Chain of Thought

Reasoning models, represented by OpenAI o1, DeepSeek-R1, and QwQ, have chain-of-thought expansion as their core mechanism, generating a large number of intermediate reasoning steps before producing a final answer.

Key techniques in Prompt Engineering include role-playing setup, few-shot learning, structured constraints, and chain-of-thought.

The process of motivated reasoning occurs within the internal chain of thought, making it difficult for external observers to monitor and detect.

Reasoning models perform internal Chain-of-Thought before returning final results, consuming large amounts of tokens and computation time in exchange for deeper understanding and lower error rates.

The core idea of the test-time compute paradigm is to shift resources from the training phase to the answer generation process, trading more inference compute for higher-quality output.

Chain-of-Thought and Tree-of-Thoughts are reasoning frameworks that enable models to systematically analyze multi-layered causal relationships.

Test-Time Compute refers to the mechanism by which new-generation models perform longer internal reasoning via chain-of-thought before generating a final answer.

Test-time compute scaling improves performance on complex reasoning tasks by investing more computation during the inference stage through methods such as extending the chain of thought, multi-path sampling with best-of-n selection, and self-verification reflection.

Chain-of-Thought technology can improve accuracy on complex tasks, but at the cost of multiplying token generation, further lengthening response time.

Jason Wei's 2022 paper demonstrated that giving 8 examples in the prompt enabled a 540-billion-parameter model to reach the then-best level on the GSM8K math word problem benchmark.

The effectiveness of CoT depends heavily on whether the model possesses the underlying knowledge needed to complete the task; if the model lacks the relevant social common sense, more reasoning steps merely spin on false premises.

Adding long-term memory and summarization modules alone made the villagers' win rate lower than the chain-of-thought baseline (villagers won about four out of ten games, winning one fewer game with memory alone).

In defining distillation attacks, Anthropic noted that distillers want to generate millions of responses, seeking not only the final answers but also the chain of thought by which the model arrives at them.

In traditional CoT, even if a reasoning step is wrong, subsequent steps may still be corrected by global context, thereby masking the propagation path of the error.

Methods such as Chain-of-Thought prompting and Test-time Compute were introduced to mitigate the problem of broken reasoning chains.

Native chain-of-thought easily degenerates into surface correlations or shallow pattern matching when handling user simulation tasks, and cannot truly capture the behavioral factors that drive real users' decisions.

Adding the instruction "please calculate step by step" (chain of thought) to a request can raise the accuracy of a large model on complex arithmetic such as multiplying a six-digit number by a seven-digit number to above 90%.

Chain of Thought (CoT) is a prompting and reasoning method proposed by Wei et al. in 2022.

The line in the video, "we thought a chain of thought might trigger a chain reaction that would destroy the entire world", uses the shared word root to stitch together chain of thought and chain reaction, creating the implication that the stronger the reasoning capability, the greater the risk.

The reasoning text generated by chain-of-thought does not guarantee a faithful reflection of the model's true basis for decisions and may be rationalizations written after the conclusion; the reasoning process should not be treated as reliable audit evidence.