<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>TKDB — The Kernel Daily Blog</title><description>TKDB is an independent technology publishing platform: multiple blogs, original papers, engineering projects, and videos in one place.</description><link>https://tkdb.pages.dev/</link><item><title>Looking back at 2017&apos;s Transformer: the translation model that dropped recurrence for attention (AI Breakthroughs)</title><link>https://tkdb.pages.dev/blogs/ai-breakthroughs/transformer-2017/</link><guid isPermaLink="true">https://tkdb.pages.dev/blogs/ai-breakthroughs/transformer-2017/</guid><description>A retrospective on &quot;Attention Is All You Need&quot;: what the 2017 paper actually showed, how an independent lab reproduced it, how it spread to GPT and BERT, and the limits that remain.</description><pubDate>Sun, 25 Oct 2026 00:00:00 GMT</pubDate><category>transformers</category><category>attention</category><category>machine-translation</category><category>retrospective</category><category>independently-replicated</category></item><item><title>One ion, no measurements: USTC reports break-even for autonomous error correction, but the baseline is contested (Quantum Computing)</title><link>https://tkdb.pages.dev/blogs/quantum-computing/single-ion-autonomous-qec/</link><guid isPermaLink="true">https://tkdb.pages.dev/blogs/quantum-computing/single-ion-autonomous-qec/</guid><description>A USTC team corrected dephasing in a qubit stored across four levels of one calcium ion without measuring anything. It outlived the best qubit stored in the same levels, though another group questions that comparison.</description><pubDate>Sun, 25 Oct 2026 00:00:00 GMT</pubDate><category>quantum-error-correction</category><category>trapped-ions</category><category>quantum-memory</category><category>peer-reviewed</category></item><item><title>Graham&apos;s 55-year-old rearrangement conjecture is settled for all large primes (Science &amp; Mathematics)</title><link>https://tkdb.pages.dev/blogs/science-mathematics/graham-rearrangement-conjecture/</link><guid isPermaLink="true">https://tkdb.pages.dev/blogs/science-mathematics/graham-rearrangement-conjecture/</guid><description>Four papers from partly independent groups together show that any set of nonzero numbers on a prime-sized clock can be ordered so no partial sum repeats, as long as the prime is large. The final piece is still a preprint.</description><pubDate>Sun, 25 Oct 2026 00:00:00 GMT</pubDate><category>combinatorics</category><category>additive-combinatorics</category><category>probabilistic-method</category><category>graham-conjecture</category><category>preprint</category></item></channel></rss>