Openai eyes researcher level ai – but roadblocks remain
OpenAI is aggressively pursuing the development of artificial intelligence systems capable of independently tackling complex technical challenges, potentially rivaling the work of human research scientists. The company’s chief scientist, Jakub Pachocki, believes recent advancements in coding and mathematical reasoning, coupled with progress in physics, are pushing them closer to a significant breakthrough.
A long-term horizon for autonomous intelligence
Pachocki emphasized that the key metric for evaluating progress is the duration a model can operate autonomously – specifically, the length of tasks it can handle without human intervention. He outlined an ambitious timeline: an ‘AI research intern’ by September 2026, followed by a fully autonomous ‘AI researcher’ by March 2028. It’s a bold, almost audacious, goal – and CEO Sam Altman himself acknowledged the potential for failure, stating OpenAI “could fail completely” in achieving it, yet stressing the importance of transparency.
The accelerating growth of coding tools is fueling this push. Agents like Codex, currently automating a significant portion of OpenAI’s coding operations, demonstrate the tangible progress. Pachocki highlighted ‘mathematical benchmarks’ as a crucial area for improvement, citing their ease of verification as a validation point. “We’re seeing this explosive growth in coding tools,” he stated. “For most people, the act of programming has changed fundamentally.”
The short-term challenge: focused autonomy
Despite the encouraging trends, Pachocki cautioned that achieving truly independent operation remains a challenge. He doesn’t foresee systems that simply ‘do what you tell them’ this year. The immediate focus is on developing AI capable of handling specific technical tasks with greater autonomy, leveraging more computational resources, and sustaining longer operational periods. “For more specific technical ideas – like ‘I have this particular idea about how to improve the models, how to execute this evaluation differently’ – I think we have the pieces, we just need to put them together,” he explained.
However, the reality is that current AI isn’t yet at the level of a fully independent researcher. “I don’t expect we’ll have systems that simply… and do it, not this year,” Pachocki clarified. The path forward requires significant investment and ongoing refinement. But with the rapid evolution of coding Technology and a clear target in sight, OpenAI’s pursuit of AI capable of independent, complex problem-solving continues unabated.