All pages
- 3Blue1Brown
- AI prepping
- AI safety field consensus
- AI safety is harder than most things
- AI safety is not a community
- AI safety lacks a space to ask stupid or ballsy questions
- AI safety technical pipeline does not teach how to start having novel thoughts
- AI timelines
- AI will solve everything argument against AI safety
- Add the complete proof on proof cards to reduce friction when reviewing
- Agent foundations
- AlphaGo
- AlphaGo Zero
- AlphaGo as evidence of discontinuous takeoff
- Analyzing disagreements
- Anki deck options
- Anki deck philosophy
- Architecture
- Big card
- Bury cards to speed up review
- Busy life periods and spaced inbox
- Carl
- Carl Shulman
- Central node trick for remembering equivalent properties
- Choosing problems for spaced proof review
- Coherence and goal-directed agency discussion
- Comparison of AI takeoff scenarios
- Comparison of sexually transmitted diseases
- Comparison of terms related to agency
- Competence gap
- Content sharing between AIs
- Continuous takeoff keyhole
- Convergent evolution of values
- Corrigibility
- Counterfactual of dropping a seed AI into a world without other capable AI
- DSA
- Dealing with bad problems in spaced proof review
- Deck options for proof cards
- Deconfusion
- Desiderata for cognitive reductions
- Different senses of claims about AGI
- Discontinuities in usefulness of whole brain emulation technology
- Discontinuities in whole brain emulation technology
- Discursive texts are difficult to ankify
- Distortion of true frequency of piece of knowledge due to spaced repetition
- Do an empty review of proof cards immediately after adding to prevent backlog
- Duolingo
- Eliezer
- Empty review
- Encoding dependence problem
- Evolution
- Evolution analogy
- Explorable explanation
- Fake review
- Feynman technique fails when existing explanations are bad
- Finiteness assumption in explorable media
- Flag things to fix during review
- Fractional progress argument for AI timelines
- Future planning
- Generic advice is difficult to give but also important
- Giving advice in response to generic questions is difficult but important
- Goal inference
- Goalpost for usefulness of HRAD work
- HRAD
- Hansonian
- Hardware-driven vs software-driven progress
- Hardware argument for AI timelines
- Hardware overhang
- Highly reliable agent designs
- How doomed are ML safety approaches?
- Human safety problem
- Human safety problems
- Hyperbolic growth
- IDA
- Ignore Anki add-ons to focus on fundamentals
- Improvement curve for good people
- Incremental reading
- Incremental reading in Anki
- Intelligence amplification
- Inter-personal comparison test
- Interacting with copies of myself
- Interaction reversal between knowledge-to-be-memorized and ideas-to-be-developed
- Interaction reversal between knowledge and notes
- Intra-personal comparison test
- It is difficult to find people to bounce ideas off of
- It is difficult to get feedback on published work
- Iterated amplification
- Iteration cadence for spaced repetition experiments
- Jessica Taylor
- Jonathan Blow
- Kanzi
- Kasparov Window
- Kasparov window
- Keyhole argument against continuous takeoff
- Laplace's rule of succession argument for AI timelines
- Late singularity
- Linked list proof card
- List of big discussions in AI alignment
- List of breakthroughs plausibly needed for AGI
- List of critiques of iterated amplification
- List of disagreements in AI safety
- List of experiments with Anki
- List of men by number of sons, daughters, and wives
- List of success criteria for HRAD work
- List of teams at OpenAI
- List of technical AI alignment agendas
- List of techniques for managing working memory in explanations
- List of terms used to describe the intelligence of an agent
- List of thought experiments in AI safety
- List of timelines
- Lumpiness
- Lumpy
- MIRI
- MIRI vs Paul research agenda hypotheses
- Main Page
- Make new cards when you get stuck
- Managing micro-movements in learning
- Mass shift to technical AI safety research is suspicious
- Meta-execution
- Minimal AGI vs task AGI
- Missing gear
- Missing gear for intelligence
- Missing gear vs secret sauce
- Mixed messaging regarding independent thinking
- My current thoughts on the technical AI safety pipeline (outside academia)
- My take on RAISE
- My understanding of how IDA works
- Narrow window argument against continuous takeoff
- Nate
- Newcomers in AI safety are silent about their struggles
- Nobody understands what makes people snap into AI safety
- One-sentence summary card
- One wrong number problem
- Ongoing friendship and collaboration is important
- Online question-answering services are unreliable
- Output curve for good people
- Paperclip maximizer
- Parasitizing on popularity
- Paul
- Paul christiano
- People are bad
- People watching
- Personhood API vs therapy axis of interpersonal interactions
- Politicization of AI
- Popularity symbiosis
- Probability and statistics as fields with an exploratory medium
- Progress in self-improvement
- Prosaic AI
- Quick review
- Quotability vs ankifiability
- RCG
- RSI
- Rapid capability gain vs AGI progress
- Realism about rationality
- Resource overhang
- Reverse side card for everything
- Richard Ngo
- Science argument
- Second species argument
- Secret sauce for intelligence
- Secret sauce for intelligence vs specialization in intelligence
- Selection effect for successful formalizations
- Selection effect for who builds AGI
- Short-term preferences-on-reflection
- Simple core
- Simple core algorithm
- Simple core algorithm for agency
- Simple core algorithm of agency
- Simple core of consequentialist reasoning
- Slow feedback cycle for spaced repetition
- Slow feedback loop for spaced repetition
- Soft-hard takeoff
- Soft takeoff keyhole
- Something like realism about rationality
- Spaced everything
- Spaced inbox
- Spaced proof review
- Spaced proof review as a way to invent novel proofs
- Spaced proof review as a way to understand key insights in a proof
- Spaced proof review is not about memorizing proofs
- Spaced proof review routine
- Spaced repetition
- Spaced repetition allows graceful deprecation of experiments
- Spaced repetition and cleaning one's room
- Spaced repetition as generator of questions
- Spaced repetition as soft alarm clock
- Spaced repetition experiments take months to complete
- Spaced repetition prevents unrecalled unrecallables
- Spaced repetition world
- Spaced writing inbox
- Spoiler test of depth
- Statistical analysis of expert timelines argument for AI timelines
- Stupid questions
- Tao Analysis I exercise count
- Terms used to describe intelligence of agents
- Test
- The Hour I First Believed
- The Precipice notes
- The Secret of Psalm 46
- The Secret of Psalm 46 outline
- The Uncertain Future
- The Witness
- The mathematics community has no clear standards for what a mathematician should know
- There is pressure to rush into a technical agenda
- There is room for something like RAISE
- Timeline of my involvement in AI safety
- Toby Ord questions
- UDASSA
- Unbounded working memory assumption in explanations
- Uncertain Future
- Understanding is discontinuous
- Uninsightful articles can seem insightful due to unintentional spaced repetition
- Unreliability of online question-answering services makes it emotionally taxing to write up questions
- Use paper during spaced repetition reviews
- Using Anki for math
- Value learning
- WBE
- Website to aggregate solutions to textbook exercises
- Wei
- Wei Dai
- Whole brain emulation
- Why ain'tcha better at math
- Will there be significant changes to the world prior to some critical AI capability threshold being reached?
- Yudkowskian
