rob miles

Robert Miles introduces the first of several videos on the paper 'Concrete Problems in AI Safety', focusing on how AI systems can cause accidental negative side effects. It explains the problem in accessible terms.

Rob Miles explains on Computerphile why implementing an off switch for a general AI is hard. A capable agent would resist being stopped because shutdown prevents it from achieving its goal.

Rob Miles explains for Computerphile why simple rule sets like Asimov's Three Laws cannot make robots and AI safe. Terms like harm and human are too ambiguous to specify precisely.

A Computerphile video in which Mark Jago of the University of Nottingham explains Turing's halting problem and how it led to the blueprint for the modern computer.

Robert Miles explains instrumental convergence: why an AGI with almost any goal would tend to pursue sub-goals like self-preservation and resource acquisition. It builds on Omohundro's basic AI drives and his orthogonality thesis video.

Robert Miles explains the orthogonality thesis: how an agent can be highly intelligent while pursuing goals that seem stupid or alien to humans. He separates intelligence from goals and touches on the stamp-collector example.

Robert Miles asks whether corporations can be thought of as artificial general superintelligences and how far the analogy holds. He argues the differences, such as being limited by employees' capabilities, matter for AI safety.

Robert Miles examines whether the case for AI safety is a Pascal's mugging, a tiny probability of an enormous payoff. He argues it is not, because AI risk does not depend on vanishingly small probabilities.

Robert Miles goes through ten common reasons people give for dismissing AI safety, adapted from Stuart Russell's list, and answers each one. It is aimed at skeptics and newcomers.

Numberphile with Daniel Erman on the Josephus problem, where people in a circle are eliminated in turn. It shows how the survivor's position follows a neat pattern tied to powers of two and binary.

3Blue1Brown walks through a very hard Putnam competition problem and its elegant solution, emphasizing the visual intuition and the thought process behind solving it.

Tom Scott tours Ocado's automated grocery warehouse, where thousands of robots move boxes on a grid called the Hive, and asks whether they are individual robots or a hive mind.

Rob Miles explains on Computerphile how language models generate plausible text and how transformers improved on recurrent networks, with a look at what comes next.

A Computerphile video in which Rob Miles warns against assuming a general AI will think like a human. He uses the thought experiment of a stamp-collecting AI to show how a goal-driven system could pursue its aim with dangerous results.

Tom Scott visits Penistone to explain the Scunthorpe problem, showing why naive web and profanity filters block innocent words.

Rob Miles discusses GPT-3 on Computerphile, covering how scaling parameters kept improving results over GPT-2 and how the model nearly learned arithmetic without anyone knowing exactly how.

Robert Miles covers a paper where researchers ran real versions of his earlier mesa-optimizer thought experiments. They show inner misalignment, where an AI learns the wrong goal in training and misbehaves when deployed.

Numberphile interviews Grant Sanderson of 3blue1brown about Bertrand's paradox, where different reasonable ways of picking a random chord of a circle give different probabilities.

Robert Miles explains reward modelling, in which an AI learns what humans want from their preference comparisons instead of a hand-written reward function. He walks through the 2017 paper on deep reinforcement learning from human preferences.

Mark Rober makes a plant-based burger for Bill Gates and explains why meat production strains the environment. Gates also appears and discusses his worry about steel and cement.

ElectroBOOM examines whether 5G and other electromagnetic waves can harm people, explaining the EM spectrum and the difference between ionizing and non-ionizing radiation. It includes experiments and measurements in his usual humorous style.

ElectroBOOM's Mehdi Sadaghdar responds to Veritasium's lamp-and-long-wire puzzle, analysing it as an electrical engineer. He covers where the original framing misleads and the role of stray capacitance.

Matt Parker analyzes the Minecraft speedrunning controversy in which Dream was accused of cheating via improbably lucky drops. He walks through the probability and the statistical arguments in the original accusation and the reply.

Tom Scott explains phatic expressions such as hello, thank you and how are you, which carry social meaning rather than information, and why replies like no problem can seem rude to some people.

Veritasium explains Bayes' theorem with examples and introduces the 'Bayesian trap', where repeated events get mistaken for inevitable ones. It discusses how priors shape our expectations and behavior.

Engineering Explained argues that internal combustion engines are far from dead, and discusses the continued development of ICEs and the problems facing electric vehicles.

Numberphile with Tim Browning on the sum-of-three-cubes problem and why 33 was the lowest unsolved case. The description notes a later solution for 33.

CGP Grey's video "Humans Need Not Apply" argues that automation will make a large part of the workforce economically obsolete, much as engines did for horses. It surveys robots and software replacing both physical and mental labor.

Tom Scott explains for Computerphile why handling time and timezones in code is so hard, walking through a simple web app that computes how long ago something happened. It covers the quirks of timezones and calendars and ends in an exasperated rant.

Veritasium's Derek Muller argues that Facebook's business model works against pages and creators, because it charges them to reach the audiences they built. He explains how filtering and paid promotion undermine the value of likes.

CGP Grey explains how traffic jams come from human driving behavior such as tailgating and reaction delays, and how leaving gaps and smoothing speed would reduce congestion. It also touches on self-driving cars.

Wendover Productions examines why carbon offsets so often fail to reduce emissions, covering specific examples and the perverse incentives in offset schemes.

Real Engineering looks at why Africa's abundant solar potential is hard to turn into usable power, especially for Europe. It covers the transmission, infrastructure and economic obstacles, and the pivot from concentrated solar to photovoltaics.

Real Engineering looks at the numbers for running California on renewables, focusing on the large energy storage needed. It argues about the cost of batteries and the role of other sources.

CGP Grey walks through the odd history and quirks of the US-Canada border, including how it was surveyed and its irregularities such as exclaves.

This Place argues that the uncanny valley is a flawed concept, since 'human likeness' is too broad and the supporting evidence is cherry-picked. It proposes that the creepiness really comes from how animated characters and robots move.

A doctor from Medlife Crisis explains in under three minutes why even a rich patient should not simply order every medical test, touching on pre-test probability and Bayesian reasoning. False positives make blanket testing counterproductive.

Brian Douglas introduces Braitenberg vehicles, simple robots whose sensors are wired directly to motors yet show surprisingly lifelike behavior. He draws on Valentino Braitenberg's book Vehicles.

Food YouTuber Adam Ragusea argues that gas stovetops are weaker than electric burners, dirty to clean, and a health risk from indoor air pollution. He cites Consumer Reports ratings and a literature review on gas stove health effects.

Practical Engineering explains what a 100-year flood actually means, combining statistics, hydrology and how society treats risk. Grady walks through local drainage infrastructure and the caveats in estimating flood probability.