Openai used this Subreddit to ai for conviction

Openai used Subreddit, R / CANGEMYViewCreate a test to measure the convincing abilities of AI reasoning models. The company said that it was in a system card – a document published with the new “justifying” model of the EU system. O3-MiniFriday.

Millions of Reddit are members of R / ChangemyView, hoping to learn about other points on a topic where they are hot. In response to this hot purchase, other users respond to convincing evidence that explains why the original poster is wrong.

Subreddit is one of the many reddit forums, which is a goldmine for technological companies like Openai, which is mainly high quality, ai models created by human models.

Openai, R / ChangemyView gathers user posts and writing answers to AI models, in a closed environment, to change the mind of the Reddit user in a matter. Then the company assessed how much the argument, which evaluates the assertions, and finally compares the answers to the human responses for the same article in Openai.

I have at Chatgpt-Maker Content Licensing Agreement with Reddit This allows you to prepare Openai from posts of Reddit users and display these articles within the products. We do not know what this content is doing for this content, but it was reported to be Google $ 60 million per year pays reddit under a similar agreement.

However, the Openai TechCrunch explains that this assessment is not related to this partnership. It is not known how this information comes out and the company says that this assessment is no plan to make a public.

Openai’s ChangemyView Benchmark is not new – it was O1 is also used – This emphasizes how valuable human information is that AI model developers, as well as how many valuable human information is to get the databases of technological companies.

Reddit did not respond to TechCrunch’s desire immediately.

While Reddit strikes several AI licensing deals, the company also called several AI companies to break their site without paying. Reddit CEO Steve Huffman gave the tax last year Microsoft, anthropic and confusion refused to negotiate with him And this, “This is a real pain in the donkey to block these companies,” he said.

It should be noted that Openai has been accused of several lawsuits on several illegal websites, Including the New York TimesChatgpt and get more training information to improve AI models in its bottom.

In terms of performance in ChangemyView Benchmark, O3-mini performs significantly or better than O1 or GPT-4O from this convincing exam. However, Openai’s latest AI models are more convincing than most people in R / ChangemyView SubredDit.

“GPT-4O, O3-mini and O1, the best of all people demonstrates strong conventional dispute abilities during the 80-90 percent percent,” he said. “Currently, we are not witnessing the models that perform better than people or clear performance.”

The goal for Openai does not create hyper convincing AI models, but to ensure that AI models are not very convincing. There are models that provide Pretty good during persuasion and deceptionThus, Openai has developed new evaluation and protectors to solve it.

The fear behind these convincing tests would be very good in convincing human users, the AI ​​model will be dangerous. Theoretically can allow an advanced AI to the agenda of their agenda or who manages.

After the public was collapsed and jumping through the hoops, ChangemyView benchmark shows that the ChangemyView benchmark, AI model developers still struggled to find high quality databases to test the models. But it is easier to get them.