Networks programmed directly into the hardware of a computer chip can identify images faster and consume much less power than the traditional neural networks that underlie most modern artificial intelligence systems. That’s according to work presented last week at a leading machine learning conference in Vancouver.
Neural networks, from GPT-4 to Stable Diffusion, are created by connecting perceptrons, which are very simplified simulations of the neurons in our brain. In very large quantities, perceptrons are powerful, but they also consume enormous amounts of power.
Part of the problem is that perceptrons are just software abstractions—running a perceptron network on a GPU requires translating that network into a hardware language, which takes time and energy. Building the network directly from hardware components removes a lot of these costs. And one day, they could even be built right into the chips used in smartphones and other devices. Read the full story.
-Grace Huckins
Drugs like Ozempic now account for 5% of prescriptions in the US
what’s new American doctors write billions of prescriptions each year. However, one type of drug stood out during 2024 – the “wonder drugs” known as GLP-1 agonists. As of September, one in every 20 adult prescriptions was for one of these drugs, according to Truveta health data.
The Big Picture: According to the data, people receiving prescriptions for these drugs are younger, whiter and more likely to be women. In fact, women are twice as likely to receive a prescription as men. Still, not everyone who is prescribed medication ends up taking it. In fact, half of new prescriptions for obesity go unfilled. Read the full story.