Openi is moving to publish the results of the safety assessments of the model in the interior more regularly on what the outfit is saying is an effort to increase transparency.
On Wednesday, Openi launched the hub of security assessments, a web page showing how the score of the company’s models on various tests for the generation of harmful content, jailbreaks and hallucinations. Openii says that he will use the hub to share metrics on a “continuous base” and that he intends to update the hub with “updates of the main models” in the future.
“As the AI assessment science, we aim to share our progress on the development of more scalable ways to measure the capacity and safety of the model,” wrote Openai in a blog post. “By sharing a subset of our security evaluation results here, we hope that this will not only make it easier to understand the safety performance of the Openai systems over time, but also supports the efforts of the community to increase transparency throughout the field.”
Openai says that it could add further hub evaluations over time.
In recent months, Openii has raised the anger of some ethics for having reported that he had brought the security tests of some top models and failing to issue technical relationships for others. The CEO of the company, Sam Altman, is also accused of Openai misleading managers regarding the revisions of the safety of the model before its brief expansion in November 2023.
At the end of last month, Openii was forced to restore an update of the default model that powered Ghatgpt, GPT-4O, after users started reporting that he responded excessively and pleasantly. X has become flooded with chatgpt screenshots that applauds all types of problematic and dangerous decisions and ideas.
Openi has said that it will implement different corrections and changes to prevent future accidents of such, including the introduction of an “Alfa opt-in phase” for some models that would allow some chatgpt users to test the models and provide feedback before launch.
