OpenAI CEO Sam Altman and a number of other different firm executives hosted an ask-me-anything (AMA) session on Thursday. The session was hosted on the social networking platform Reddit and customers have been instructed to ask questions concerning the AI agency’s merchandise akin to ChatGPT or common queries about synthetic intelligence (AI) and synthetic common intelligence (AGI). Through the session, Altman stated that GPT-5 won’t be launched this yr, nonetheless, the corporate plans to introduce “some excellent releases” earlier than the top of 2024.
OpenAI Employees Host AMA on Reddit
The AMA session was hosted on the ChatGPT subreddit. Calling it “our Reddit launch”, a number of executives together with Altman, OpenAI CPO Kevin Weil, SVP of Analysis Mark Chen, VP Engineering Srinivas Narayanan, and Chief Scientist Jakub Pachocki participated within the question-and-answer publish. The official X (previously generally known as Twitter) deal with of OpenAI additionally posted concerning the Reddit AMA.
Answering a query concerning the timeline for GPT-5 or its equal’s launch, Altman stated, “We’ve some excellent releases coming later this yr! nothing that we’re going to name gpt-5, although.” This appears on par with what a number of reviews have confirmed with most anticipating OpenAI to launch the subsequent flagship mannequin someday in 2025.
One other consumer requested concerning the worth that SearchGPT or the ChatGPT Search function brings, Altman stated that he finds it to be a sooner and simpler strategy to get to the knowledge. He additionally highlighted that the online search performance might be extra helpful for advanced analysis. “I additionally stay up for a future the place a search question can dynamically render a customized net web page in response,” he added.
Weil additionally fielded a number of questions from customers. One requested concerning the delay in Sora, to which the OpenAI CPO stated that the delay was prompted as a consequence of further time taken to excellent the mannequin, getting security and impersonation proper, and the necessity to scale compute. Nonetheless, he didn’t reveal a date for its launch.
Weil additionally highlighted that the ‘o’ collection AI fashions, akin to GPT-4o and o1-preview, will develop into a mainstay within the firm’s lineup and can make an look even after the discharge of GPT-5. Moreover, he additionally revealed that the ChatGPT Superior Voice Mode might be tweaked so as to add a singing voice to the AI.
One consumer recommended, “Can we please get a hands-free strategy to finish a voice dialog with ChatGPT? Utilizing the motion button shortcut on iPhone, however the voice chat wants manually dismissed each time.” Weil appeared to be impressed with the concept and stated, “I really like this concept. Sharing with the staff now!”
OpenAI SVP or Analysis Mike Chen additionally answered an essential consumer query about AI hallucination. Explaining why hallucinations from AI fashions are usually not fully gone, he known as it a basically exhausting drawback. It is because the AI fashions be taught from human-written textual content, and people can typically make errors, which then are added to the core dataset of the massive language fashions (LLMs).
“Our fashions are enhancing at citing, which grounds their solutions in trusted sources, and we additionally consider that RL will assist with hallucinations as effectively – after we can programmatically examine whether or not fashions hallucinate, we will reward it for not doing so,” Chen added.
