Get low (ldl cholesterol’s model)
An experimental genetic therapy referred to as VERVE-101 can deactivate a cholesterol-raising gene in folks with hypercholesterolemia, Meghan Rosen reported in “Base enhancing can decrease ldl cholesterol” (SN: 1/27/24, p. 8).
Rosen wrote that researchers are testing to see what dosage of VERVE-101 is handiest. On condition that the therapy edits a gene, reader Linda Ferrazzara questioned why the dose issues.
Too low a dose might imply that not sufficient VERVE-101 makes it to the liver, the place it turns off the gene, Rosen says. If too few cells have the gene switched off, sufferers won’t expertise the drug’s cholesterol-lowering results. If levels of cholesterol stay excessive after an preliminary therapy, a second infusion of the drug might assist, Rosen says. However the builders want for the therapy to be one dose.
Reader Jack Miller requested whether or not VERVE-101 impacts germ cells, which give rise to sperm and egg cells.
In mice, scientists have discovered that many of the drug results in the liver, and none goes to the germ line, Rosen says. The offspring of handled mice are additionally unaffected by the drug. So if the kids of handled sufferers even have hypercholesterolemia, these youngsters would nonetheless want their very own therapy, she says.
AI etiquette
To develop higher safeguards, scientists are learning how folks have tricked AI chatbots into answering dangerous questions that the AI have been educated to say no, akin to learn how to construct a harmful weapon, Emily Conover reported in “Chatbots behaving badly” (SN: 1/27/24, p. 18).
Reader Linda Ferrazzara questioned if AI chatbots have been educated on languages aside from English.
AI chatbots like ChatGPT are primarily based on giant language fashions, or LLMs, a sort of generative AI usually educated on huge swaths of web content material. Lots of the largest, most succesful LLMs proper now are tailor-made to English audio system, Conover says. Though these LLMs have some capacity to put in writing in different languages, most of their coaching knowledge is in English. However there are language fashions designed to make use of different languages, she says. Efforts up to now have centered on languages which can be extensively spoken and written, and for which giant quantities of coaching knowledge can be found, akin to Mandarin.
Ferrazzara additionally requested if boosting computing energy may assist the bots higher defend in opposition to trickery.
LLMs already use a variety of computing energy, and it’ll solely enhance as folks use LLMs an increasing number of, Conover says. However even when elevated energy would make establishing safeguards simpler, we have to acknowledge that greenhouse gasoline emissions linked to such energy-intensive calculations contribute to local weather change, she says. “The time and vitality wanted to reply to a chatbot question will not be one thing we need to overlook whereas ready for computer systems to enhance.”
Lots of the defensive strategies described within the story display screen for dangerous questions. Reader Mike Speciner questioned if filtering the responses to these questions can be simpler.
Some filters like this are already utilized on some chatbots, Conover says. For instance, Microsoft’s Bing AI tends to chop off its solutions if it wades into forbidden territory. These filters are extra normal, moderately than focused particularly at one form of assault. “To keep away from letting unhealthy stuff slip via, they might solid too broad of a web, filtering out innocuous solutions in addition to harmful ones and making the consumer’s expertise worse,” Conover says. What’s extra, an attacker who is aware of how the LLM’s self-filtering works might determine a technique to idiot that filter.
Correction
“Saving lives with secure injection” incorrectly described Elizabeth Samuels of UCLA as an epidemiologist and emergency drugs doctor (SN: 2/10/24, p. 16). She is an emergency and habit drugs doctor. That story additionally mistakenly said that drug coverage guide Edward Krumpotich helped write the 2023 laws in Minnesota that approved funding for an overdose prevention middle. He advocated for that laws however didn’t assist write it.
