Big Tech is rushing to make products that can compete with OpenAI's ChatGPT, which is a popular AI chatbot and also the app with the fastest growth rate of all time. Bard, Google's entry, will be released "in the coming weeks," according to a Google blog post. However, it's already doing a great job of imitating ChatGPT by giving out wrong information.
In Google's blog post about Bard, there is an animated graphic meant to show how Bard works. However, the AI in the graphic is wrong, and says that the James Webb Space Telescope took the first picture ever of an exoplanet. Google also sent out a tweet with the animation and this message.
Webb took its first picture of an exoplanet in September of last year. However, that wasn't the first picture ever taken of an exoplanet; that happened in 2004. (Opens in a new tab).
It's not clear what happened, but the fact that Bard's claim about James Webb is so new makes me raise an eyebrow. In a language model, information is not read from a list of facts. Instead, it is given out by systems that are very good at completing sentences. Sentences about the very recent past might have more mistakes than usual for an AI because the information in them hasn't been written as many times. This is probably one reason why ChatGPT's model won't tell you much after 2021. (Opens in a new tab). https://ejtandemonium.com/
This mistake shows that all generative AI products still have a problem with ignoring truth value, which could be a good reason for people to keep using old-fashioned search engines. Microsoft is working on adding an answer engine similar to ChatGPT to its Bing search engine. However, Microsoft says that "Bing will sometimes misrepresent the information it finds, and you may see responses that sound convincing but are incomplete, wrong, or inappropriate." http://sentrateknikaprima.com/
In Google's blog post about Bard, there is an animated graphic meant to show how Bard works. However, the AI in the graphic is wrong, and says that the James Webb Space Telescope took the first picture ever of an exoplanet. Google also sent out a tweet with the animation and this message.
Webb took its first picture of an exoplanet in September of last year. However, that wasn't the first picture ever taken of an exoplanet; that happened in 2004. (Opens in a new tab).
It's not clear what happened, but the fact that Bard's claim about James Webb is so new makes me raise an eyebrow. In a language model, information is not read from a list of facts. Instead, it is given out by systems that are very good at completing sentences. Sentences about the very recent past might have more mistakes than usual for an AI because the information in them hasn't been written as many times. This is probably one reason why ChatGPT's model won't tell you much after 2021. (Opens in a new tab). https://ejtandemonium.com/
This mistake shows that all generative AI products still have a problem with ignoring truth value, which could be a good reason for people to keep using old-fashioned search engines. Microsoft is working on adding an answer engine similar to ChatGPT to its Bing search engine. However, Microsoft says that "Bing will sometimes misrepresent the information it finds, and you may see responses that sound convincing but are incomplete, wrong, or inappropriate." http://sentrateknikaprima.com/
