 |
ML Newsnew | past | best | rss | submit | about |
| The unreasonable effectiveness of pattern matching (arxiv.org) | |
| 7 points by headalgorithm 258 days ago | hide | past | pdf | 1 comment on HN | |
|
In plain words: Large language models were given sentences with nearly every content word replaced by nonsense, like "He dwushed a ghanc zawk," and asked to recover the meaning. They did so from sentence structure alone, showing pattern matching is a real ingredient of understanding. AbstractWe report on an astonishing ability of large language models (LLMs) to make sense of "Jabberwocky" language in which most or all content words have been randomly replaced by nonsense strings, e.g., translating "He dwushed a ghanc zawk" to "He dragged a spare chair". This result addresses ongoing controversies regarding how to best think of what LLMs are doing: are they a language mimic, a database, a blurry version of the Web? The ability of LLMs to recover meaning from structural patterns speaks to the unreasonable effectiveness of pattern-matching. Pattern-matching is not an alternative to "real" intelligence, but rather a key ingredient. | |
| add comment on HN |
They also tend to overextend this idea to human intelligence:
"We think a more promising approach lies in studying how our pattern-matching minds are extended by cognitive prostheses which allow us to formulate and manipulate progressively more abstract and larger patterns."
I would put it the other way around: how our cognitive system can delegate some of the processing to simple pattern matching prostheses? When I think of concepts such as LLMs there is more than pattern matching of all that I read about them, but there is a vague idea of what they are, how they are build and how I feel about them, and the name "LLM" is just a simple tag for this concept. I am wondering what cognitive scientist would make of this paper.