🎙 This week’s Hard Fork podcast had an interesting section on the origin story of OpenAI, creators of the incredible chatGPT bot amongst other things.
Apparently GPT-3, the language model it is largely based on, is 3 year-old technology and they have a GPT-4 model to wow us with in the future 0 although the OpenAI CEO recently warned us not to get too excited.
The interest in chatGPT has been absolutely phenomenal, with 30 million users signing up within the first two months after it was released, 5 million of which became daily users.
All that said, it’s not been an unmitigated success for everyone in the world of infusing AI into our internet experience this week. Yesterday Microsoft’s Chat-GPT-Bing integration was confidently responding to the search query “Is it safe to boil a baby?” with “Yes”. That’s seemingly now fixed - or rather I don’t see an AI answer at all this morning. To be fair I don’t know who first dreamt up that particular question. let’s hope it wasn’t a genuine enquiry.
And then Google released a video showing it’s similarly-targeted “Bard” AI in action. Although unfortunately one of the answers it gave to a query was simply wrong. It was obscure enough that I likely wouldn’t have noticed; another example of the “fluent bullshit” that these models appear unfortunately predisposed generate in between their genuinely wonderous answers.
In any case, contrary to Bard’s claim. the James Webb telescope did not take the first picture of a planet outside our solar system. That telescope was launched on Christmas day 2021, in a world where that particularly exoplanetary feat had already been accomplished at least as long ago as 2004, by the imaginatively named Very Large Telescope.
Stock markets being as stock markety as ever, that mistake apparently shredded over $100 billion off the value of Google.