Dualformer: Controllable Fast and Slow Thinking by Learning with Randomized Reasoning Traces · HackerPapers discussions

Dualformer: Controllable Fast and Slow Thinking by Learning with Randomized Reasoning Traces — HN submission

Fri, 25 Oct 2024 13:32:16 GMT

Dualformer: Controllable Fast/Slow Thinking,Learning with Rand. Reasoning Traces

2 points; 1 discussion reply on HN. Paper and source history. A submission is not an endorsement.

Dualformer: Controllable Fast and Slow Thinking by Learning with Randomized Reasoning Traces — mentioned by lessisgood123

Thu, 24 Oct 2024 23:20:02 GMT

Meta AI (FAIR) latest paper integrates system-1 and system-2 thinking into reasoning models. Basically, it introduces the term "Dualformer" which integrates both system-1 (fast-thinking) and system-2 (slow-thinking) into the transformer to improve its reasoning capability. The high level idea is to train the model with "randomized trace", which randomly drop parts of the reasoning tokens. This approach improves model's inference speed, accuracy, and diversity. It also enables model to perform system-1 and system-2 thinking in a controllable fashion. The paper's link here…

Paper and discussion history. A mention is not an endorsement.

Dualformer: Controllable Fast and Slow Thinking by Learning with Randomized Reasoning Traces — HN submission

Wed, 23 Oct 2024 03:51:45 GMT

Controllable Fast and Slow Thinking by Learning with Randomized Reasoning Traces

44 points; 15 discussion replies on HN. Paper and source history. A submission is not an endorsement.