Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Openai moves to spread the results of the typical safety assessments of the internal spontaneity regulation regularly in what the outfit says is an attempt to increase transparency.
On Wednesday, Openai launched HUB safety reviews,, Web page shows how the company’s models are recorded in various tests to generate harmful content, broken, and hallucinations. Openai says he will use the axis to exchange standards on a “continuous” basis and that it intends to update the axis through “main model updates” to move forward.
“With the development of artificial intelligence evaluation science, we aim to share our progress in developing more developmental ways to measure the ability of the model and safety,” Openai books in A. Blog post. “Through the participation of a sub -set of our safety evaluation results here, we hope that this is not only easy to understand the safety performance of Openai with time, but also support society’s efforts – to increase transparency throughout the field.”
Openai says it may add additional reviews to the axis over time.
In recent months, Openai has angered some ethics It is said Hurry to test safety for some of the leading models Not issuing technical reports to others. The CEO of the company, Sam Al -Tamman, also A defendant stands From Openai CEOs of Officer Officer on Model Safety Reviews before him The overthrow is short In November 2023.
Late last month, Openai The update is forced To the virtual model that runs ChatGPT, GPT-4O, after users began reporting that he responded in an excessive healthy and acceptable way. X immersed screenshots of Chatgpt screen constructing all kinds of problems, dangerous Decisions And ideas.
Openai He said This will Running Many repairs and changes to prevent such incidents in the future, including the introduction of the “Alpha stage” for some models that would allow some Chatgpt users to test models and make notes before launch.