limits

Not Just Bikes explains how North American speed limits are set and why posted limits matter far less than street design. It argues that drivers follow the road's cues, so safe speeds have to be built into the street.

Lex Fridman's long podcast interview with Yann LeCun on Meta AI, open source, the limits of LLMs, AGI and the future of AI.

Kurzgesagt explains why, even with science-fiction technology, most of the universe is permanently out of reach because of cosmic expansion and finite resources.

Yann LeCun on the Big Technology Podcast discusses why current AI models cannot make discoveries despite their knowledge. He argues systems need abstract world models and also covers limits of AI research, investor expectations, and open source after DeepSeek.

Physicist and YouTuber Angela Collier argues in this hour-long video that AI as commonly described does not really exist. She covers the limits of the tools, biased datasets, and why they should not be used to make decisions.

Alex O'Connor shows that ChatGPT cannot draw a wine glass filled to the brim, and ties the failure to David Hume's missing shade of blue. It asks what the failure says about how image models learn.

Stuff Made Here builds a device that lets a blind person perceive their surroundings by touch, using an iPad's lidar and a mechanical tactile display driven by motors. The video walks through the build and the limits he hit.

Associate Professor Simon Michaux presents estimates of the metals needed to build one generation of renewable technology to replace fossil fuels. He argues current production and reserves fall far short, with copper a key shortfall.

Dwarkesh Patel interviews Richard Sutton, reinforcement learning pioneer, Turing Award winner and author of The Bitter Lesson. Sutton argues LLMs are a dead end because they cannot learn on the job, and that continual learning from experience will require a new architecture.

Jeremy Howard's 90-minute practical tour of language models. It covers how they work, GPT-4's capabilities and limits, using the OpenAI API, code writing, data analysis and fine-tuning.

Evan Czaplicki, creator of Elm, examines how programming languages are funded. He looks at where compiler engineers' salaries come from, why Go has about 5 engineers while Dart has about 30, and how this limits language design.

Veritasium's explainer on Godel's incompleteness theorems and the limits of mathematical proof, tracing set theory, Hilbert's program and the road to Turing and modern computing.

Herb Sutter's CppCon 2022 talk on whether C++ can become ten times simpler and safer. He introduces his experimental Cpp2 syntax and the cppfront transpiler, explaining the limits on incremental improvement.

Veritasium explains why analog computing may suit AI workloads, covering old analog computers, The Analog Thing and Mythic's analog chips that use flash memory for neural network inference. It argues digital hardware is hitting limits.

In an Ars Technica War Stories episode, Prince of Persia creator Jordan Mechner describes how he made the game on the Apple II, fighting hardware memory limits and the platform's decline. It covers the rotoscoped animation and the engineering tricks involved.

John Oliver's Last Week Tonight segment on PFAS, the 'forever chemicals' linked to health problems. It covers their widespread use in products like nonstick coatings and repellent fabrics and the limits of regulation.

Benno Rice's linux.conf.au 2020 talk asks whether UNIX was the best design we could have had. He looks at what alternatives existed, what was gained and lost, and what could be re-learned, including the limits of treating everything as a file.

Tantacrul walks through how the MuseScore 4 team was built and how design decisions were made for the overhaul of a huge legacy notation app. It covers playback and engraving quality improvements.

Veritasium investigates PFAS forever chemicals, from Teflon's history to pollution of the global water system and their presence in the US population. It covers the corporate cover-up and the host's own blood tests.

Martin Kleppmann surveys current research on CRDTs, the algorithms for merging concurrent edits across offline replicas. He covers the hard parts such as rich text, moves and merge results that are technically correct but poor.

John McCall of Apple's Swift team presents at CppNow 2023 on introducing a memory-safe successor language into large C++ code bases. He argues incremental improvement of C++ hits hard limits and explains Swift's C++ interoperability for gradual migration.

Lex Fridman interviews physicist Roger Penrose on consciousness, the limits of computation, quantum mechanics and his cyclic cosmology. It covers his views on AI and the big bang.

Sabine Hossenfelder weighs the arguments for and against nuclear power, covering waste, fuel supply and newer reactor designs. She is skeptical that fast breeders and thorium solve the problems.

Sabine Hossenfelder explains that waste heat from ever-growing energy use could eventually boil the oceans in a few centuries, whatever the energy source. She discusses proposed ways to cool the planet.

Asianometry explains why Russia cannot replace TSMC after sanctions halted chip shipments. It looks at Russian chip designers such as Baikal, MCST and Yadro, and at Russia's lagging domestic manufacturing capacity.

Linus Tech Tips opens up an M1 MacBook Air to test whether better thermal pads could improve performance in the fanless design. It looks at what Apple left on the table.

UC Berkeley EECS colloquium by OpenAI's John Schulman on reinforcement learning from human feedback. He covers why RL helps with truthfulness, hallucination, and the limits of behavior cloning.

debuglive makes a 1990-style dance track on an Amiga 500 with a Stereo Master sampler and cheap secondhand records, showing how sampling and trackers worked within tight memory limits.

PBS Space Time considers whether life could be based on silicon instead of carbon, comparing the two elements' chemistry and why silicon is the most plausible alternative.

Kurzgesagt asks whether advanced civilizations could have existed on Earth in the deep past and left little evidence. It looks at time as a dimension for searching for alien civilizations and the limits of the geological record.

EposVox reviews a very cheap USB 2.0 HDMI capture dongle sold under names like BlueAVS/Goodan, testing whether it works as an alternative to the Elgato Cam Link. He covers its quality and limits for webcam-style and camera capture use.

Jim Keller, CTO of Tenstorrent, gives a keynote at Stanford's System X conference on change, understanding and the future of computing. He touches on Moore's Law, how far transistors are from physical limits, and ideas like DNA origami.

Allan Jude of Klara presents work on making ZFS scale on fast NVMe storage, covering fixes to the TaskQueue interface and removal of bottlenecks. The talk was given at EuroBSDcon 2022 and describes raising write throughput from about 3 to over 7 GB/sec.

Boston Dynamics introduces Stretch, a prototype robot that moves boxes in warehouses and distribution centers. Its mobile base lets it unload trucks, build pallets and assemble orders.

Dashcam-style footage of a Tesla Model 3 crashing into an overturned truck on a highway in Taiwan during the morning commute.

Dwarkesh Patel interviews OpenAI chief scientist Ilya Sutskever on time to AGI, what comes after generative models, data limits, alignment, and competing with Google. The title centers on whether next-token prediction could surpass human intelligence.

The video argues that machine-learning-based aimbots, which read the screen and drive the controls externally, could make cheating in FPS games undetectable. It explains how conventional aimbots work, how the new approach differs, and speculates on possible detection.

Machine Learning Street Talk interviews Paul Lessard about Categorical Deep Learning, which uses category theory and monads to make neural network architectures more interpretable and composable. They also discuss neural networks' limits in reasoning and unbounded computation.