unhobbling

Improvements that release capabilities latent in a trained model by removing default limitations rather than by further scaling of pretrainingpretraining, n.: The first and usually the most expensive stage of building a modern AI model, in which it learns general patterns from enormous amounts of raw text, images or code before being specialized for particular uses.: for example giving a model toolstool use, n.: An AI model's ability to reach beyond its own knowledge by operating outside tools — searching the web, running code, using a calculator or an app — and folding the results into its work., memory and longer context windowscontext window, n.: How much an AI model can read and keep in view at once: the instructions, the conversation so far and any attached documents. Text that does not fit has to be cut, summarized or left out.; letting it think step by step (chain of thoughtchain of thought, n.: Getting an AI to write out its reasoning step by step before giving an answer, the way a student shows their working. This often makes answers to math and logic problems more accurate., reasoning trainingreasoning model, n.: An AI model trained to work through a problem step by step in hidden 'thinking' text before it answers. Letting it think longer generally makes it better at math, coding and planning.); wrapping it in scaffoldingscaffolding, n.: The code, prompts and tools built around an AI model so it can carry out tasks it could not manage from a bare prompt — like the scaffolding that lets builders reach every part of a building. and agent loops so that it can act, including through computer usecomputer use, n.: An AI's ability to work a computer through its normal screen and controls: it studies pictures of what is displayed and sends mouse and keyboard actions, so it can use ordinary apps that were built for people.; or post-training a base model into a usable assistant with RLHFreinforcement learning from human feedback, n.: A way of training AI using people's judgments: humans compare pairs of answers, a scoring model learns their preferences, and the AI is then trained to produce answers that score well.. In Leopold Aschenbrenner's accounting of AI progress it is the third driver, alongside growth in physical compute and algorithmic efficiencies (effective computeeffective compute, n.: A way of measuring AI progress that counts better methods as if they were extra computing power: if a new technique lets a model match one trained with twice the computing, it counts as doubling effective compute.); he wrote that its gains are hard to put on the same effective-compute scale but clearly large, and expected them to turn chatbots into agents. The concept emphasizes that measured capability depends on how a model is deployed, not only on its pretraining.

Most of this year's jump came from unhobbling — tools and longer reasoning — not a bigger base model.