There had been murmurs that AI progress had slowed down after the release of GPT-5, which didn’t end up being as much of a jump over competitors as some had hoped, but one of the godfathers of AI seems to believe that AI progress isn’t showing any signs of slowing down.
Yoshua Bengio, the Turing Award-winning computer scientist whose pioneering work on deep learning helped spark the current AI revolution, has pushed back against the narrative of plateauing progress. Speaking about the trajectory of artificial intelligence capabilities, Bengio pointed to concrete scientific evidence that tells a different story—one of continued, and in some areas exponential, advancement.
“The most important point is the scientific data on many benchmarks of AI capabilities is very clear and trending up, in some cases exponentially fast,” Bengio said. “In large part over the last year, this has been thanks to progress in the reasoning abilities of these systems and the planning abilities of these systems, so that they can be more capable of achieving tasks that take time and planning.”
The observation is particularly significant coming from Bengio, who has been vocal about AI safety concerns and is not typically prone to hype. His assessment focuses on measurable performance across standardized benchmarks rather than anecdotal improvements or marketing claims.
“This doesn’t look, scientifically speaking, like we are hitting a wall,” Bengio continued. “But of course, nobody has a crystal ball. Maybe progress will continue at the same rate. Maybe it will stop. Maybe it will accelerate. Some people think that the advances in programming, for example, might be used to accelerate the development of new algorithms.”
The implications of Bengio’s assessment are substantial for the tech industry and beyond. His emphasis on reasoning and planning capabilities addresses what many researchers consider the next frontier in AI development—moving beyond pattern recognition and text generation toward genuine problem-solving abilities. Recent developments support this view: GPT-5, Claude 4.5 and a range of Chinese models are all quickly improving upon their previous iterations The notion that AI systems could increasingly contribute to their own development—using programming advances to accelerate algorithm discovery—suggests a potential feedback loop that could sustain or even accelerate progress. For businesses navigating AI investment decisions and policymakers considering regulatory frameworks, Bengio’s evidence-based optimism suggests the transformative impact of AI may be far from reaching its peak, even as the path forward remains fundamentally uncertain.