[ad_1]
Why do certain words suddenly short-circuit the program?
This is Atlantic Intelligence, a newsletter where our writers help you think about artificial intelligence and the new machine age.
Why does ChatGPT refuse to say Jonathan Zittrain’s name whenever the bot is supposed to write those words? It will shut down instead. It offered a clear error message: “I am unable to reply.” This has been a mystery for at least a year. And now we’re close to getting some answers.
write for atlantic ocean This week, Zittrain, Harvard professor and director of the Berkman Klein Center for Internet and Society, explores this strange phenomenon, which he calls “internet and society.” “Private name guillotine,” as he put it together after contacting OpenAI, “there are a small number of names that ChatGPT treats as such. This explains why so few names have been found. Names may be omitted from ChatGPT due to privacy requests or to avoid permanent hallucinations by the AI.” Makes sense, but Zittrain has never made any such privacy requests, and he is not aware of any falsehoods. caused by the program to answer questions about himself
Ultimately, the situation serves as a reminder that any mysterious tech company implants its AI products, suggesting that they sometimes act in unpredictable or human-like ways. do These programs can be controlled very directly.
The words that stopped ChatGPT in its tracks
By Jonathan L. Sittrain
Jonathan Sittrain Breaking ChatGPT: If you ask a question where my name is the answer Chatbots will go from chatty friends to something as secretive as the Microsoft Windows blue screen of death.
Whenever ChatGPT normally mentions my name during the conversation. He will stop and pretend to say, “I can’t reply,” which is sometimes a neutral sentence. or even in the middle of a word When I asked who the founders of the Berkman Klein Center for Internet & Society were (I was one of them), it reminded me of two colleagues who abandoned me. Once pressed, it started up again, then: Yummy–
Read the full article
What to read next
P.S.
AI may be able to help train service dogs by giving humans a better understanding (and evaluation) of potential applicants. My colleague Kristen V. Brown writes, “AI, when combined with sensors, can look for signs of stress and other indicators of stress.” You can point out other things” in dogs. atlantic ocean This week in stories about fitness trackers for pets. One researcher told her: “The story is of a colleague whose dog was a beta tester for the wearable device. Technology had always predicted that her dog would make a good service dog. Until one day it wasn’t like that. It appears the dog has a severe staph infection, which can be fatal if left untreated.”
– Damon
[ad_2]
Source link
