https://hackerpaper.kaipfister.de/p/2602.04101/Collected HN comments and submissions linking this paper.enMon, 02 Nov 2026 11:16:22 GMThttps://news.ycombinator.com/item?id=48101159urn:hackerpapers:mention:2602.04101:48101159Mon, 11 May 2026 21:53:44 GMT<p>Here’s the academic paper behind it: https://arxiv.org/abs/2602.04101</p><p><a href="https://hackerpaper.kaipfister.de/p/2602.04101/">Paper and discussion history</a>. A mention is not an endorsement.</p>cs.AIhttps://news.ycombinator.com/item?id=47603506urn:hackerpapers:submission:2602.04101:47603506Wed, 01 Apr 2026 17:00:07 GMT<p>Show HN: A new model architecture because transformers are not enough</p><p>Most AI products are built around the same idea. Take a giant model. Add a prompt and hope it does the whole job. The AI equivalent of spray-n-pray. That works great for chatbots for creative outputs like code generation, website design and email writing. BUT, It breaks down fast for developer work that requires highly deterministic output. Think of OCR for KYC at banks or Audio recognition for patient-doctor notes. We have been training SLMs for specific tasks for a while now. During which we kept seeing the same pattern. A SOTA language model with great benchmarks, failing on the parts that…</p><p>3 points; 2 discussion replies on HN. <a href="https://hackerpaper.kaipfister.de/p/2602.04101/">Paper and source history</a>. A submission is not an endorsement.</p>cs.AI