Arwa

arwaadel.bsky.social

17 October 2026 13:47

😍

ScienceConcierge編集部

hydezero-message.bsky.social

16 October 2026 14:08

What remains after light? A theory tracks electron wave phases to calculate orbital angular momentum in a graphene model. Transient peaks differ from residual response; this is not a memory device or proof of permanent storage. https://note.com/hydezero/n/nea6886ec7d05?hl=en #Physics #Graphene

Mohammed Tahri

mohammedtahri.bsky.social

14 October 2026 22:26

Phase 11 done: Transformers & Vector Spaces, as a journal issue. Five claims checked against code and papers: the analogy needs its inputs hidden, close isn't synonym, attention predates 2017, weights are contested clues, the FFN holds 2/3 of an encoder layer. #MT24Learning #Transformers #NLP

Mohammed Tahri

mohammedtahri.bsky.social

14 October 2026 10:56

Sprint 35/90: Inside a Transformer. I built one layer in numpy. In each encoder layer the FFN holds 0.67 of the weights, not attention. With no mask and no position codes, "dog bites man" and "man bites dog" look like one set of words. #MT24Learning #Transformers #DeepLearning

Mohammed Tahri

mohammedtahri.bsky.social

13 October 2026 11:30

Sprint 34/90: How Attention Works. I ran softmax(QKᵀ/√dk)V on real GloVe vectors. In "she took a loan from the bank", loan climbs from bank's 52nd neighbour to 12th. Without √dk, softmax picks one winner (0.95). Weights are clues, not reasons. #MT24Learning #Transformers #NLP