scratchpad

Reasoning

Text that a language modellarge language model, n.: A computer program trained on enormous amounts of text to predict the words that come next. At vast scale, that simple skill lets it chat, answer questions, translate, summarize and write code. generates on the way to an answer, such as partial sums or a trace of program state, and then reads back while producing later tokenstoken, n.: The small chunk of text an AI language model reads and writes — often a word, part of a word or a punctuation mark. AI usage, limits and prices are counted in tokens.. Because a transformer performs a fixed amount of computation per token, writing intermediate results down lets it split a long calculation into many short ones; in Maxwell Nye and colleagues' 2021 experiments, models taught to use a scratchpad became far more accurate at multi-digit addition and at predicting what a piece of code outputs. The idea fed directly into 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. prompting and today's reasoning modelsreasoning 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., whose reasoning tracesreasoning trace, n.: The written-out 'thinking' an AI reasoning model produces before its final answer — notes to itself full of attempts, checks and corrections, which users may or may not get to see. are scratchpads at scale.

With a scratchpad, the model carried every digit correctly through twelve-digit addition.

Agents

An area outside the main conversation where an agentAI agent, n.: An AI system that does not just answer but acts: given a goal, it decides on steps, uses tools such as a browser or code editor, checks the results and keeps going until the job is done. records notes, plans, to-do lists or intermediate findings — often a file such as NOTES.md, a state object or a dedicated 'think' tool — and re-reads them later. It complements compactioncompaction, n.: Like replacing a stack of rough notes with a one-page digest: when an AI's conversation grows too long for its memory, the history is condensed into a summary and the work continues from there. and agent memoryagent memory, n.: The ways an AI assistant or agent keeps information — your preferences, past conversations, lessons from earlier tasks — and brings it back later, beyond what fits in a single conversation. by keeping progress available after older context has been summarized or cleared.