There are a lot of “confuse” operations on Openai this week
First, many accounts were banned on Sunday, especially those with IP addresses in Taiwan, Japan, and Hong Kong. The hit rate was 40% by visual inspection. Registration of new users is closed on Saturdays. The second thing is that since last Tuesday, plus users have been unable to log in on their mobile phones, and ordinary users have been unable to log in. Starting Wednesday, the google web version login needs email confirmation to log in.
In summary, we guess three reasons:
First, there are malicious data feeds. Asia and other places frequently input what it thinks is wrong information. Its confrontation training gives an early warning, so openai decides to seal people’s accounts and even seals related reports without distinction. AI chips will be delivered by the end of the second quarter at the fastest. To prevent unnecessary waste, stop new user registration and non-plus user login. Third, Google uses gpt to generate answers to train models similar to large models. The industry discloses secrets, and other companies do the same because there are too little data in real human communication scenarios. Still, openai is not expected to pursue it because its data is also a lot of gray areas. Restricting access to Asia Is there any consideration?
Let’s talk about Bloomberg GPT’s 50 billion parameters, 700 billion tokens training semantic library, based on the model Unigram! Overseas guesses that gpt3.5 turbo may also be a model with tens of billions of parameters, which has been slimmed down. Many talents are currently focusing on developing vertical models, hoping that the excellent vertical model can achieve an effect comparable to that of large models.
Disclaimer: Community is offered by Moomoo Technologies Inc. and is for educational purposes only.
Read more
Comment
Sign in to post a comment