On Sunday, OpenAI pointed out the insufficient training data as its Orion model posts smaller gains over its predecessor, GPT-4.
Based on reports, the development of the large language model (LLM) may have reached its limit. More concretely, the slowdown of artificial intelligence (AI) from the company creates an impact on the whole industry.
In addition, Orion had less significant quality improvement than the earlier version. It also does not outperform GPT-4 in terms of programming.
While the LLM shows improvements in language capabilities, the model is more expensive to run in data centers.
On top of that, researchers at OpenAI reveal that the slowdown of Orion is rendered by insufficient high-quality training data. They noted that the team has already utilized most texts and data available to the public.
For this reason, Nick Ryden led the organization in building the Foundations Team to address the problem. The company targets using training material made by AI models or synthetic data to assist in closing the gap.
According to Chief Executive Officer Sam Altman, the plan of OpenAI is shifting to learning from less data despite a sufficient amount already available.
Meanwhile, the halt in the AI industry also involves the upcoming Gemini 2.0 from Google, which failed to achieve its internal targets.
ChatGPT of OpenAI Rejects Deepfake Requests
As indicated by reports, the ChatGPT of OpenAI rejected about 250,000 deepfakes requests during the United States elections. It is a result of heightened measures that Altman took to prevent misinformation.
The organization stressed that the AI tool has denied prompts to render photos of politicians, including President-elect Trump, Vice President Harris, and President Biden.
Moreover, the OpenAI made sure that ChatGPT did not provide any political preferences or recommendations to individuals.
At the same time, it partnered with the National Association of Secretaries of State to redirect users asking about different voting-related questions.
