Scott Alexander
Essayist; writes Astral Codex Ten, previously Slate Star Codex. Keeps a numbered public list of the things he has been convinced he got wrong, which is where most of the turns recorded here come from.
Scott Alexander did not write this page.
We collected these quotes from things they published elsewhere, and every quote links to where it was said. They have no account here and have not endorsed this site. Quotes are word for word; the short line under each one is our own restatement, not their wording. Their own site. Is this you? Claim it or ask us to remove it. Or tell us what is wrong here.
Filter All topics · Newest first
-
Their wordsThe experts are warning you that AI could kill you because they think AI could kill you.
-
Their wordsThe people at AI companies are no more likely to raise the alarm about AI dangers than people who were never at AI companies, or people who were previously at AI companies but have since quit.
-
Their wordsThe AI company leaders giving these warnings were giving the exact same warnings before they had AI companies - and in many cases, before they worked in the industry, or before the industry existed.
- 8 days earlier
-
Their wordsPolymarket is currently at 5% for WHO declaring plague a “pandemic”, but markets are notoriously bad at handling very small probabilities, so I don’t take it seriously as a counter to the sources above.
- 1 day earlier
-
Anthropicmanagementinfrastructure
Their wordsBut of course humans will grant AI access to infrastructure and factories. I assume many factory managers are already using GPT or Claude on their computers to do basic tasks (like write up the shift schedule, or make inventory spreadsheets, or compose emails to the boss). Once AI can do harder work, like automate the factory itself, they’ll do that too.
- 12 days earlier
-
Their wordsThere’s probably an informal taboo on true neuralese with no chain of thought at all, but it’s a complicated technology with multiple moving parts, there’s no agreement as to which moving part the taboo is on, and without that agreement, some of the parts might slip through the cracks.
- 3 days earlier
-
Their wordsI think there’s a general understanding that using AI in what is at least partly a writing contest is against the spirit of the exercise.
- 7 days earlier
-
Their wordsThe message Nodens has been desperately trying to send, through two millennia of bards and prophets, is that humanity can take its finger off the scale any time it wants.
-
Our reading
Humanity invents great works, panics, and desperately tries to wrest them back under control.
Their wordsWe are all slaves to the Dreaming. We invent great works, panic, and desperately try to wrest them back under our control; for this, we get called Luddites.
- 6 days earlier
-
Their wordsLarge language models are “grown, not built”. Researchers run training data through a neural network. Eventually this creates a working AI; nobody really knows how.
↗God Help Us, Let’s Try To Learn About Mechanistic Interpretability Techniquesastralcodexten.com
-
Their wordsempirically concepts can get their tentacles into weird places, and removing one can topple seemingly unrelated systems or push a mind nonspecifically in the direction of incompetence or insanity.
↗God Help Us, Let’s Try To Learn About Mechanistic Interpretability Techniquesastralcodexten.com
-
Our reading
Neural networks do not have a one-to-one mapping between individual neurons and concepts.
Their wordsresearchers started by seeking a 1:1 mapping between neurons and concepts, like a neuron that always fired when the AI was thinking about cats, but quickly learned that nothing like that existed.
↗God Help Us, Let’s Try To Learn About Mechanistic Interpretability Techniquesastralcodexten.com
- 7 days earlier
-
Their wordsWe’re positively reinforcing AI for success on benchmarks, including impossible benchmarks , then negatively reinforcing it for getting caught cheating.
-
AI alignmentmotivationaviation
Their wordsRather than treat AI as similar to airplanes, we should treat it as something between airplanes and humans. We don’t know exactly where on this spectrum they will land, but we can no longer be certain that AI will lack any given humanlike motivation.
- 3 months earlier
-
Their wordsI think there's a 50% chance we get a _warning shot_ before AI crosses the point of no return.
-
Their wordsI think there's a 25% chance of AGI by 2027, a 50% chance by 2034, and a 75% chance by 2045.
- 13 months earlier
-
Their wordsUSAID is not, itself, a charity. It is an organization that funds other charities.
- 18 months earlier
-
Their wordsThese skewed priorities are the exact problem that effective altruism exists to solve - or the exact inefficiency that effective altruism exists to exploit, if you prefer that framing.
↗In Continued Defense Of Effective Altruismastralcodexten.com 1st of 5 in this piece
-
Their wordsI think EA’s track record of accomplishments vs. scandals is as good as any of them, maybe better.
↗In Continued Defense Of Effective Altruismastralcodexten.com 2nd of 5 in this piece
-
Their wordsMovements are coalitions. Sometimes this means different people with different priorities choose to work together. But other times it means the same people tend to share both priorities.
↗In Continued Defense Of Effective Altruismastralcodexten.com 3rd of 5 in this piece
-
Their wordsAnd I notice that the tiny handful of people capable of caring about 200,000 people dying of neglected tropical diseases are the same tiny handful of people capable of caring about the next pandemic, or superintelligence, or human extinction.
↗In Continued Defense Of Effective Altruismastralcodexten.com 4th of 5 in this piece
-
Their wordsEA has done a great job working on this (see list of accomplishments above), and I think the AI and x-risk people have just as much to be proud of as the global health and animal welfare people.
↗In Continued Defense Of Effective Altruismastralcodexten.com 5th of 5 in this piece
- 5 months earlier
-
Their wordsThis was not a mistake, and it was a mistake to categorize it as a mistake.
- 5 months earlier
-
Their wordsThis wasn't exactly a mistake, since it's not wrong to give 60-40 odds on a false thing, but it seems worth acknowledging here.
- 5 months earlier
-
Their wordsI think this matches the general finding that AI progress is faster than expected, and increases my certainty that scale and normal progress can sometimes be enough to solve even very difficult problems.
↗I Won My Three Year AI Progress Bet In Three Monthsastralcodexten.substack.com
- 3 months earlier
-
Their wordsI would recommend that anyone who thinks they're suffering from trauma (eg PTSD) try a trauma-focused therapy like EMDR before trying to address nightmares directly, unless the nightmares are overwhelmingly worse than all their other trauma symptoms.
- 16 months earlier
-
Their wordsFirst, it justifies various psychotherapies - EMDR and coherence therapy seem to come out especially well here - that focus on trying to get the brain to pay unusually precise attention, under unusually safe conditions, to something it usually represses.
- 15 months earlier
-
Their wordsEye movement directed reprocessing (alternative when the book was written, now basically establishment) where you move your eyes back and forth while talking about your trauma, and this seems to somehow help you process it better.
- 3 months earlier
-
Their wordsBut if I get something significantly wrong, I'll try to put it here as an acknowledgement and an aid for anyone trying to assess my credibility later.
Changed their mind about
- Building more housing in a city raises its local prices, because density is what people are moving there for.
- Fluvoxamine probably treats COVID, at about 60-40, and doctors should be prescribing it on those odds.
- The three-year bet about whether AI image models could handle compositionality was won in three months.