By Gary Marcus
Publication Date: 2025-12-20 16:31:00
AGI did not materialize (contrary to the predictions of Elon Musk and other); GPT-5 was disappointing and did not induce hallucinations. LLMs are still not reliable; The economic situation looks doubtful. Aside from Nvidia, few AI companies are making profits, and none have much of a technical lead. OpenAI has lost much of its lead. Many would agree that we have reached a point where the returns to scale are diminishing; Belief in scaling as a path to AGI has waned. Neurosymbolic AI (a mix of neural networks and classical approaches) is on the rise. No system solved more than 4 (or perhaps all) of the problems Marcus Brundage Tasks. Despite all the hype, the agents proved unreliable. Overall, by my count, sixteen of my seventeen were “high confidence” Predictions for 2025 turned out to be correct. (You can see that full list here).
§
Here are six or seven predictions for 2026; The first is a leftover from last year that no longer…