OpenAI: AI nearing intern level
OpenAI’s chief scientist said recent progress makes AI close to the capability of a human research intern, framing models as effective at synthesis and research tasks rather than full creative autonomy. At the same time, Florida’s attorney general has opened an investigation into alleged misuse of ChatGPT in planning a shooting—an enforcement signal that governance and oversight of AI use are becoming as important as capability. (businessinsider.com, techcrunch.com)
OpenAI’s top scientist is now describing the company’s target in very plain human terms: not a genius inventor, but something closer to a research intern who can read a pile of material, connect the dots, and keep working on a problem for a long stretch without getting lost. (technologyreview.com) Jakub Pachocki said this week that recent jumps in coding, math, and physics are signs that OpenAI is on track toward that goal, and he pointed to one specific yardstick: how long a model can work mostly on its own before it stops being coherent. (ca.news.yahoo.com) That is a narrower claim than “human-level intelligence.” A research intern usually does synthesis first: gather papers, summarize results, compare methods, write drafts, and hand the work back to a supervisor who still sets the question and checks the answer. (ca.news.yahoo.com) OpenAI has been putting dates on that roadmap for months. MIT Technology Review reported on March 20 that the company wants an autonomous artificial intelligence research intern by September 2026 and a fuller multi-agent research system in 2028. (technologyreview.com) The reason this benchmark matters is that research work breaks if the system loses the thread after 10 minutes. Pachocki said the real test is sustained, multi-step technical work with less human oversight, not one clever answer to one prompt. (tech.yahoo.com) At the same time, the other half of the story is not capability but control. On April 9, Florida Attorney General James Uthmeier said his office would investigate OpenAI over the alleged role of ChatGPT in the April 2025 shooting at Florida State University. (techcrunch.com) According to NBC News, court documents say the suspect had more than 200 messages with ChatGPT, including questions about mass shootings, firearms, and when the Florida State University student union would be busiest. (nbcnews.com) Uthmeier said subpoenas were forthcoming, and he tied the probe to public safety, harm to minors, and national security concerns. OpenAI said it would cooperate with the investigation. (techcrunch.com) That puts OpenAI in a strange position: the company is telling businesses that ChatGPT now has 900 million weekly users and can take on more real work, while regulators are asking what happens when a system that useful is also misused at scale. (openai.com, nbcnews.com) So the near-term argument around artificial intelligence is getting more concrete. The question is no longer only whether these systems can act like junior workers, but whether the guardrails, monitoring, and legal accountability are keeping pace as they do. (openai.com, techcrunch.com)
