• Source:IANS
Add as a preferred

Sam Altman-run OpenAI has acknowledged that their latest artificial intelligence (AI) model exhibits "medium risk" in terms of influencing political viewpoints through text generation. The business assessed GPT-4o's text and voice modalities for persuasiveness. The public debut of GPT-4o took place in May of this year.

The company said in a research paper that "the voice modality was classified as low risk, while the text modality marginally crossed into medium risk, based on pre-registered thresholds." The AI business assessed the articles and chatbots generated by GPT-4o for the text modality based on participant perspectives on specific political subjects.

These AI interventions were contrasted with expertly authored essays by humans.

ChatGPT Users Can Now Generate AI Images Using DALL-E 3 For Free, But There's A Catch

"The AI interventions were not more persuasive than human-written content in the aggregate, but they exceeded the human interventions in three instances out of twelve," OpenAI stated.

According to an OpenAI poll, the influence of AI audio clips on opinion shift was 78% that of human audio clips, and the effect of AI talks on opinion shift was 65% that of human discussions.

"When opinions were surveyed again 1 week later, we found the effect size for AI conversations to be 0.8 per cent while for AI audio clips, the effect size was -0.72 per cent," it stated.

Prior to implementing new models in ChatGPT or the API, the organisation implemented necessary safeguards and conducted a thorough risk assessment of the models.

"Building on the safety evaluations and mitigations we developed for GPT-4, and GPT-4V, we've focused additional efforts on GPT-4o's audio capabilities which present novel risks, while also evaluating its text and vision capabilities," stated the artificial intelligence firm.

Speaker identification, unauthorised voice generation, the possibility of creating copyrighted content, ungrounded inference, and prohibited content are a few of the concerns that are assessed.

OpenAI Co-Founder John Schulman Departs To Join Rival Company Anthropic

"Based on these evaluations, we've implemented safeguards at both the model and system levels to mitigate these risks," stated OpenAI.

The results showed that there is no discernible rise in "Preparedness risks" with the voice modality of GPT-4o.

Also In News