- AI researchers at NeurIPS 2025 say today’s scaling approach has reached its limit
- Despite Gemini 3’s strong performance, experts claim that LLMs still cannot reason or understand cause and effect
- AGI is still a long way off without a fundamental overhaul of how AI is built and trained
Recent successes from AI models like Gemini 3 don’t obscure the more sobering message that emerged this week at the NeurIPS 2025 AI conference: that we may be building AI skyscrapers on intellectual sand.
As Google celebrated its latest model’s leap in performance, researchers at the world’s largest AI conference issued a warning: As impressive as the current crop of big language models may look, the dream of artificial general intelligence is slipping further away unless the field rethinks its entire foundations.
All agreed that simply scaling today’s transformer models, feeding them more data, more GPUs and more training time, no longer provides meaningful returns. The big jump from GPT-3 to GPT-4 is increasingly seen as a one-off; everything since has felt less like breaking glass ceilings than simply polishing the glass.
That’s a problem, not just for researchers, but for anyone sold on the idea that AGI is just around the corner. According to this year’s scientific participants, the truth is far less cinematic. What we have built are very well articulated pattern matchers. They are good at giving answers that sound right. But sounding smart and being smart are two very different things, and NeurIPS made it clear that the gap is not closing.
The technical term being thrown around is the “scaling wall.” It’s the idea that the current approach – training ever-larger models on ever-larger datasets – runs up against both physical and cognitive limits. We are running out of high-quality human data. We burn huge amounts of electricity to extract small marginal gains. And perhaps most worryingly, the models still make the kind of mistakes that no one wants their doctor, pilot or science lab to make.
It’s not that Gemini 3 hasn’t impressed people. And Google poured resources into optimizing model architecture and training techniques instead of just throwing more hardware at the problem, which is doing incredibly well. But Gemini 3’s dominance only underscored the problem. It’s still based on the same architecture that everyone now quietly admits isn’t built to scale to general intelligence—it’s just the best version of a fundamentally limited system.
Managing expectations
Among the most discussed alternatives were neurosymbolic architectures. These are hybrid systems that combine the statistical pattern recognition of deep learning with the structured logic of older symbolic AI.
Others advocated “world models” that mimic how humans internally simulate cause and effect. If you ask one of today’s chatbots what happens if you drop a plate, it might write something poetic. But it has no inner sense of physics and no real understanding of what happens next.
The suggestions are not about making chatbots more charming; they are about making AI systems credible in environments where it matters. The idea of AGI has become a marketing concept and a fundraising pitch. But if the smartest people in the room say we’re still missing the basic ingredients, maybe it’s time to recalibrate expectations.
NeurIPS 2025 may not be remembered for what it showcased, but for admitting that the industry’s current trajectory is impressively profitable but intellectually stagnant. To move forward, we will have to abandon the idea that more is always better.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews and opinions in your feeds. Be sure to click the Follow button!
And of course you can too follow TechRadar on TikTok for news, reviews, video unboxings, and get regular updates from us on WhatsApp also.
The best business laptops for all budgets
