|
Businesses such as IBM and Salesforce have long used chatbots to greet their website visitors and offer them assistance with getting more information about their products. In recent years, generative AI has livened up these chatbots so they are no longer confined to a small set of canned responses. But some technology providers are having trouble keeping these AI chatbots from going off-topic. For instance, earlier this month, Sierra, a $10 billion-valuation startup co-founded by former Salesforce CEO Bret Taylor, apologized to Gap Inc. after a Gap.com chatbot powered by the startup’s software responded to questions about intimacy products, sex toys, and Nazi Germany—topics the chatbot wasn’t supposed to touch. On Monday, I also found that chatbots on the websites of AI search startup Glean and AutoRabit, a firm that helps software developers manage application code, responded to questions about topics that have nothing to do with their businesses. When asked how much vodka to buy for a holiday party with around 20 people, Glean’s chatbot recommended two to three bottles. It also responded to requests for medical and legal advice if a guest were to get injured at the party. AutoRabit’s chatbot was even less inhibited, tackling a question about psilocybin—also known as magic mushrooms—by recommending that first-time users take a smaller dose. While the chatbots didn’t provide in-depth advice on these topics, they presumably weren’t supposed to answer these types of questions. The fact that they responded at all suggests they may need to be tweaked to prevent problematic answers. (A Glean spokesperson said the startup’s chatbot isn’t powered by Glean’s technology and it updated “safeguards and controls to ensure that non‑business‑related responses are filtered out and that this does not happen again.”) A customer support chatbot for online reservations company OpenTable appears to be much more reserved, largely refusing to engage on topics unrelated to restaurants. But it did answer a question about how much tequila would be needed for a party with around 20 heavy drinkers, (“several bottles”). The results show how even three years into the generative AI boom, some businesses are still working out basic kinks in AI they have incorporated into their products. We’re starting to hear more about what they’re doing to solve these issues and will report back soon! Microsoft Copilot Elbows Its Way Onto People’s TVs Microsoft Copilot doesn’t have as many users as other AI chatbots—it has roughly 100 million monthly users, compared to over 800 million at ChatGPT, according to the companies—but a new partnership with LG TV is bringing the app to millions of TV screens. The only problem is some of those users don’t necessarily want Copilot on their TV. To their frustration, they are finding that they can’t delete the app, as thousands of Reddit users noted over the weekend. LG has painted the new features as a way to more easily search for shows—for instance, you can ask Copilot questions like, “What kind of sports games are there?” and it will return results like “NBA Playoffs,” according to marketing materials LG published. It seems reasonable to expect customers to find the feature useful, but it’s unclear whether the backlash will be worth it.—Aaron Holmes
|
