Lower Bounds for Chain-of-Thought Reasoning in Hard-Attention Transformers · HackerPapers discussions

Lower Bounds for Chain-of-Thought Reasoning in Hard-Attention Transformers — mentioned by nyrikki

Thu, 17 Jul 2025 14:20:52 GMT

Subbarao Kambhampati and his team has several papers strongly suggesting that the intermediate tokens are of no symantic import.[0] My hunch is that training explicitly for human acceptance of the intermediate/scratch tokens will end up being expensive. What you need is to converge to the correct answer, and scratch space dramatically improves expressivity[1], but the relationship to learning and retrieval is still open unless I missed something. Remember even access to a pure random Oracle dramatically improves compute time on some problems. [0] https://arxiv.org/abs/2505.13775 [1]…

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

Lower Bounds for Chain-of-Thought Reasoning in Hard-Attention Transformers — mentioned by nyrikki

Sat, 29 Mar 2025 16:27:47 GMT

As an intuition on why many people see this as different. PAC Learning is about compression, KF/EKF is more like Taylor expansion. The specific types of PAC Learning that this paper covers has problems with a simplicity bias, and fairly low sensitivity. While based on UHATs, this paper may provide some insights. https://arxiv.org/abs/2502.02393 Obviously LLM and LRMs are the most studied, but even the recent posts on here from anthropic show that without a few high probability entries in the k-top results, confabulations are difficult for transformers. Obviously there are PAC Learning methods…

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

Lower Bounds for Chain-of-Thought Reasoning in Hard-Attention Transformers — mentioned by nyrikki

Thu, 27 Mar 2025 18:07:12 GMT

In the field of AI wishful mnemonics is the rule, arguments about prescription are of limited value. CoT scratch space extends LLMs from DLOGTIME-UNIFORM TC0 to PTIME with polynomial sized scratch space. https://arxiv.org/abs/2502.02393 Yes prompt engineering is probably better though of as prompt augmentation or stearing. But the systems identification problem and Rice's theorm rigorously debunk the above links core claims. It is a craft that can improve domain specificity and usefulness. All models are wrong (even formalized engineering ones), but some are useful. The price one has to pay…

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

Lower Bounds for Chain-of-Thought Reasoning in Hard-Attention Transformers — mentioned by nyrikki

Mon, 10 Mar 2025 17:39:42 GMT

We can't even define what intelligence is. We know or have strong hints at the limits of math/computation related to LLMs + CoT Note how PARITY and MEDIAN is hard here: https://arxiv.org/abs/2502.02393 We also know HALT == open frame == symbol grounding == system identification problems. The definition of AGI is also not well defined, but given the following: > Strong AI, also called artificial general intelligence, refers to machines possessing generalized intelligence and capabilities on par with human cognition. We know enough for any mechanical methods with either current machines or even…

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

Lower Bounds for Chain-of-Thought Reasoning in Hard-Attention Transformers — mentioned by nyrikki

Fri, 07 Mar 2025 18:12:27 GMT

Subbarao Kambhampati, who seems to only use X is a good resource. He points out how the CoT text is not of semantic importantce. This work from his team shows how few 'reasoning' traces are valid. https://atharva.gundawar.com/searchformer_response_analysis.... This paper shows how the scratch space gets transformers to PTIME from TC0 without it. https://arxiv.org/abs/2502.02393 OpenAI may be able to do more in the long term because they don't show the <think> and can spend more of that scratch space on improving answers vs appeasing users, but time will show. Remember that probabilistic…

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