Wi-Fi-connected kitchen appliances have struggled to prove their worth, but Samsung’s not entirely new app could help realize the potential of connected cooking.
WSJ reports on declining usage numbers for ChatGPT, Midjourney, and Jasper, leading to a cooling on venture capital’s unbridled optimism for AI startups. They’re not going anywhere, but maybe our expectations of their value should be tapered. That’s what the people with the money are saying, at least.
[The Wall Street Journal]
But it’s not clear how much actual income the company will make, The Information reports. That’s still a pretty big leap from last year, when OpenAI had only $28 million in revenue. The Information cites a single anonymous source with “direct knowledge,” so that’s someone who thought this was important to put in industry press. Wonder who!
My Assistant, a generative AI-powered feature of its larger Me@Campus app, will be rolled out to Walmart employees working in the company’s different offices in the US. Around 50,000 employees can use it to help summarize large documents or draft questions.
The company would not tell The Verge who it partnered with to built the tool, but said it will be working on bringing more generative AI to employees, including to those working in its many stores in the US and internationally.
I asked Bing the same question and it told me to buy a USB lighter. This is a true story.
Google engineers developed a new method they call SayTap, which they said bridges the gap between a prompt and the precise code that tells a robot dog, or other four-legged robots, where to put its feet down. LLMs generally have a hard time processing such low-level commands given to robots, which usually involve typing in the exact angle for a leg.
Researchers can write a prompt like “back off” and then input a set pattern of numbers that indicate if the legs should be on the ground or lifted up.
[blog.research.google]
Google publicly released its Cloud TPU v5e, a processor specifically built to handle faster training performance for large language models. The company said TPU v5e “delivers up to 2x higher training performance per dollar” compared to the previous TPU v4. Google also announced the newest version of its supercomputer to run more generative AI models. The A3 VM is powered by Nvidia H100 Tensor Core GPUs.
