Like most things Microsoft gets into they're racing straight to third place from the sizable lead OpenAI has established. ChatGPT took the world by storm just 6 weeks ago, being the first large-scale experiment of RLHF, but competitors are biting at their heels already.
From what's being said on social media, the most common complaints are that each update seemingly trades off capability for more safety. A sound strategy if their goal is to eventually integrate it into Teams as Clippy 2.0. Enterprises would be unhappy about a chatbot that actually answered every question honestly. But that strategy will not win them the consumer market.
Large language models at their core are the application of techniques, data, and training time meaning there is virtually no barrier to entry. The techniques are widely known, the data is widely accessible, and a couple million dollars is not unreachable even for individuals. Just like DALL-E is all but forgotten in the footsteps of Stable Diffusion and Midjourney, ChatGPT will soon be eclipsed by free and open models as well.
Sounds like Microsoft is trying to leap frog ahead of Google, but I worry more about the consequences of ChatGPT. Not much thought has been put into whether incorporating it in Bing will be good or bad for users.
Has there been any studies on how incorporating ChatGPT into a search engine will affect the search results and in turn millions of people that use it? I would like to read it.
Were there any studies on incorporating initial search to the young internet prior to deploying that to millions of people? Google started in a similar way. Worst case you get bad search results which is where Bing is already.
The internet isn't young anymore. There are millions of people using Bing right now. There also wasn't any thought when people incorporated the like button in social media. Look how it turned out for its users. Search engines are a mature product with enough users. Microsoft isn't some startup and it has capital to do this.
From what's being said on social media, the most common complaints are that each update seemingly trades off capability for more safety. A sound strategy if their goal is to eventually integrate it into Teams as Clippy 2.0. Enterprises would be unhappy about a chatbot that actually answered every question honestly. But that strategy will not win them the consumer market.
Large language models at their core are the application of techniques, data, and training time meaning there is virtually no barrier to entry. The techniques are widely known, the data is widely accessible, and a couple million dollars is not unreachable even for individuals. Just like DALL-E is all but forgotten in the footsteps of Stable Diffusion and Midjourney, ChatGPT will soon be eclipsed by free and open models as well.