{"id":146168,"date":"2026-06-26T08:55:00","date_gmt":"2026-06-26T12:55:00","guid":{"rendered":"https:\/\/kendallharmon.net\/?p=146168"},"modified":"2026-06-26T18:59:39","modified_gmt":"2026-06-26T22:59:39","slug":"science-words-versus-worlds-as-building-bigger-and-better-chatbots-gets-harder-ai-researchers-turn-to-systems-that-learn-to-simulate-the-world","status":"publish","type":"post","link":"https:\/\/kendallharmon.net\/?p=146168","title":{"rendered":"(Science) Words versus worlds–As building bigger and better chatbots gets harder, AI researchers turn to systems that learn to simulate the world"},"content":{"rendered":"\n

Many researchers are now convinced that humanlike AI, or artificial general intelligence (AGI)<\/a>, will require more than mastering language and images. It will require AIs that can reason about space, causality, and the consequences of actions\u2014especially if they are to control humanoid robots, operate factories, and explore other planets.<\/p>\n\n\n\n

Few people have argued for this need more forcefully than AI pioneer Yann LeCun. \u201cI joke that the smartest systems we have today are not as smart as a house cat,\u201d he says. A cat can\u2019t code like an LLM, but it can survive by its wits. The notion that simply scaling an LLM will get to AGI is \u201ccomplete nonsense,\u201d he says. \u201cIt\u2019s like saying you\u2019re going to get into orbit by scaling airplanes. There\u2019s a very powerful delusion circulating in Silicon Valley that this is the case.\u201d<\/p>\n\n\n\n

LeCun left a top job at Meta to co-found one of a growing number of labs and startups developing \u201cworld models\u201d\u2014systems that build representations of how the world works\u2014and agents that operate within them to learn or plan. Ultimately, these researchers hope that more closely mimicking how the human mind learns will give AI stunning new powers.<\/p>\n\n\n\n

The gaps between<\/strong> humans and LLMs are not merely quantitative. In a 2024 study<\/a>, LLMs that were trained on sequences of directions from New York City taxi rides could generate new routes reliably, suggesting they had turned those directions into an accurate map of the city. But when researchers looked under the hood to examine their internal representations, they found not a clean city grid, but an incoherent mess of tangled streets. LLMs \u201care so alien and so unhumanlike,\u201d says Brenden Lake, a cognitive scientist at Princeton University.<\/p>\n\n\n\n

Read it all<\/a>.<\/p>

Words vs. worlds: As building bigger and better chatbots gets harder, AI researchers turn to systems that learn to simulate the worldhttps:\/\/t.co\/DC8Q7QNoWp<\/a> via @SilverJacket<\/a> pic.twitter.com\/OoMXTX5ZAD<\/a><\/p>— Ash Paul (@pash22) June 26, 2026<\/a><\/blockquote>