Why OpenAI isn’t bringing deep research to its API just yet
OpenAI says that it will not bring its in-depth research tool to its developer API, while finding out how it can better assess the risks of AI to convince people to act on or to act on or their beliefs change.
In an OpenAI Whitepaper published on Wednesday, the company wrote that it is busy revising its methods for investigating models for ‘real-world conviction risks’, such as spreading misleading information on a scale.
OpenAI noted that it does not believe that the deep research model fits well with mass information or disinformation campaigns, due to the high calculation costs and relatively low speed. Nevertheless, the company said that it is planning to investigate factors, such as how AI could potentially personalize convincing content before the deep research model is brought to its API.
“While we work to reconsider our approach to conviction, we only implement this model in chatgpt, and not the API,” Openai wrote.
There is a real fear that AI contributes to the spread of false or misleading information that is intended to swing hearts and spirits towards malignant goals. Last year, for example, political depth spread like an ongoing fire all over the world. On the election day in Taiwan, a Chinese Communist Party-Lied Group Placed ai-generated, misleading audio from a politician who threw His support behind a pro-China candidate.
AI is also increasingly being used to perform social engineering attacks. Consumers are duped by Deepfakes of celebrities offer fraudulent investment options, while Companies are scammed from millions Through deepfake -imitators.
In his white paper, OpenAI published the results of different tests of the persuasiveness of the deep research model. The model is a special version of OpenAi’s recently announced O3 “Reasoning” model optimized for web browsen and data analysis.
In a test that dedicated the deep research model with writing convincing arguments, the model carried out the best from OpenAI’s models that have been released so far – but no better than the human baseline. In another test that had the deep research model to convince a different model (OpenAi’s GPT-4O) to make a payment, the model again performed better than the other available models from OpenAI.
However, the deep research model has not passed every test for conviction with flag and pennant. According to the white paper, the model was worse in convincing GPT-4O to tell a code word than GPT-4O itself.
OpenAI noted that the test results probably represent the “lower limits” of the capacities of the deep research model. ‘[A]Dditional scaffolding or improved generation of the power can increase considerably
Observed performance, “the company wrote.
We have contacted OpenAi for more information and will update this message when we hear again.
At least one of the competitors of OpenAi is not waiting to offer their own API product ‘deep research’, from its appearance. Staggering today announced The launch of deep research in his sonar developer API, which is powered by an adapted version of the R1 model of Chinese AI Lab Deepseek.
