urgent issues or potential crisis related to AI regularly reach the desk of the most senior level principals in the US government (e.g. Chairman of the Federal Reserve Jerome Powell and Secretary of the Treasury Scott Bessent convening leaders of top banks to discuss AI cybersecurity risks last week, or the Anthropic DOW conflict). • The version of GPT used for the ChatGPT release was trained with about 10^23 FLOP of training compute. Claude Mythos may been trained on as much as 10^27 FLOP (10,000x more computations). (this is an extremely incomplete list, curious to hear other examples of striking comparisons) What will things look like in AI world 1230 days in the future from today?

For the would-be Great Man of History, the final boss is always the pope.

Politics makes for interesting bedfellows as the usually virulent critics of Catholic moral teaching on human sexuality and marriage, rally to defend the pope. We live in very strange days.

Maybe a very obvious point, but it only just struck me that, so far at least, intelligence seems to require four valence electrons.

“A non-writing writer is a monster courting insanity.” ― Franz Kafka

I think that over the next five years we are likely to see both substantial progress toward something like 'weak AGI', i.e. systems that can do most cognitive tasks humans can do, and growing diminishing returns to raw frontier model improvement in the economic sense. The point isn't that scaling stops working but rather that (a) achieving each additional increment of capability at the model level will require disproportionately greater expenditure of compute, data, engineering effort, and capital; and (b) 'weak AGI' will probably come from the combination of strong models with scaffolding, tools, memory, retrieval, planning, decomposition, verification, and other system-level affordances around them. As a result, deployment design and scaffolding becomes more important over time, not less. The old view that wrappers are disposable because the next model jump will wash them away seems naive. If frontier gains become more input-intensive, then the question increasingly becomes how much capability you can extract, route, verify, and compose from a model within a given budget. Recent developments point in this direction too. What seems to matter is whether a given pipeline is structured to exploit capabilities well: assigning subproblems appropriately, using division of labour intelligently, and compensating for weaknesses with tools and process. It seems quite plausible that there's more alpha on the harness side at the moment, than in merely betting on scaling alone.

@ThomasWayneRil1 Love this part of the country! Just outstanding to gaze upon.

The Painted Desert.

@deredleritt3r Why do you think it would be more risky to develop in this manner?

The events of the past few days have pushed us closer to a world where AGI is developed in an underground bunker owned and operated by the U.S. government. This would be a bad world - one in which the risks posed by powerful AI would be significantly more pronounced. I hope that humanity, in its collective wisdom, would find a way to steer clear from such an outcome.

I confidently predict Pope Leo will react to the Trump tweet exactly the same way I did, and respond accordingly.

I guess it’s time to pick a side

meu "gênero" literário favorito tem sido "fábula onírica sobre terras remotas permeada de lirismo melancólico e um senso de estagnação ancestral ocasionalmente perturbado por vagos pressentimentos de algum cataclismo sinistro" preciso de um nome mais curto só. e recomendações!

These particular words of Pope Leo called out to me greatly today: "We must not allow ourselves to be overwhelmed by the pace of a world that does not know what it is chasing. Rather, we must return to serving the rhythm of life, the harmony of creation and healing its wounds."

@rSanti97 As usual, the answer to how to do things is to just do them. Walther’s piece was a great kick in the butt.

@rSanti97 Very true re: discipline I have not read him in this topic but am looking forward to doing so shortly!

I'm unemployed for the next three days AMA

Going down the rabbit hole of vintage giant cacti photos

A few months ago my friend @nabeelqu asked me for a reading list. It turned into a bit of an adventure. Should I just list my "favorite" books? Or the ones that influenced me personally the most? Or the ones I deem most "important" in some broader sense? I deliberated uselessly for months. The result is a fun little project I call "Coutellerie." Michel Foucault once wrote that "knowledge is not made for understanding; it is made for cutting." Very few ideas animate my work and thinking more than this one. Coutellerie is a French word that describes the manufacturing, trade, and appreciation of knives. The list of books and essays I ultimately created for Nabeel, and now share with you, is not necessarily a list of my favorites, or the ones I deem most influential. Instead it is a collection of the sharpest blades I have found after 20 years of searching. It is the knowledge in these books that I have found most amenable to the hard work of cutting. I hope they are useful to you too. I'll probably update the list over time. It'll live permanently in the header of my low-volume personal blog, copied below.