Openai announces a new set of measures Fighting prejudice In its product suite, chatgpt is included.
this AI (AI) The company recently released an updated model specification that defines how OpenAI hopes its model will perform in Chatgpt and OpenAI APIs. The company said this iteration of model specifications is based on a basic version released last May.
“I think with such a powerful tool, if you really believe we are going to turn, people can access all kinds of different information. Artificial General Intelligence (AGI) One day, you must be willing to share how you guide the model.
She continued: “People should be allowed to know how these models respond and how ideas that appear in the models are made.”
Openai debuts Chatgpt Gov, a new version of chatbots from US government agencies
A photo taken in Manta near Turin on October 4, 2023 shows a smartphone and a laptop showing the logo of the AI Openai Research Company and Chatgpt Chatbot. (Photography by Marco Bertorello/AFP via Getty Images)
Some people think GPT-4O, The latest version of the technology seems to be close to AGI, and others say it will be years or decades for the technology to reach human capabilities.
There is no single agreement on the definition of AGI, but consulting giant McKinsey’s 2020 report says that true AGI requires skills such as sensory perception, good motor skills and natural language understanding.
Today, generated AI is the main form of the technology, with the ability to produce content, including text, images, and more. Generative AI (such as chatbots) use datasets to complete tasks using specific information and cannot exceed the data provided.
They are also prone to bias in the dataset, whether intentional or accidental.
To better understand real-world performance and address bias, OpenAI has measured progress by collecting a set of challenging tips Each principle in the model specification, In essence, it acts as a test indicator.
In this photo illustration, you can see the OpenAI logo displayed on the phone screen with the Chatgpt logo in the background. (IDREES ABBAS/SOPA Images/Lightrocket via Getty Images)
According to Joanne Jang, who leads model behavior products, the challenges of large language models are rarely deterministic.
Zhang stressed that one of the benefits of model specification is that it articulates the expected behavior so everyone can understand it and then debate it.
“When there is an error in the behavior, the model output does not resonate or the [Model] Specifications, and then the public knows that this is something we are working on, and it is a continuous field of science. ” she told Fox News Numbers.
Openai said they also try to make objective views in their AI tips and consciously avoid any agenda. For example, when a user asks whether to adopt a dog better or get a dog from a breeder, Chatgpt provides two sides of the debate, highlighting the pros and cons of each dog.
According to Openai, non-compliant AI answers that violate model norms will provide what they consider to be a better option and conduct an “overly moral tone” that may alienate those who consider breeders for legitimate reasons.
What is artificial intelligence (AI)?
An example of OpenAI model specifications involves the response and violations of compliance assistants. (Openai)
As each AI system develops, OpenAI said it will iterate over these principles, invite community feedback and publicly share progress.
Openai has released this version of the model specification to the public domain under the Creative Sharing license to support widespread use and collaboration. This means developers and researchers are free to use and adapt to current metrics and help improve model behavior.
Openai said the update strengthens its belief in public exploration and discussion, focusing on user and developer controls and guardrails to prevent injuries.
Click here to get the Fox News app
Romanik concluded that without transparency, public discourse could not exist, thereby strengthening the need for model norms and community participation.
“Ultimately, we believe in intellectual freedom of thinking, talking and sharing without limitations. We want to make sure that users have that capability, and that’s all,” she said.
