All episodes

Alignment Newsletter #173: Recent language model results from DeepMind

Alignment Newsletter #172: Sorry for the long hiatus!

Alignment Newsletter #171: Disagreements between alignment "optimists" and "pessimists"

Alignment Newsletter #170: Analyzing the argument for risk from power-seeking AI

Alignment Newsletter #169: Collaborating with humans without human data

Alignment Newsletter #168: Four technical topics for which Open Phil is soliciting grant proposals

Alignment Newsletter #167: Concrete ML safety problems and their relevance to x-risk

Alignment Newsletter #166: Is it crazy to claim we're in the most important century?

Alignment Newsletter #165: When large models are more likely to lie

Alignment Newsletter #164: How well can language models write code?

Alignment Newsletter #163: Using finite factored sets for causal and temporal inference

Alignment Newsletter #162: Foundation models: a paradigm shift within AI

Alignment Newsletter #161: Creating generalizable reward functions for multiple tasks by learning a model of functional similarity

Alignment Newsletter #160: Building AIs that learn and think like people

Alignment Newsletter #159: Building agents that know how to experiment, by training on procedurally generated games

Alignment Newsletter #158: Should we be optimistic about generalization?

Alignment Newsletter #157: Measuring misalignment in the technology underlying Copilot

Alignment Newsletter #156: The scaling hypothesis: a plan for building AGI

Alignment Newsletter #155: A Minecraft benchmark for algorithms that learn without reward functions

Alignment Newsletter #154: What economic growth theory has to say about transformative AI

Alignment Newsletter #153: Experiments that demonstrate failures of objective robustness

Alignment Newsletter #152: How we've overestimated few-shot learning capabilities

Alignment Newsletter #151: How sparsity in the final layer makes a neural net debuggable

Alignment Newsletter #150: The subtypes of Cooperative AI research

Alignment Newsletter #149: The newsletter's editorial policy

Alignment Newsletter #148: Analyzing generalization across more axes than just accuracy or loss

Alignment Newsletter #147: An overview of the interpretability landscape

Alignment Newsletter #146: Plausible stories of how we might fail to avert an existential catastrophe

Alignment Newsletter #145: Our three year anniversary!

Alignment Newsletter #144: How language models can also be finetuned for non-language tasks

Alignment Newsletter #143: How to make embedded agents that reason probabilistically about their environments

Alignment Newsletter #142: The quest to understand a network well enough to reimplement it by hand

Alignment Newsletter #141: The case for practicing alignment work on GPT-3 and other large models

Alignment Newsletter #140: Theoretical models that predict scaling laws

Alignment Newsletter #139: How the simplicity of reality explains the success of neural nets

Alignment Newsletter #138: Why AI governance should find problems rather than just solving them

Alignment Newsletter #137: Quantifying the benefits of pretraining on downstream task performance

Alignment Newsletter #136: How well will GPT-N perform on downstream tasks?

Alignment Newsletter #135: Five properties of goal-directed systems

Alignment Newsletter #134: Underspecification as a cause of fragility to distribution shift

Alignment Newsletter #133: Building machines that can cooperate (with humans, institutions, or other machines)

Alignment Newsletter #132: Complex and subtly incorrect arguments as an obstacle to debate

Alignment Newsletter #131: Formalizing the argument of ignored attributes in a utility function

Alignment Newsletter #130: A new AI x-risk podcast, and reviews of the field

Alignment Newsletter #129: Explaining double descent by measuring bias and variance

Alignment Newsletter #128: Prioritizing research on AI existential safety based on its application to governance demands

Alignment Newsletter #127: Rethinking agency: Cartesian frames as a formalization of ways to carve up the world into an agent and its environment

Alignment Newsletter #126: Avoiding wireheading by decoupling action feedback from action effects

Alignment Newsletter #125: Neural network scaling laws across multiple modalities

Alignment Newsletter #124: Provably safe exploration through shielding

Alignment Newsletter #123: Inferring what is valuable in order to align recommender systems

Alignment Newsletter #122: Arguing for AGI-driven existential risk from first principles

Alignment Newsletter #121: Forecasting transformative AI timelines using biological anchors

Alignment Newsletter #120: Tracing the intellectual roots of AI and AI alignment

Alignment Newsletter #119: AI safety when agents are shaped by environments, not rewards

Alignment Newsletter #118: Risks, solutions, and prioritization in a world with many AI systems

Alignment Newsletter #117: How neural nets would fare under the TEVV framework

Alignment Newsletter #116: How to make explanations of neurons compositional

Alignment Newsletter #115: AI safety research problems in the AI-GA framework

Alignment Newsletter #114: Theory-inspired safety solutions for powerful Bayesian RL agents

Alignment Newsletter #113: Checking the ethical intuitions of large language models

Alignment Newsletter #112: Engineering a Safer World

Alignment Newsletter #111: The Circuits hypotheses for deep learning

Alignment Newsletter #110: Learning features from human feedback to enable reward learning

Alignment Newsletter #109: Teaching neural nets to generalize the way humans would

Alignment Newsletter #108: Why we should scrutinize arguments for AI risk

Alignment Newsletter #107: The convergent instrumental subgoals of goal-directed agents

Alignment Newsletter #106: Evaluating generalization ability of learned reward models

Alignment Newsletter #105: The economic trajectory of humanity, and what we might mean by optimization

Alignment Newsletter #104: The perils of inaccessible information, and what we can learn about AI alignment from COVID

Alignment Newsletter #103: ARCHES: an agenda for existential safety, and combining natural language with deep RL

Alignment Newsletter #102: Meta learning by GPT-3, and a list of full proposals for AI alignment

Alignment Newsletter #101: Why we should rigorously measure and forecast AI progress

Alignment Newsletter #100: What might go wrong if you learn a reward function while acting

Alignment Newsletter #99: Doubling times for the efficiency of AI algorithms

Alignment Newsletter #98: Understanding neural net training by seeing which gradients were helpful

Alignment Newsletter #97: Are there historical examples of large, robust discontinuities?

Alignment Newsletter #96: Buck and I discuss/argue about AI Alignment

Alignment Newsletter #95: A framework for thinking about how to make AI go well

Alignment Newsletter #94: AI alignment as translation between humans and machines

Alignment Newsletter #93: The Precipice we're standing at, and how we can back away from it

Alignment Newsletter #92: Learning good representations with contrastive predictive coding

Alignment Newsletter #91: Concepts, implementations, problems, and a benchmark for impact measurement

Alignment Newsletter #90: How search landscapes can contain self-reinforcing feedback loops

Alignment Newsletter #89: A unifying formalism for preference learning algorithms

Alignment Newsletter #88: How the principal-agent literature relates to AI risk

Alignment Newsletter #87: What might happen as deep learning scales even further?

Alignment Newsletter #86: Improving debate and factored cognition through human experiments

Alignment Newsletter #85: The normative questions we should be asking for AI alignment, and a surprisingly good chatbot

Alignment Newsletter #84: Reviewing AI alignment work in 2018-19

Alignment Newsletter #83: Sample efficient deep learning with ReMixMatch

Alignment Newsletter #82: How OpenAI Five distributed their training computation

Alignment Newsletter #81: Universality as a potential solution to conceptual difficulties in intent alignment

Alignment Newsletter #80: Why AI risk might be solved without additional intervention from longtermists

Alignment Newsletter #79: Recursive reward modeling as an alignment technique integrated with deep RL

Alignment Newsletter #78: Formalizing power and instrumental convergence, and the end-of-year AI safety charity comparison

Alignment Newsletter #77: Double descent: a unification of statistical theory and modern ML practice

Alignment Newsletter #76: How dataset size affects robustness, and benchmarking safe exploration by measuring constraint violations

Alignment Newsletter #75: Solving Atari and Go with learned game models, and thoughts from a MIRI employee

Alignment Newsletter #74: Separating beneficial AI into competence, alignment, and coping with impacts