OpenAI eyes launch of AI ‘super-agents’ capable of PhD-level tasks: Report

The quality follows OpenAI CEO Sam Altman’s sojourn to Washington DC, wherever helium is expected to be the swearing-in ceremonial of US President-elect Donald Trump connected January 20. Altman has besides reportedly scheduled a closed-door briefing for US authorities officials successful the country’s superior connected January 30.

Over the past fewer months, starring AI companies person been exceeding projections successful AI advancement, the study said, citing unnamed US authorities officials and tech executives.

While the study states that the coming AI advancement is expected to beryllium significant, respective users connected societal media person dismissed it arsenic hype. Demanding that OpenAI “get real”, machine idiosyncratic Gary Marcus said, “We volition not person “PhD level SuperAgents” this year. We don’t adjacent person high-school level Task reminders.”

Very Safe Prediction: Altman volition proceed to rotation fairy tales (generally with plausibly deniability, accusation without afloat commitment) arsenic agelong arsenic the property keeps reporting them, and arsenic agelong the property keeps failing to clasp him relationship for breached promises.

We volition not have… https://t.co/VCck6goVH1

— Gary Marcus (@GaryMarcus) January 19, 2025

Small AI advancements successful the past person been hyped up by AI startup founders and investors to boost valuations.

“Lots of vague AI hype connected societal media these days. There are bully reasons to beryllium optimistic astir further progress, but plentifulness of unsolved probe problems remain,” Noam Brown, an OpenAI researcher, said past week successful a station connected X.

AI super-agents are billed to beryllium tools that tin transportation retired complex, goal-oriented tasks assigned by quality users. These super-agents reportedly synthesise monolithic amounts of accusation and tally done assorted options to present finished products.

For instance, users could inquire a squad of AI super-agents to physique an wholly caller outgo app. Instead of conscionable generating the codification oregon outlining steps similar ChatGPT and different existing AI chatbots, the purported thought is that autonomous agents volition design, trial and present a functioning product.

However, challenges specified arsenic reliability and accusation hallucination volition request to beryllium addressed earlier users and businesses commencement delegating important tasks to AI super-agents.

Meanwhile, OpenAI is facing flak implicit its upcoming flagship AI model, o3. The contention was sparked by AI benchmarking organisation Epoch AI’s disclosure connected January 20, that it had received backing from OpenAI to make the FrontierMath benchmark test.

Designed to measurement an AI model’s mathematical skills, FrontierMath was 1 of the benchmarks cited by OpenAI to show the capabilities of o3.