Loading activity…
@aaroth
CS prof at Penn. Amazon Scholar at AWS. Author of The Ethical Algorithm (w/ Michael Kearns). I study machine learning, privacy, game theory, and uncertainty.
AI
One post argues that calling LLMs incapable of thought is unscientific without a rigorous definition of thinking and evidence that their mechanisms fundamentally differ from human cognition.
AI
A researcher urges junior researchers to hone their writing and think carefully about what individual words mean and evoke in readers.
AI
A research thread describes a condition weaker than requiring reviewer agents to match your utility exactly—and cautions that it is not guaranteed to hold.
AI
A research thread claims that, under a non-negative span condition, every Nash equilibrium leaves the principal no worse off than baseline—even when the driver agent and reviewers optimize their long-run discounted payoffs.
AI
A thread describes a condition weaker than requiring each reviewer to share the principal’s exact utility. Under it, the author claims every Nash equilibrium leaves the principal no worse off than baseline.
AI
A user studying reviewer agents defines safety as guaranteeing that the user’s utility—what they want to maximize—is no worse than under a baseline policy.
AI
For a simple one-decision model, a thread claims safety holds exactly when the principal’s utility is a weighted sum of reviewer utilities with no negative weights.
AI
A thread describes a one-decision model where safety holds exactly when the principal's utility is a non-negative weighted combination of reviewer utilities, then extends the characterization to long-running agents.
AI
A user questions the reassurance of having another agent grant permission: why should the reviewer deserve more trust than the agent doing the work?
AI
A user describes Codex and Claude Code's auto-approval as asking another agent for permission—and questions why that reviewer should be trusted.
AI
The user describes computationalism as the idea that human cognition obeys the same laws of computation as silicon, but stresses that it remains a conjecture.
AI
Worst-case complexity has been an “awful guide” to machine learning, the user argues, pointing to the difficulty of learning even a linear classifier.
AI
The critique centers on worst-case complexity, which the writer calls an “awful guide” in machine learning.
AI
A post on theoretical computer science predicts that AI could shift scarce intellectual work toward formulating questions, models, theories, definitions and conjectures.
AI
One post argues that “essentially zero direct progress” on proving P ≠ NP leaves advanced AI little to build on, unless P = NP.
