Formal Language Recognition by Hard Attention Transformers: Perspectives from Circuit Complexity · HackerPapers discussionshttps://hackerpaper.kaipfister.de/p/2204.06618/Collected HN comments and submissions linking this paper.enMon, 02 Nov 2026 11:16:22 GMTFormal Language Recognition by Hard Attention Transformers: Perspectives from Circuit Complexity — mentioned by nyrikkihttps://news.ycombinator.com/item?id=39487869urn:hackerpapers:mention:2204.06618:39487869Sat, 24 Feb 2024 00:23:34 GMT<p>No, just a programmer that has to apply math. Transformers can only do what is called weak negation. if (not (goal X)), then (assert not x) One example I have seen is: &quot;you can cross if you have no information on a train coming” VS: &quot;you can cross if you have evidence that no train is coming&quot; See the difference? As to what &#x27;transformers&#x27; can encode is ambiguous and depends on many factors but here is some light reading. https://direct.mit.edu/tacl/article/doi/10.1162/tacl_a_00493... https://arxiv.org/abs/2204.06618</p><p><a href="https://hackerpaper.kaipfister.de/p/2204.06618/">Paper and discussion history</a>. A mention is not an endorsement.</p>cs.CC