70: Open-source AI models performed just as well — or better — than proprietary models at solving complex medical problems, according to a new study by Harvard researchers published on Friday. Notably, Meta’s Llama model correctly diagnosed patients 70% of the time as opposed to OpenAI’s GPT-4, which did so only 64% of the time. This signals that the gap between open- and closed-source models, with the former being largely free to use and customizable, is closing.
7.5 billion: The surveillance company Flock Safety raised $275 million in a new funding round Thursday. The round was led by Bedrock Capital that values Flock, which uses computer vision — a type of artificial intelligence — at $7.5 billion. The firm not only sells to private businesses, but also police departments, sparking concerns from civil liberties advocates such as the ACLU.
30: California’s state legislature is currently considering 30 bills about AI. The proposals are varied: one requires human drivers in autonomous vehicles while another mandates comprehensive safety testing. The state legislature could try to rein in the largely California-based technology sector just as the Trump administration rolls back Biden-era guardrails on AI.
31.8: South Korean chip sales to China dropped 31.8% in February, the second straight month of plummeting sales to the country, according to South Korea’s Ministry of Trade, Industry and Energy. This dropoff among Korean companies such as Samsung and SK Hynix could be attributable to new US export controls initiated at the end of the Biden administration, which restrict the sales of advanced semiconductors — even those using US parts — to China.
25: AI systems are very, very bad at reading clocks, according to new research from the University of Edinburgh, which tested models’ understanding of visual inputs. The AI systems tested correctly read analog clocks less than 25% of the time.
