It's pretty easy to see the problem here: The Internet is brimming with misinformation, and most large language models are trained on a massive body of text obtained from the Internet.
Ideally, having substantially higher volumes of accurate information might overwhelm the lies. But is that really the case? A new study by researchers at New York University examines how much medical information can be included in a large language model (LLM) training set before it spits out inaccurate answers. While the study doesn't identify a lower bound, it does show that by the time misinformation accounts for 0.001 percent of the training data, the resulting LLM is compromised.
While the paper is focused on the intentional "poisoning" of an LLM during training, it also has implications for the body of misinformation that's already online and part of the training set for existing LLMs, as well as the persistence of out-of-date information in validated medical databases.
© Just_Super
On Friday, the US Department of Agriculture (USDA) announced that it would begin a nationwide testing program for the presence of the H5N1 flu virus, also known as the bird flu. Testing will focus on pre-pasteurized milk at dairy processing facilities (pasteurization inactivates the virus), but the order that's launching the program will require anybody involved with milk production before then to provide samples to the USDA on request. That includes "any entity responsible for a dairy farm, bulk milk transporter, bulk milk transfer station, or dairy processing facility."
The ultimate goal is to identify individual herds where the virus is circulating and use the agency's existing powers to do contact tracing and restrict the movement of cattle, with the ultimate goal of eliminating the virus from US herds.
At the time of publication, the CDC had identified 58 cases of humans infected by the H5N1 flu virus, over half of them in California. All but two have come about due to contact with agriculture, either cattle (35 cases) or poultry (21). The virus's genetic material has appeared in the milk supply and, although pasteurization should eliminate any intact infectious virus, raw milk is notable for not undergoing pasteurization, which has led to at least one recall when the virus made its way into raw milk. And we know the virus can spread to other species if they drink milk from infected cows.
© Credit: mikedabell
Cancer incidence is on the rise, particularly among younger adults, but most new drug compounds fail to progress through clinical trials. According to French entrepreneur Fanny Jaulin (on the right in the picture above), the problem lies with clinical trial design. “Just because a drug isn’t successful in clinical trials doesn’t mean it’s not a […]
© 2024 TechCrunch. All rights reserved. For personal use only.
