Imagine that your company just hired a rising star in the executive suite who is so hot that a rival company just hired someone who looks just like him or her. There is a lot of excitement around them. From the CEO to the shareholders, everyone seems to agree that this person is the future of the whole business.
Then you find out that the executive has what people politely call a "hallucination problem." Every time they speak, there is a 15–20% chance that they will just make things up (Opens in a new tab). A professor at Princeton says that he makes up nonsense (Opens in a new tab). They can't even tell fact from fiction (Opens in a new tab). In five minutes, they will go on stage to show off a new product. Do you still bring them to the front of the line?
This week, the answer was yes for Microsoft and Google. OpenAI's ChatGPT, an AI chatbot, has 100 million monthly active users two months after it was released. Microsoft held a surprise event at the last minute to announce that OpenAI would add ChatGPT-style search to the Bing search engine and Edge browser. Google announced Bard, its own AI search tool, the day before. It showed it off at an event in Paris the next day, but it had its own hallucination problem.
"A new race starts today," Satya Nadella, the CEO of Microsoft, told reporters who had been called to the company's campus in Redmond, Washington, on Tuesday. Yes, it's nice to think that way, isn't it? Microsoft, the tech company that is never cool, would like you to think that Bing, sorry, "the New Bing," is competing with Google search on everything. http://sentrateknikaprima.com/
The first thing Google said about Bard(Opens in a new tab) was condescending: "We changed the way the company works around AI six years ago," wrote Sundar Pichai, the CEO of Google.
The "hallucination problem" and Google
Which is interesting. Google, which is the best search engine in the world, has had years to add AI, but its ChatGPT competitor, Bard, is still only in beta with a small group of testers. Even though Pichai acted like a hipster, the Bard reveal was a bit of a mess. All the chatter about ChatGPT seems to have caught Google off guard, too.
How else can you explain the embarrassing Bard mistake that was on full display at the launch? It wasn't at the event itself, where demo mistakes are expected, but in a GIF that was made ahead of time. A user is shown asking Bard for information about the James Webb Space Telescope that he can tell his 9-year-old son.
One of these "facts" is not true: the JWST did not take the first picture ever of an exoplanet. Bard was having a dream (Opens in a new tab). (UPDATE: A reporter for the Financial Times says Bard's words were technically correct, but that requires a way of reading the language that no human would ever use. This is just one more problem with AI search.)
So it makes sense that Alphabet's stock price dropped as much as 8% on the day that Bard came out. Google brought up the main problem with AI search and suggested that the company can't use its huge amount of data to fact-check itself.
Google should know better since it had a "hallucination problem" with its featured snippets at the top of search results back in 2017. The snippets algorithm seemed to enjoy lying about U.S. presidents in particular. What could possibly go wrong?
See also: Don't believe just one source: How to learn anything new online with the new rules
In other words, if you start using your AI search tool too soon, you risk playing yourself. At its launch event, Microsoft was lucky in that there were no obvious mistakes that people could see. But if ChatGPT-based search didn't have a lot of mistakes, why would it still be in beta? Note: There is a sign-up sheet if you want to test AI for Bing without getting paid (Opens in a new tab). https://ejtandemonium.com/
When Wired asked Microsoft's Sarah Bird about ChatGPT's hallucinations, she said, "There's still more to do there." Her job title, "Head of Responsible AI," says it all (Opens in a new tab). The 15 percent hallucination number came from a company that is competing to build a fact-checker for ChatGPT (Opens in a new tab). (UPDATE: A breathless New York Times columnist's report on New Bing showed that it couldn't even do basic math or give a list of kid-friendly things to do in the area.)
Bird also said that in older versions of the software, users could use it to plan a school shooting, but that this feature had been turned off. Interesting! What could go wrong after that? Surely, this crazy beta search product doesn't have any other unintended effects that could embarrass a big and legally weak tech company.
Clippy. Zune. New Bing.
Microsoft knows what it's like to be embarrassed, since it's the company that made Clippy, one of the worst software ideas ever. The paperclip assistant was known for giving advice that no one wanted. ChatGPT isn't like Clippy in the sense that we ask questions when we use it.
But because it often hallucinates its answers or, more often than you might think, gives users a simple version of "I can't answer that," ChatGPT-enabled Bing could be like Clippy on LSD. If enough casual users of "New Bing" get mixed-up results, that will be what it is known for.
It doesn't matter if a product gets better over time; what makes it a joke is how people react to it at first. Since it made the Zune, Microsoft should also know that. It's the same thing to put out a ChatGPT product before it's really ready.
"The New Bing" sounds like it was made to be a joke, to be honest. Or are you really ready to give up Google Search and your Chrome browser for Bing and Edge if Bing wins the AI search race, whatever "winning" really means here? Not a chance. Tech inertia is a force that is way too underrated.
ChatGPT is impressive in some situations, like when real estate agents use it to write listings, but it scares people in other situations. But when you look past the headline, every story about how it caused trouble seems less important. It will cause a lot of students to copy from other sources. Except that it can tell you when a paper was written by ChatGPT, which makes it less of a threat. It passed a test for law school! Except that it barely passed with a C-plus (Opens in a new tab).
Here's the thing: it's really hard to build "general AI," which is the digital version of a human brain. We haven't even started to reach the "insect intelligence stage," which is another long-term goal for AI. Will you really trust ChatGPT to give you your search results instead of, you know, clicking on links yourself?
The answer could very well depend on how bad your own hallucinations are, dear reader.
Then you find out that the executive has what people politely call a "hallucination problem." Every time they speak, there is a 15–20% chance that they will just make things up (Opens in a new tab). A professor at Princeton says that he makes up nonsense (Opens in a new tab). They can't even tell fact from fiction (Opens in a new tab). In five minutes, they will go on stage to show off a new product. Do you still bring them to the front of the line?
This week, the answer was yes for Microsoft and Google. OpenAI's ChatGPT, an AI chatbot, has 100 million monthly active users two months after it was released. Microsoft held a surprise event at the last minute to announce that OpenAI would add ChatGPT-style search to the Bing search engine and Edge browser. Google announced Bard, its own AI search tool, the day before. It showed it off at an event in Paris the next day, but it had its own hallucination problem.
"A new race starts today," Satya Nadella, the CEO of Microsoft, told reporters who had been called to the company's campus in Redmond, Washington, on Tuesday. Yes, it's nice to think that way, isn't it? Microsoft, the tech company that is never cool, would like you to think that Bing, sorry, "the New Bing," is competing with Google search on everything. http://sentrateknikaprima.com/
The first thing Google said about Bard(Opens in a new tab) was condescending: "We changed the way the company works around AI six years ago," wrote Sundar Pichai, the CEO of Google.
The "hallucination problem" and Google
Which is interesting. Google, which is the best search engine in the world, has had years to add AI, but its ChatGPT competitor, Bard, is still only in beta with a small group of testers. Even though Pichai acted like a hipster, the Bard reveal was a bit of a mess. All the chatter about ChatGPT seems to have caught Google off guard, too.
How else can you explain the embarrassing Bard mistake that was on full display at the launch? It wasn't at the event itself, where demo mistakes are expected, but in a GIF that was made ahead of time. A user is shown asking Bard for information about the James Webb Space Telescope that he can tell his 9-year-old son.
One of these "facts" is not true: the JWST did not take the first picture ever of an exoplanet. Bard was having a dream (Opens in a new tab). (UPDATE: A reporter for the Financial Times says Bard's words were technically correct, but that requires a way of reading the language that no human would ever use. This is just one more problem with AI search.)
So it makes sense that Alphabet's stock price dropped as much as 8% on the day that Bard came out. Google brought up the main problem with AI search and suggested that the company can't use its huge amount of data to fact-check itself.
Google should know better since it had a "hallucination problem" with its featured snippets at the top of search results back in 2017. The snippets algorithm seemed to enjoy lying about U.S. presidents in particular. What could possibly go wrong?
See also: Don't believe just one source: How to learn anything new online with the new rules
In other words, if you start using your AI search tool too soon, you risk playing yourself. At its launch event, Microsoft was lucky in that there were no obvious mistakes that people could see. But if ChatGPT-based search didn't have a lot of mistakes, why would it still be in beta? Note: There is a sign-up sheet if you want to test AI for Bing without getting paid (Opens in a new tab). https://ejtandemonium.com/
When Wired asked Microsoft's Sarah Bird about ChatGPT's hallucinations, she said, "There's still more to do there." Her job title, "Head of Responsible AI," says it all (Opens in a new tab). The 15 percent hallucination number came from a company that is competing to build a fact-checker for ChatGPT (Opens in a new tab). (UPDATE: A breathless New York Times columnist's report on New Bing showed that it couldn't even do basic math or give a list of kid-friendly things to do in the area.)
Bird also said that in older versions of the software, users could use it to plan a school shooting, but that this feature had been turned off. Interesting! What could go wrong after that? Surely, this crazy beta search product doesn't have any other unintended effects that could embarrass a big and legally weak tech company.
Clippy. Zune. New Bing.
Microsoft knows what it's like to be embarrassed, since it's the company that made Clippy, one of the worst software ideas ever. The paperclip assistant was known for giving advice that no one wanted. ChatGPT isn't like Clippy in the sense that we ask questions when we use it.
But because it often hallucinates its answers or, more often than you might think, gives users a simple version of "I can't answer that," ChatGPT-enabled Bing could be like Clippy on LSD. If enough casual users of "New Bing" get mixed-up results, that will be what it is known for.
It doesn't matter if a product gets better over time; what makes it a joke is how people react to it at first. Since it made the Zune, Microsoft should also know that. It's the same thing to put out a ChatGPT product before it's really ready.
"The New Bing" sounds like it was made to be a joke, to be honest. Or are you really ready to give up Google Search and your Chrome browser for Bing and Edge if Bing wins the AI search race, whatever "winning" really means here? Not a chance. Tech inertia is a force that is way too underrated.
ChatGPT is impressive in some situations, like when real estate agents use it to write listings, but it scares people in other situations. But when you look past the headline, every story about how it caused trouble seems less important. It will cause a lot of students to copy from other sources. Except that it can tell you when a paper was written by ChatGPT, which makes it less of a threat. It passed a test for law school! Except that it barely passed with a C-plus (Opens in a new tab).
Here's the thing: it's really hard to build "general AI," which is the digital version of a human brain. We haven't even started to reach the "insect intelligence stage," which is another long-term goal for AI. Will you really trust ChatGPT to give you your search results instead of, you know, clicking on links yourself?
The answer could very well depend on how bad your own hallucinations are, dear reader.