{ "source_file": "0930", "source_platform": "Substack", "author": "AI by Hand ✍️ Prof. Tom Yeh", "title_listed": "Transformer: Six Levels of Understanding", "keywords": "Transformer,理解层次", "domain": "大模型算法", "content_type": "教程", "is_paid_flag": "否", "url": "https://www.byhand.ai/p/transformer-six-levels-of-understanding", "publish_date": "2026-02", "corpus_stage": "预训练/SFT", "notes": "Transformer六层理解", "id": "773bcd9301ac", "collected_at": "2026-10-01T20:20:38", "title": "Transformer - Six Levels of Understanding", "body_markdown": "Library\n›\nSeminar Series 2026\nManifold-Constrained Hyper Connections (mHC) from DeepSeek (Jan 9, 2026)\nHow Small Models Learn Tool Use (Jan 11, 2026)\nGenerative AI (Jan 11, 2026)\nAttention (Jan 15, 2026)\nGoogle Ironwood TPU: From Bits to HBM (Jan 19, 2026)\n9 AI Eval Formulas You Must Know (Jan 24, 2026)\nMeta Superintelligence Labs vs Facebook AI Research (Jan 30, 2026)\nTransformer: Six Levels of Understanding (Feb 13, 2026)\nOpenClaw Seminar (Feb 23, 2026)\nPPO → DPO → GRPO → Rubrics (Mar 2, 2026)\nGemma 4 (May 7, 2026)\nQwen 3.6 (May 21, 2026)\nKimi 3 (Aug 6, 2026)\nIn this seminar, I taught Transformers. Instead of jumping straight into attention diagrams and equations, I used the opportunity to guide everyone through six levels of understanding.\nWe first practiced on something small and concrete, an\nartificial neuron\n, going deeper level by level from (1) black box to (2) components, to (3) tensors, to (4) math, to (5) Excel, and finally to (6) coding. Then we applied the exact same progression to the\ntransformer\n. My goal was to show the audience how to fish, not just hand them a fish, so they can use this framework to unpack any architecture they encounter next.\nArtificial Neuron\nLevel 1, Black Box:\nYou see an artificial neuron as a function that takes inputs and produces a single output.\nLevel 2, Components:\nYou identify its internal parts, inputs, weights, bias, activation function, and output.\nLevel 3, Tensors (AI by Hand ✍️):\nYou trace how each input aligns with a corresponding weight, forms a dot product, adds bias, and passes through an activation gate.\nLevel 4, Math:\nYou express it symbolically as\ny = wx + b\nand understand what each symbol represents.\nLevel 5, Excel:\nYou replicate the full computation step by step in Excel, without libraries, abstractions, or training wheels, so you can directly observe how changing numbers affects the output.\nLevel 6, Coding:\nYou implement it in code only after going through the earlier levels and\nearning the right to code\n, so clearly verbalizing its logic, and you can even use a coding model like OpenAI’s Codex or an Anthropic’s Opus, so long as you can describe what you want.\nTransformer\nLevel 1, Black Box:\nYou see a transformer as a function that takes tokens as input and produces the next-token probability distribution as output.\nLevel 2, Components:\nYou identify its internal parts, embeddings, positional encoding, self-attention, feed-forward networks, residual connections, layer normalization, and the final linear + softmax head.\nLevel 3, Tensors (AI by Hand ✍️):\nYou trace how tokens become vectors, how encoder layers can run in parallel during\nprefill\n, how decoder layers must run one token at a time during\ndecoding\n, and how shapes stay consistent as representations flow through attention, cross-attention, and feed-forward layers.\n**Level 4, Math:**You write the equations directly on top of the tensor visualization and understand what each matrix multiplication corresponds to.\nLevel 5, Excel:\nYou replicate attention, projection, and softmax step by step in Excel, without libraries, abstractions, or training wheels, so you can directly observe how probabilities emerge from matrix operations.\nLevel 6, Coding\n: You move to implementation only after progressing through the earlier levels and earning the right to build, able to explain embeddings, attention flow, projection, and sampling in plain language, provided you can specify the mechanics clearly rather than leaning on high-level buzzwords.\nFeedback\nRecording & Workbook\nMembers:\nthe full recording and the Excel workbook are below ↓", "metadata": { "title": "Transformer - Six Levels of Understanding", "description": "AI by Hand ✍️ Seminar Series", "keywords": null, "author": "Prof. Tom Yeh", "og:type": "article", "og:title": "Transformer - Six Levels of Understanding", "og:description": "AI by Hand ✍️ Seminar Series", "og:image": "https://substackcdn.com/image/fetch/$s_!azrY!,w_1200,h_675,c_fill,f_jpg,q_auto:good,fl_progressive:steep,g_auto/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F78b1e600-14fd-4181-8b0d-c65d833603f2_1650x1275.gif", "og:url": "https://www.byhand.ai/p/transformer-six-levels-of-understanding", "twitter:title": "Transformer - Six Levels of Understanding", "twitter:description": "AI by Hand ✍️ Seminar Series", "twitter:image": "https://substackcdn.com/image/fetch/$s_!gie7!,f_auto,q_auto:best,fl_progressive:steep/https%3A%2F%2Faibyhand.substack.com%2Fapi%2Fv1%2Fpost_preview%2F187799771%2Ftwitter.jpg%3Fversion%3D4", "twitter:card": "summary_large_image", "article:modified_time": "2026-08-20T14:29:11.404Z" }, "paywall": { "listed_paid": false, "detected": true, "marker": "this post is for paid subscribers", "public_only": true, "policy": "付费/会员内容只保留页面公开可见部分，不绕过付费墙" }, "crawl": { "success": true, "status_code": 200, "error": "" }, "code_blocks": [], "images": [ { "status": "ok", "url": "https://substackcdn.com/image/fetch/$s_!azrY!,w_1456,c_limit,f_auto,q_auto:good,fl_lossy/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F78b1e600-14fd-4181-8b0d-c65d833603f2_1650x1275.gif", "bytes": 634097, "reused": true, "local_path": "attachments/773bcd9301ac/images/026.gif", "src": "https://substackcdn.com/image/fetch/$s_!azrY!,w_1456,c_limit,f_auto,q_auto:good,fl_lossy/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F78b1e600-14fd-4181-8b0d-c65d833603f2_1650x1275.gif", "alt": "", "downloaded": true }, { "status": "ok", "url": "https://substackcdn.com/image/fetch/$s_!ios9!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Feb60b69d-fab0-4409-ba4e-9e27415964b2_1106x590.png", "bytes": 70226, "reused": true, "local_path": "attachments/773bcd9301ac/images/037.png", "src": "https://substackcdn.com/image/fetch/$s_!ios9!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Feb60b69d-fab0-4409-ba4e-9e27415964b2_1106x590.png", "alt": "", "downloaded": true }, { "status": "ok", "url": "https://substackcdn.com/image/fetch/$s_!e0qc!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa038fc96-f7f9-4282-99cc-2b33a1d93e19_1400x714.png", "bytes": 67712, "reused": false, "local_path": "attachments/773bcd9301ac/images/body-54c7a5ad4a368d6c924b.png", "src": "https://substackcdn.com/image/fetch/$s_!e0qc!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa038fc96-f7f9-4282-99cc-2b33a1d93e19_1400x714.png", "alt": "", "downloaded": true }, { "status": "ok", "url": "https://substackcdn.com/image/fetch/$s_!UKJA!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff5488826-9407-4a2c-8130-1f46f504e7fa_1414x1388.png", "bytes": 149620, "reused": false, "local_path": "attachments/773bcd9301ac/images/body-1ef559c0ac9bcd493bc0.png", "src": "https://substackcdn.com/image/fetch/$s_!UKJA!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff5488826-9407-4a2c-8130-1f46f504e7fa_1414x1388.png", "alt": "", "downloaded": true }, { "status": "ok", "url": "https://substackcdn.com/image/fetch/$s_!yN--!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Feddd0a54-c23b-4aa2-80c7-78a17bb52201_1702x854.png", "bytes": 126322, "reused": false, "local_path": "attachments/773bcd9301ac/images/body-e95b26a7abbd38a1e024.png", "src": "https://substackcdn.com/image/fetch/$s_!yN--!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Feddd0a54-c23b-4aa2-80c7-78a17bb52201_1702x854.png", "alt": "", "downloaded": true }, { "status": "ok", "url": "https://substackcdn.com/image/fetch/$s_!VDAT!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fab1fc962-f7b4-44f3-b014-e63289dc2436_1628x1692.png", "bytes": 383915, "reused": false, "local_path": "attachments/773bcd9301ac/images/body-eb97065b33e730af735d.png", "src": "https://substackcdn.com/image/fetch/$s_!VDAT!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fab1fc962-f7b4-44f3-b014-e63289dc2436_1628x1692.png", "alt": "", "downloaded": true } ], "audio": [], "videos": [], "links": [ { "href": "https://www.byhand.ai/p/library", "label": "Library", "type": "a" }, { "href": "https://www.byhand.ai/p/library-series-videos-seminars-2026", "label": "Seminar Series 2026", "type": "a" }, { "href": "https://www.byhand.ai/p/manifold-constrained-hyper-connections", "label": "Manifold-Constrained Hyper Connections (mHC) from DeepSeek (Jan 9, 2026)", "type": "a" }, { "href": "https://www.byhand.ai/p/how-small-models-learn-tool-use", "label": "How Small Models Learn Tool Use (Jan 11, 2026)", "type": "a" }, { "href": "https://www.byhand.ai/p/introduction-to-gen-ai", "label": "Generative AI (Jan 11, 2026)", "type": "a" }, { "href": "https://www.byhand.ai/p/seminar-foundation-attention-to-frontier", "label": "Attention (Jan 15, 2026)", "type": "a" }, { "href": "https://www.byhand.ai/p/google-ironwood-tpu-from-bits-to", "label": "Google Ironwood TPU: From Bits to HBM (Jan 19, 2026)", "type": "a" }, { "href": "https://www.byhand.ai/p/ai-eval-equations-you-must-know", "label": "9 AI Eval Formulas You Must Know (Jan 24, 2026)", "type": "a" }, { "href": "https://www.byhand.ai/p/meta-superintelligence-labs-vs-facebook", "label": "Meta Superintelligence Labs vs Facebook AI Research (Jan 30, 2026)", "type": "a" }, { "href": "https://www.byhand.ai/p/openclaw-seminar", "label": "OpenClaw Seminar (Feb 23, 2026)", "type": "a" }, { "href": "https://www.byhand.ai/p/recording-ppo-dpo-grpo-rubrics", "label": "PPO → DPO → GRPO → Rubrics (Mar 2, 2026)", "type": "a" }, { "href": "https://www.byhand.ai/p/gemma-4-seminar-recording", "label": "Gemma 4 (May 7, 2026)", "type": "a" }, { "href": "https://www.byhand.ai/p/qwen-3-6-seminar-recording", "label": "Qwen 3.6 (May 21, 2026)", "type": "a" }, { "href": "https://www.byhand.ai/p/library-videos-seminars-2026-kimi-3-seminar-recording", "label": "Kimi 3 (Aug 6, 2026)", "type": "a" } ], "repair_version": 2, "repaired_at": "2026-10-14T10:47:13Z", "source_url": "https://www.byhand.ai/p/transformer-six-levels-of-understanding", "papers": [], "actual_video_files": [], "body_chars": 3597, "expected_body_images": 6, "discovery_status": "ok", "delivery_status": "ok", "missing_reason": "", "raw_html": "attachments/773bcd9301ac/pages/source.html", "local_html": "attachments/773bcd9301ac/pages/article.html", "body_base_path": "attachments/773bcd9301ac/pages/", "body_markdown_path": "attachments/773bcd9301ac/pages/article.md" }