Group items tagged

"Transistors use grooves etched in silicon to guide electrons around the chip. The channels do a similar job to that of wires, but on a much smaller scale. Making these grooves just 7nm wide means you can fit more transistors on the chips. For comparison a strand of human hair, at 100,000nm thick, is about 10,000 times wider than the channel. A red blood cell is a thousand times bigger, at 7,500nm in diameter. A strand of DNA is in the same order of magnitude, but slightly smaller at just 2.5nm wide."

"In recent months, people have shared digitally altered sexual images of the new deputy prime minister Angela Rayner and celebrities including Taylor Swift. But you don't need to be famous to appear in one of these images or videos - the technology is readily accessible, and can easily be used by ex-partners or strangers to humiliate and degrade. As a tech luddite, I was still under the impression that one needed some digital skills to commit this kind of abuse. Not so. You can simply take someone's image, put it into a "nudify" app, and the app's AI will generate a fake nude picture. "It's quick and easy to create these images, even for anyone with absolutely no technical skills," Jake Moore, an adviser at a cybersecurity firm, told me."

"The turning point happened in 2005, when the transistors, while continuing to double in numbers, were neither faster nor more energy-efficient at the same rates as before."

"The gate, a chip component that switches transistors on and off, is a critical measure of transistor size. Previous research had already pushed gate lengths to one nanometer and below. By scaling gate lengths down to the size of single atoms, the latest work sets a new mark that'll be hard to beat. "In the future, it will be almost impossible for people to make a gate length smaller than 0.34 nm," the paper's senior author Tian-Ling Ren told IEEE Spectrum. "This could be the last node for Moore's Law.""

"Training a massive AI model, the size of GPT-4, would currently take about 8,000 H100 chips, and 15 megawatts of power, Nvidia said - enough to power about 30,000 typical British homes."

"On a zoomed-out time scale, technological progress follows an exponential curve. Compare how the world looked 15 years ago (no smartphones, really), 150 years ago (no combustion engine, no home electricity), 1,500 years ago (no industrial machines), and 15,000 years ago (no agriculture). The coming change will center around the most impressive of our capabilities: the phenomenal ability to think, create, understand, and reason. To the three great technological revolutions-the agricultural, the industrial, and the computational-we will add a fourth: the AI revolution. This revolution will generate enough wealth for everyone to have what they need, if we as a society manage it responsibly. The technological progress we make in the next 100 years will be far larger than all we've made since we first controlled fire and invented the wheel. We have already built AI systems that can learn and do useful things. They are still primitive, but the trendlines are clear."

"Our estimate of the length of tasks that an agent can complete depends on methodological choices like the tasks used and the humans whose performance is measured. However, we're fairly confident that the overall trend is roughly correct, at around 1-4 doublings per year. If the measured trend from the past 6 years continues for 2-4 more years, generalist autonomous agents will be capable of performing a wide range of week-long tasks."