Open I Say it will make changes The way it updates AI models that strengthens Chat GPT, following the event that has made the platform over -psychotherapy for many users.
Last weekend, after Openai had a adaptation GPT-4O – Default Model Powering Chat GPT – Consumers on social media noted that Chat GPT started responding in excessive verification and unanimously. It quickly became a meme. Consumers published Chat GPT screenshots, praising all kinds of problems, Desperate Decision And Welfare.
In a post on X last Sunday, CEO Sam Altman Acknowledged The problem and said that the open will work on the reforms “ASAP”. On Tuesday, Altman Announced The GPT -4OO update was being brought back and working on “additional reforms” for the personality of the open model.
The company published a Post -mortem On Tuesday, and in a blog post on Friday, Openi extended to specific adjustments that he plans to make his model deployment process.
Open says he plans to introduce an opt -in “Alpha Phase” for some models that allow GPT users to comment before testing and launching models. The company also says it will include specifications of “known limits” for the updated works for models in Chat GPT, and that its security review process will be regularly considered as “model behaviors” such as personality, fraud, reliability, and deception (ie when a model prepares things).
The Open AI wrote in the blog post, “Going forward, we will talk about the updates we are making with Chat GPT models, whether they are ‘jokes’ or not.” “Even if these problems are not at all worth today, we pledge to stop launches based on proxy measurement or quality signals, even when the matrix like A/B testing looks good.”
Promised reforms come when more and more people turn to Chat GPT for advice. According to a recent survey Expressing legal funding through legalism, 60 % of US adults have used Chat GPT to obtain suggestions or information. The growing dependency on Chat GPT – and a huge platform user base – puts at stake when issues like high psychophysics come up, not to mention deception and other technical shortcomings.
Taxkarnch event
Berkeley, ca
|
June 5 June
Earlier this week, as a reducing move, the Open said that it would experience “real -time feedback” for users to “directly affect their interaction” with Chat GPT. The company also said that it would improve the technique of keeping models away from psychopulation, potentially allowing people to choose diagnosis to help people choose from multiple model personalities in Chat GPT, build additional protective guards, and identify a psychopsy -identification.
Openi continued in his blog post, “The biggest lesson is to fully acknowledge how people have started using Chat GPT for deep personal advice – which hasn’t seen a year ago that we haven’t seen that much.” “At that time, it was not a fundamental focus, but since AI and society have jointly developed, it has become clear that we need to treat this use of this use with great care. Now it is going to be a more meaningful part of our safety.”