{
  "schema_version": 1,
  "generated_from": "Markdown front matter and bodies",
  "entries": [
    {
      "id": "alex-krizhevsky",
      "kind": "person",
      "title": "Alex Krizhevsky",
      "updated": "2026-10-05",
      "sources": [
        {
          "id": "alexnet-paper",
          "url": "https://proceedings.neurips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html",
          "title": "ImageNet Classification with Deep Convolutional Neural Networks",
          "checked": "2026-10-05"
        }
      ],
      "related": [
        "breakthroughs/imagenet-alexnet.md"
      ],
      "breakthroughs": [
        "breakthroughs/imagenet-alexnet.md"
      ],
      "path": "people/alex-krizhevsky.md",
      "content": "## Bio and education\nThe primary source checked here identifies Krizhevsky as an AlexNet paper author. Education and a current official personal/work profile were not verified in sources checked; do not infer them from name matches.\n\n## Contributions — documented\nFirst author of the 2012 AlexNet paper, coauthored with Ilya Sutskever and Geoffrey Hinton. The paper describes the network and its ImageNet classification results. [Source: alexnet-paper]\n\n## Key insight — interpretation\nThe work is a case study in combining scale, engineering, and an evaluation setting that made a capability legible.\n\n## Links\n- Paper: https://proceedings.neurips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html\n",
      "markdown": "---\nid: alex-krizhevsky\nkind: person\ntitle: Alex Krizhevsky\nupdated: 2026-10-05\nsources:\n  - id: alexnet-paper\n    url: https://proceedings.neurips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html\n    title: ImageNet Classification with Deep Convolutional Neural Networks\n    checked: 2026-10-05\nrelated:\n  - breakthroughs/imagenet-alexnet.md\nbreakthroughs:\n  - breakthroughs/imagenet-alexnet.md\n---\n## Bio and education\nThe primary source checked here identifies Krizhevsky as an AlexNet paper author. Education and a current official personal/work profile were not verified in sources checked; do not infer them from name matches.\n\n## Contributions — documented\nFirst author of the 2012 AlexNet paper, coauthored with Ilya Sutskever and Geoffrey Hinton. The paper describes the network and its ImageNet classification results. [Source: alexnet-paper]\n\n## Key insight — interpretation\nThe work is a case study in combining scale, engineering, and an evaluation setting that made a capability legible.\n\n## Links\n- Paper: https://proceedings.neurips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html\n"
    },
    {
      "id": "alphafold2",
      "kind": "breakthrough",
      "title": "AlphaFold2 — Protein Structure Prediction at Scale",
      "updated": "2026-10-05",
      "sources": [
        {
          "id": "alphafold-nature",
          "url": "https://www.nature.com/articles/s41586-021-03819-2",
          "title": "Highly accurate protein structure prediction with AlphaFold (Jumper et al., 2021)",
          "checked": "2026-10-05"
        },
        {
          "id": "alphafold-deepmind",
          "url": "https://deepmind.google/science/alphafold/",
          "title": "AlphaFold — Google DeepMind project and impact page",
          "checked": "2026-10-05"
        },
        {
          "id": "alphafold-db",
          "url": "https://alphafold.ebi.ac.uk/",
          "title": "AlphaFold Protein Structure Database — EMBL-EBI and Google DeepMind",
          "checked": "2026-10-05"
        }
      ],
      "related": [
        "breakthroughs/transformer.md",
        "breakthroughs/imagenet-alexnet.md"
      ],
      "contributors": [
        "people/john-jumper.md",
        "people/demis-hassabis.md"
      ],
      "path": "breakthroughs/alphafold.md",
      "content": "## What was blocked\nExperimental structure determination is labor-intensive, and known structures covered only a small fraction of protein sequences. The bottleneck limited structural biology and downstream investigation. Protein structure prediction had a long history, including community benchmarks such as CASP; the challenge was improving accuracy and coverage enough for practical use. [Source: alphafold-nature]\n\n## What changed\nAlphaFold2 combined learned representations of sequences and residue relationships with iterative refinement to predict 3D structures. In CASP14 it achieved accuracy competitive with experimental structures for many targets, according to the authors. The 2021 paper has 24 authors, led by John Jumper and Demis Hassabis; credit belongs to the team and prior structural-biology community, not two people alone. Google DeepMind and EMBL-EBI subsequently made a large predicted-structure database openly accessible. [Sources: alphafold-nature, alphafold-deepmind, alphafold-db]\n\n## Why it mattered — interpretation\nThe breakthrough connected a benchmarked prediction capability to broad access: a research result became infrastructure scientists could query. It reduced a structural-information bottleneck, enabling new hypotheses and prioritization, rather than replacing experimental biology. This is an impact interpretation supported by the paper and database/project materials.\n\n## Practical leverage\nUse predicted structures to generate or prioritize hypotheses, inspect likely domains and residues, and decide where experiments may be most informative. Check confidence metrics and biological context; treat the model as a guide to experiments rather than a definitive observation. For an AI deployment, the lesson is to pair a capability with accessible, well-documented infrastructure and domain-specific validation.\n\n## Limits and open questions\nPredicted structures are not experimental structures, and confidence varies across residues and regions. A static structure does not fully describe dynamics, complexes, cellular context, or function. Database coverage and model suitability vary by protein. Downstream biological conclusions still need domain expertise and, where appropriate, experimental validation.\n\n## Sources\n- [alphafold-nature] Primary paper: https://www.nature.com/articles/s41586-021-03819-2\n- [alphafold-deepmind] Project history, methods, and case studies: https://deepmind.google/science/alphafold/\n- [alphafold-db] Searchable database: https://alphafold.ebi.ac.uk/\n",
      "markdown": "---\nid: alphafold2\nkind: breakthrough\ntitle: AlphaFold2 — Protein Structure Prediction at Scale\nupdated: 2026-10-05\nsources:\n  - id: alphafold-nature\n    url: https://www.nature.com/articles/s41586-021-03819-2\n    title: Highly accurate protein structure prediction with AlphaFold (Jumper et al., 2021)\n    checked: 2026-10-05\n  - id: alphafold-deepmind\n    url: https://deepmind.google/science/alphafold/\n    title: AlphaFold — Google DeepMind project and impact page\n    checked: 2026-10-05\n  - id: alphafold-db\n    url: https://alphafold.ebi.ac.uk/\n    title: AlphaFold Protein Structure Database — EMBL-EBI and Google DeepMind\n    checked: 2026-10-05\nrelated:\n  - breakthroughs/transformer.md\n  - breakthroughs/imagenet-alexnet.md\ncontributors:\n  - people/john-jumper.md\n  - people/demis-hassabis.md\n---\n## What was blocked\nExperimental structure determination is labor-intensive, and known structures covered only a small fraction of protein sequences. The bottleneck limited structural biology and downstream investigation. Protein structure prediction had a long history, including community benchmarks such as CASP; the challenge was improving accuracy and coverage enough for practical use. [Source: alphafold-nature]\n\n## What changed\nAlphaFold2 combined learned representations of sequences and residue relationships with iterative refinement to predict 3D structures. In CASP14 it achieved accuracy competitive with experimental structures for many targets, according to the authors. The 2021 paper has 24 authors, led by John Jumper and Demis Hassabis; credit belongs to the team and prior structural-biology community, not two people alone. Google DeepMind and EMBL-EBI subsequently made a large predicted-structure database openly accessible. [Sources: alphafold-nature, alphafold-deepmind, alphafold-db]\n\n## Why it mattered — interpretation\nThe breakthrough connected a benchmarked prediction capability to broad access: a research result became infrastructure scientists could query. It reduced a structural-information bottleneck, enabling new hypotheses and prioritization, rather than replacing experimental biology. This is an impact interpretation supported by the paper and database/project materials.\n\n## Practical leverage\nUse predicted structures to generate or prioritize hypotheses, inspect likely domains and residues, and decide where experiments may be most informative. Check confidence metrics and biological context; treat the model as a guide to experiments rather than a definitive observation. For an AI deployment, the lesson is to pair a capability with accessible, well-documented infrastructure and domain-specific validation.\n\n## Limits and open questions\nPredicted structures are not experimental structures, and confidence varies across residues and regions. A static structure does not fully describe dynamics, complexes, cellular context, or function. Database coverage and model suitability vary by protein. Downstream biological conclusions still need domain expertise and, where appropriate, experimental validation.\n\n## Sources\n- [alphafold-nature] Primary paper: https://www.nature.com/articles/s41586-021-03819-2\n- [alphafold-deepmind] Project history, methods, and case studies: https://deepmind.google/science/alphafold/\n- [alphafold-db] Searchable database: https://alphafold.ebi.ac.uk/\n"
    },
    {
      "id": "ashish-vaswani",
      "kind": "person",
      "title": "Ashish Vaswani",
      "updated": "2026-10-05",
      "sources": [
        {
          "id": "transformer-paper",
          "url": "https://arxiv.org/abs/1706.03762",
          "title": "Attention Is All You Need",
          "checked": "2026-10-05"
        },
        {
          "id": "vaswani-profile",
          "url": "https://www.ashishvaswani.com/",
          "title": "Ashish Vaswani — official site",
          "checked": "2026-10-05"
        }
      ],
      "related": [
        "breakthroughs/transformer.md"
      ],
      "breakthroughs": [
        "breakthroughs/transformer.md"
      ],
      "path": "people/ashish-vaswani.md",
      "content": "## Bio and education\nVaswani is a computer scientist and entrepreneur. The official site and paper were retrieved; education details are not included here because not verified in the checked sources.\n\n## Contributions — documented\nFirst-listed author of the eight-author 2017 paper “Attention Is All You Need,” which introduced the Transformer architecture. [Sources: transformer-paper, vaswani-profile]\n\n## Key insight — interpretation\nArchitectural leverage can come from removing a serial dependency in the training process while preserving useful interactions across sequence positions.\n\n## Links\n- Official site: https://www.ashishvaswani.com/\n- Original paper: https://arxiv.org/abs/1706.03762\n",
      "markdown": "---\nid: ashish-vaswani\nkind: person\ntitle: Ashish Vaswani\nupdated: 2026-10-05\nsources:\n  - id: transformer-paper\n    url: https://arxiv.org/abs/1706.03762\n    title: Attention Is All You Need\n    checked: 2026-10-05\n  - id: vaswani-profile\n    url: https://www.ashishvaswani.com/\n    title: Ashish Vaswani — official site\n    checked: 2026-10-05\nrelated:\n  - breakthroughs/transformer.md\nbreakthroughs:\n  - breakthroughs/transformer.md\n---\n## Bio and education\nVaswani is a computer scientist and entrepreneur. The official site and paper were retrieved; education details are not included here because not verified in the checked sources.\n\n## Contributions — documented\nFirst-listed author of the eight-author 2017 paper “Attention Is All You Need,” which introduced the Transformer architecture. [Sources: transformer-paper, vaswani-profile]\n\n## Key insight — interpretation\nArchitectural leverage can come from removing a serial dependency in the training process while preserving useful interactions across sequence positions.\n\n## Links\n- Official site: https://www.ashishvaswani.com/\n- Original paper: https://arxiv.org/abs/1706.03762\n"
    },
    {
      "id": "demis-hassabis",
      "kind": "person",
      "title": "Demis Hassabis",
      "updated": "2026-10-05",
      "sources": [
        {
          "id": "alphafold-nature",
          "url": "https://www.nature.com/articles/s41586-021-03819-2",
          "title": "Highly accurate protein structure prediction with AlphaFold",
          "checked": "2026-10-05"
        },
        {
          "id": "alphafold-deepmind",
          "url": "https://deepmind.google/science/alphafold/",
          "title": "AlphaFold — Google DeepMind",
          "checked": "2026-10-05"
        }
      ],
      "related": [
        "breakthroughs/alphafold.md"
      ],
      "breakthroughs": [
        "breakthroughs/alphafold.md"
      ],
      "path": "people/demis-hassabis.md",
      "content": "## Bio and education\nGoogle DeepMind's official project history describes Hassabis's role in establishing the AlphaFold effort; the primary paper lists him among 24 authors. Education details are not provided here because they were not verified from the sources checked.\n\n## Contributions — documented\nCoauthor of the 2021 AlphaFold paper and a senior scientific leader associated with the project. Attribute the technical result to the full research team and broader structural-biology community. [Sources: alphafold-nature, alphafold-deepmind]\n\n## Key insight — interpretation\nOne path to impact is to choose a hard, valuable scientific bottleneck and build a sustained team effort around measurable external evaluation.\n\n## Links\n- Google DeepMind AlphaFold page: https://deepmind.google/science/alphafold/\n- Original paper: https://www.nature.com/articles/s41586-021-03819-2\n",
      "markdown": "---\nid: demis-hassabis\nkind: person\ntitle: Demis Hassabis\nupdated: 2026-10-05\nsources:\n  - id: alphafold-nature\n    url: https://www.nature.com/articles/s41586-021-03819-2\n    title: Highly accurate protein structure prediction with AlphaFold\n    checked: 2026-10-05\n  - id: alphafold-deepmind\n    url: https://deepmind.google/science/alphafold/\n    title: AlphaFold — Google DeepMind\n    checked: 2026-10-05\nrelated:\n  - breakthroughs/alphafold.md\nbreakthroughs:\n  - breakthroughs/alphafold.md\n---\n## Bio and education\nGoogle DeepMind's official project history describes Hassabis's role in establishing the AlphaFold effort; the primary paper lists him among 24 authors. Education details are not provided here because they were not verified from the sources checked.\n\n## Contributions — documented\nCoauthor of the 2021 AlphaFold paper and a senior scientific leader associated with the project. Attribute the technical result to the full research team and broader structural-biology community. [Sources: alphafold-nature, alphafold-deepmind]\n\n## Key insight — interpretation\nOne path to impact is to choose a hard, valuable scientific bottleneck and build a sustained team effort around measurable external evaluation.\n\n## Links\n- Google DeepMind AlphaFold page: https://deepmind.google/science/alphafold/\n- Original paper: https://www.nature.com/articles/s41586-021-03819-2\n"
    },
    {
      "id": "fei-fei-li",
      "kind": "person",
      "title": "Fei-Fei Li",
      "updated": "2026-10-05",
      "sources": [
        {
          "id": "fei-fei-profile",
          "url": "https://profiles.stanford.edu/fei-fei-li",
          "title": "Stanford profile — Fei-Fei Li",
          "checked": "2026-10-05"
        },
        {
          "id": "imagenet-challenge",
          "url": "https://image-net.org/challenges/LSVRC/2012/",
          "title": "ImageNet 2012 challenge",
          "checked": "2026-10-05"
        }
      ],
      "related": [
        "breakthroughs/imagenet-alexnet.md"
      ],
      "breakthroughs": [
        "breakthroughs/imagenet-alexnet.md"
      ],
      "path": "people/fei-fei-li.md",
      "content": "## Bio and education — documented\nStanford's official profile identifies Li as a computer scientist and describes her research and academic roles. Her profile documents a BS in physics from Princeton and a PhD in electrical engineering from Caltech. [Source: fei-fei-profile]\n\n## Contributions\nLi and collaborators initiated ImageNet as a large-scale visual database; she also presented the 2012 challenge results workshop according to the official challenge page. ImageNet's creation was collaborative, and the challenge infrastructure is distinct from AlexNet's competing model. [Sources: fei-fei-profile, imagenet-challenge]\n\n## Key insight — interpretation\nData infrastructure and shared evaluation can be field-shaping research contributions in their own right.\n\n## Links\n- Official Stanford profile: https://profiles.stanford.edu/fei-fei-li\n- ImageNet 2012 challenge: https://image-net.org/challenges/LSVRC/2012/\n",
      "markdown": "---\nid: fei-fei-li\nkind: person\ntitle: Fei-Fei Li\nupdated: 2026-10-05\nsources:\n  - id: fei-fei-profile\n    url: https://profiles.stanford.edu/fei-fei-li\n    title: Stanford profile — Fei-Fei Li\n    checked: 2026-10-05\n  - id: imagenet-challenge\n    url: https://image-net.org/challenges/LSVRC/2012/\n    title: ImageNet 2012 challenge\n    checked: 2026-10-05\nrelated:\n  - breakthroughs/imagenet-alexnet.md\nbreakthroughs:\n  - breakthroughs/imagenet-alexnet.md\n---\n## Bio and education — documented\nStanford's official profile identifies Li as a computer scientist and describes her research and academic roles. Her profile documents a BS in physics from Princeton and a PhD in electrical engineering from Caltech. [Source: fei-fei-profile]\n\n## Contributions\nLi and collaborators initiated ImageNet as a large-scale visual database; she also presented the 2012 challenge results workshop according to the official challenge page. ImageNet's creation was collaborative, and the challenge infrastructure is distinct from AlexNet's competing model. [Sources: fei-fei-profile, imagenet-challenge]\n\n## Key insight — interpretation\nData infrastructure and shared evaluation can be field-shaping research contributions in their own right.\n\n## Links\n- Official Stanford profile: https://profiles.stanford.edu/fei-fei-li\n- ImageNet 2012 challenge: https://image-net.org/challenges/LSVRC/2012/\n"
    },
    {
      "id": "geoffrey-hinton",
      "kind": "person",
      "title": "Geoffrey Hinton",
      "updated": "2026-10-05",
      "sources": [
        {
          "id": "hinton-profile",
          "url": "https://www.cs.toronto.edu/~hinton/",
          "title": "University of Toronto — Geoffrey Hinton",
          "checked": "2026-10-05"
        },
        {
          "id": "alexnet-paper",
          "url": "https://proceedings.neurips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html",
          "title": "AlexNet paper",
          "checked": "2026-10-05"
        }
      ],
      "related": [
        "breakthroughs/imagenet-alexnet.md"
      ],
      "breakthroughs": [
        "breakthroughs/imagenet-alexnet.md"
      ],
      "path": "people/geoffrey-hinton.md",
      "content": "## Bio and education — documented\nHinton is a computer scientist and cognitive psychologist whose work helped shape neural-network research. His official University of Toronto page records a BA in experimental psychology from Cambridge (1970) and a PhD in artificial intelligence from Edinburgh (1978). [Source: hinton-profile]\n\n## Contributions\nCoauthor of AlexNet with Alex Krizhevsky and Ilya Sutskever; the paper reports its 2012 ImageNet result. The broader contribution of his research career is not exhaustively summarized here. [Source: alexnet-paper]\n\n## Key insight — interpretation\nA useful lesson from the AlexNet episode is to study the enabling stack, not just the winning architecture: data, compute, optimization choices, and measurement interacted.\n\n## Links\n- Official profile: https://www.cs.toronto.edu/~hinton/\n- Work/paper: https://proceedings.neurips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html\n",
      "markdown": "---\nid: geoffrey-hinton\nkind: person\ntitle: Geoffrey Hinton\nupdated: 2026-10-05\nsources:\n  - id: hinton-profile\n    url: https://www.cs.toronto.edu/~hinton/\n    title: University of Toronto — Geoffrey Hinton\n    checked: 2026-10-05\n  - id: alexnet-paper\n    url: https://proceedings.neurips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html\n    title: AlexNet paper\n    checked: 2026-10-05\nrelated:\n  - breakthroughs/imagenet-alexnet.md\nbreakthroughs:\n  - breakthroughs/imagenet-alexnet.md\n---\n## Bio and education — documented\nHinton is a computer scientist and cognitive psychologist whose work helped shape neural-network research. His official University of Toronto page records a BA in experimental psychology from Cambridge (1970) and a PhD in artificial intelligence from Edinburgh (1978). [Source: hinton-profile]\n\n## Contributions\nCoauthor of AlexNet with Alex Krizhevsky and Ilya Sutskever; the paper reports its 2012 ImageNet result. The broader contribution of his research career is not exhaustively summarized here. [Source: alexnet-paper]\n\n## Key insight — interpretation\nA useful lesson from the AlexNet episode is to study the enabling stack, not just the winning architecture: data, compute, optimization choices, and measurement interacted.\n\n## Links\n- Official profile: https://www.cs.toronto.edu/~hinton/\n- Work/paper: https://proceedings.neurips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html\n"
    },
    {
      "id": "ilya-sutskever",
      "kind": "person",
      "title": "Ilya Sutskever",
      "updated": "2026-10-05",
      "sources": [
        {
          "id": "alexnet-paper",
          "url": "https://proceedings.neurips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html",
          "title": "AlexNet paper",
          "checked": "2026-10-05"
        }
      ],
      "related": [
        "breakthroughs/imagenet-alexnet.md"
      ],
      "breakthroughs": [
        "breakthroughs/imagenet-alexnet.md"
      ],
      "path": "people/ilya-sutskever.md",
      "content": "## Bio and education\nSources checked identify Sutskever as an author on AlexNet. Education and a personal official profile were not verified in sources checked.\n\n## Contributions — documented\nCoauthored the 2012 AlexNet paper with Alex Krizhevsky and Geoffrey Hinton. [Source: alexnet-paper]\n\n## Key insight — interpretation\nA researcher's trajectory should be represented through verifiable work links; shared influence is not authorship.\n\n## Links\n- AlexNet paper: https://proceedings.neurips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html\n",
      "markdown": "---\nid: ilya-sutskever\nkind: person\ntitle: Ilya Sutskever\nupdated: 2026-10-05\nsources:\n  - id: alexnet-paper\n    url: https://proceedings.neurips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html\n    title: AlexNet paper\n    checked: 2026-10-05\nrelated:\n  - breakthroughs/imagenet-alexnet.md\nbreakthroughs:\n  - breakthroughs/imagenet-alexnet.md\n---\n## Bio and education\nSources checked identify Sutskever as an author on AlexNet. Education and a personal official profile were not verified in sources checked.\n\n## Contributions — documented\nCoauthored the 2012 AlexNet paper with Alex Krizhevsky and Geoffrey Hinton. [Source: alexnet-paper]\n\n## Key insight — interpretation\nA researcher's trajectory should be represented through verifiable work links; shared influence is not authorship.\n\n## Links\n- AlexNet paper: https://proceedings.neurips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html\n"
    },
    {
      "id": "imagenet-alexnet",
      "kind": "breakthrough",
      "title": "ImageNet–AlexNet — Deep Learning's Computer-Vision Inflection",
      "updated": "2026-10-05",
      "sources": [
        {
          "id": "alexnet-paper",
          "url": "https://proceedings.neurips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html",
          "title": "ImageNet Classification with Deep Convolutional Neural Networks (Krizhevsky, Sutskever & Hinton, 2012)",
          "checked": "2026-10-05"
        },
        {
          "id": "imagenet-challenge",
          "url": "https://image-net.org/challenges/LSVRC/2012/",
          "title": "ImageNet Large Scale Visual Recognition Challenge 2012",
          "checked": "2026-10-05"
        },
        {
          "id": "hinton-profile",
          "url": "https://www.cs.toronto.edu/~hinton/",
          "title": "Geoffrey Hinton — University of Toronto profile",
          "checked": "2026-10-05"
        },
        {
          "id": "fei-fei-profile",
          "url": "https://profiles.stanford.edu/fei-fei-li",
          "title": "Fei-Fei Li — Stanford profile",
          "checked": "2026-10-05"
        }
      ],
      "related": [
        "breakthroughs/transformer.md",
        "breakthroughs/alphafold.md"
      ],
      "contributors": [
        "people/alex-krizhevsky.md",
        "people/ilya-sutskever.md",
        "people/geoffrey-hinton.md",
        "people/fei-fei-li.md"
      ],
      "path": "breakthroughs/imagenet-alexnet.md",
      "content": "## What was blocked\nComputer vision had strong hand-engineered pipelines, but learning robust hierarchical features from data at scale remained difficult. Progress was constrained by the interaction among labeled data, model capacity, computation, and comparable evaluation—not by a single missing invention.\n\n## What changed\nAlexNet's authors trained a deep convolutional network on ImageNet's 1.3-million-image, 1,000-class training set, using two GPUs, ReLUs, dropout, and data augmentation. At the 2012 ImageNet challenge, the system substantially outperformed competing entries; the paper reports top-5 error of 15.3% versus 26.2% for the second-best entry. The paper authors were **Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton**. [Sources: alexnet-paper, imagenet-challenge]\n\nImageNet and the ImageNet Large Scale Visual Recognition Challenge, led by a broader research community including Fei-Fei Li's Stanford group, supplied an unusually large labeled dataset and a public benchmark. AlexNet was one team's system and result; benchmark creation and scientific infrastructure are distinct contributions. [Sources: imagenet-challenge, fei-fei-profile]\n\n## Why it mattered — interpretation\nThe inflection was a system-level convergence: data + GPUs + deep convolutional architecture + training tricks + a competitive benchmark. AlexNet made the performance gap visible enough to shift research and industry attention toward learned representations at scale. It did not originate deep learning, convolutional networks, GPUs, or ImageNet individually.\n\n## Practical leverage\nFor a learning system, identify whether the constraint is data quality/quantity, compute, architecture, or measurement. A benchmark can coordinate effort and expose progress, but only if its labels and test setting reflect real use. When building a narrow vision feature today, begin with a pretrained model and representative evaluation rather than reproducing AlexNet-era training from scratch.\n\n## Limits and open questions\nImageNet label coverage and taxonomy are imperfect; benchmark performance is not the same as robust perception or deployment value. The 2012 result does not isolate the causal contribution of each ingredient. High compute and large labeled datasets were material advantages. Later models, datasets, and practices changed the field substantially.\n\n## Sources\n- [alexnet-paper] Full paper landing page: https://proceedings.neurips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html\n- [imagenet-challenge] Official 2012 challenge page: https://image-net.org/challenges/LSVRC/2012/\n- [hinton-profile] Official profile: https://www.cs.toronto.edu/~hinton/\n- [fei-fei-profile] Stanford profile: https://profiles.stanford.edu/fei-fei-li\n",
      "markdown": "---\nid: imagenet-alexnet\nkind: breakthrough\ntitle: ImageNet–AlexNet — Deep Learning's Computer-Vision Inflection\nupdated: 2026-10-05\nsources:\n  - id: alexnet-paper\n    url: https://proceedings.neurips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html\n    title: ImageNet Classification with Deep Convolutional Neural Networks (Krizhevsky, Sutskever & Hinton, 2012)\n    checked: 2026-10-05\n  - id: imagenet-challenge\n    url: https://image-net.org/challenges/LSVRC/2012/\n    title: ImageNet Large Scale Visual Recognition Challenge 2012\n    checked: 2026-10-05\n  - id: hinton-profile\n    url: https://www.cs.toronto.edu/~hinton/\n    title: Geoffrey Hinton — University of Toronto profile\n    checked: 2026-10-05\n  - id: fei-fei-profile\n    url: https://profiles.stanford.edu/fei-fei-li\n    title: Fei-Fei Li — Stanford profile\n    checked: 2026-10-05\nrelated:\n  - breakthroughs/transformer.md\n  - breakthroughs/alphafold.md\ncontributors:\n  - people/alex-krizhevsky.md\n  - people/ilya-sutskever.md\n  - people/geoffrey-hinton.md\n  - people/fei-fei-li.md\n---\n## What was blocked\nComputer vision had strong hand-engineered pipelines, but learning robust hierarchical features from data at scale remained difficult. Progress was constrained by the interaction among labeled data, model capacity, computation, and comparable evaluation—not by a single missing invention.\n\n## What changed\nAlexNet's authors trained a deep convolutional network on ImageNet's 1.3-million-image, 1,000-class training set, using two GPUs, ReLUs, dropout, and data augmentation. At the 2012 ImageNet challenge, the system substantially outperformed competing entries; the paper reports top-5 error of 15.3% versus 26.2% for the second-best entry. The paper authors were **Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton**. [Sources: alexnet-paper, imagenet-challenge]\n\nImageNet and the ImageNet Large Scale Visual Recognition Challenge, led by a broader research community including Fei-Fei Li's Stanford group, supplied an unusually large labeled dataset and a public benchmark. AlexNet was one team's system and result; benchmark creation and scientific infrastructure are distinct contributions. [Sources: imagenet-challenge, fei-fei-profile]\n\n## Why it mattered — interpretation\nThe inflection was a system-level convergence: data + GPUs + deep convolutional architecture + training tricks + a competitive benchmark. AlexNet made the performance gap visible enough to shift research and industry attention toward learned representations at scale. It did not originate deep learning, convolutional networks, GPUs, or ImageNet individually.\n\n## Practical leverage\nFor a learning system, identify whether the constraint is data quality/quantity, compute, architecture, or measurement. A benchmark can coordinate effort and expose progress, but only if its labels and test setting reflect real use. When building a narrow vision feature today, begin with a pretrained model and representative evaluation rather than reproducing AlexNet-era training from scratch.\n\n## Limits and open questions\nImageNet label coverage and taxonomy are imperfect; benchmark performance is not the same as robust perception or deployment value. The 2012 result does not isolate the causal contribution of each ingredient. High compute and large labeled datasets were material advantages. Later models, datasets, and practices changed the field substantially.\n\n## Sources\n- [alexnet-paper] Full paper landing page: https://proceedings.neurips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html\n- [imagenet-challenge] Official 2012 challenge page: https://image-net.org/challenges/LSVRC/2012/\n- [hinton-profile] Official profile: https://www.cs.toronto.edu/~hinton/\n- [fei-fei-profile] Stanford profile: https://profiles.stanford.edu/fei-fei-li\n"
    },
    {
      "id": "john-jumper",
      "kind": "person",
      "title": "John Jumper",
      "updated": "2026-10-05",
      "sources": [
        {
          "id": "alphafold-nature",
          "url": "https://www.nature.com/articles/s41586-021-03819-2",
          "title": "Highly accurate protein structure prediction with AlphaFold",
          "checked": "2026-10-05"
        },
        {
          "id": "alphafold-deepmind",
          "url": "https://deepmind.google/science/alphafold/",
          "title": "AlphaFold — Google DeepMind",
          "checked": "2026-10-05"
        }
      ],
      "related": [
        "breakthroughs/alphafold.md"
      ],
      "breakthroughs": [
        "breakthroughs/alphafold.md"
      ],
      "path": "people/john-jumper.md",
      "content": "## Bio and education\nThe primary paper identifies Jumper as first author and the Google DeepMind project page describes his role in AlphaFold. Education details are not stated here because not verified in the checked sources.\n\n## Contributions — documented\nLed the 2021 paper, one of a 24-author team, describing AlphaFold's highly accurate protein-structure predictions. The result builds on CASP and structural biology work across a larger community. [Sources: alphafold-nature, alphafold-deepmind]\n\n## Key insight — interpretation\nA benchmark breakthrough becomes more consequential when translated into usable scientific infrastructure and a way for domain experts to interrogate its output.\n\n## Links\n- Google DeepMind AlphaFold page: https://deepmind.google/science/alphafold/\n- Original paper: https://www.nature.com/articles/s41586-021-03819-2\n",
      "markdown": "---\nid: john-jumper\nkind: person\ntitle: John Jumper\nupdated: 2026-10-05\nsources:\n  - id: alphafold-nature\n    url: https://www.nature.com/articles/s41586-021-03819-2\n    title: Highly accurate protein structure prediction with AlphaFold\n    checked: 2026-10-05\n  - id: alphafold-deepmind\n    url: https://deepmind.google/science/alphafold/\n    title: AlphaFold — Google DeepMind\n    checked: 2026-10-05\nrelated:\n  - breakthroughs/alphafold.md\nbreakthroughs:\n  - breakthroughs/alphafold.md\n---\n## Bio and education\nThe primary paper identifies Jumper as first author and the Google DeepMind project page describes his role in AlphaFold. Education details are not stated here because not verified in the checked sources.\n\n## Contributions — documented\nLed the 2021 paper, one of a 24-author team, describing AlphaFold's highly accurate protein-structure predictions. The result builds on CASP and structural biology work across a larger community. [Sources: alphafold-nature, alphafold-deepmind]\n\n## Key insight — interpretation\nA benchmark breakthrough becomes more consequential when translated into usable scientific infrastructure and a way for domain experts to interrogate its output.\n\n## Links\n- Google DeepMind AlphaFold page: https://deepmind.google/science/alphafold/\n- Original paper: https://www.nature.com/articles/s41586-021-03819-2\n"
    },
    {
      "id": "oriol-vinyals",
      "kind": "person",
      "title": "Oriol Vinyals",
      "updated": "2026-10-05",
      "sources": [
        {
          "id": "transformer-paper",
          "url": "https://arxiv.org/abs/1706.03762",
          "title": "Attention Is All You Need",
          "checked": "2026-10-05"
        }
      ],
      "related": [
        "breakthroughs/transformer.md"
      ],
      "breakthroughs": [
        "breakthroughs/transformer.md"
      ],
      "path": "people/oriol-vinyals.md",
      "content": "## Bio and education\nThe checked primary source identifies Vinyals as a Transformer paper coauthor. Education and a personal official biography were not verified in sources checked.\n\n## Contributions — documented\nCoauthor of the 2017 “Attention Is All You Need” paper. The paper provides the verifiable contribution; this profile does not assign an individual component of the team design. [Source: transformer-paper]\n\n## Key insight — interpretation\nFor team science, credit the shared publication unless reliable evidence supports finer attribution.\n\n## Links\n- Paper: https://arxiv.org/abs/1706.03762\n",
      "markdown": "---\nid: oriol-vinyals\nkind: person\ntitle: Oriol Vinyals\nupdated: 2026-10-05\nsources:\n  - id: transformer-paper\n    url: https://arxiv.org/abs/1706.03762\n    title: Attention Is All You Need\n    checked: 2026-10-05\nrelated:\n  - breakthroughs/transformer.md\nbreakthroughs:\n  - breakthroughs/transformer.md\n---\n## Bio and education\nThe checked primary source identifies Vinyals as a Transformer paper coauthor. Education and a personal official biography were not verified in sources checked.\n\n## Contributions — documented\nCoauthor of the 2017 “Attention Is All You Need” paper. The paper provides the verifiable contribution; this profile does not assign an individual component of the team design. [Source: transformer-paper]\n\n## Key insight — interpretation\nFor team science, credit the shared publication unless reliable evidence supports finer attribution.\n\n## Links\n- Paper: https://arxiv.org/abs/1706.03762\n"
    },
    {
      "id": "transformer",
      "kind": "breakthrough",
      "title": "The Transformer — Attention Is All You Need",
      "updated": "2026-10-05",
      "sources": [
        {
          "id": "transformer-paper",
          "url": "https://arxiv.org/abs/1706.03762",
          "title": "Attention Is All You Need (Vaswani et al., 2017)",
          "checked": "2026-10-05"
        },
        {
          "id": "vaswani-profile",
          "url": "https://www.ashishvaswani.com/",
          "title": "Ashish Vaswani — official site",
          "checked": "2026-10-05"
        }
      ],
      "related": [
        "breakthroughs/imagenet-alexnet.md",
        "breakthroughs/alphafold.md"
      ],
      "contributors": [],
      "path": "breakthroughs/transformer.md",
      "content": "## What was blocked\nSequence-to-sequence systems commonly relied on recurrent or convolutional networks. Recurrence imposes sequential computation across positions, limiting parallelization during training. Prior attention mechanisms typically supplemented those architectures rather than replacing recurrence altogether. [Source: transformer-paper]\n\n## What changed\nThe 2017 paper proposed an encoder–decoder architecture based solely on attention, dispensing with recurrence and convolutions. It introduced multi-head self-attention, positional information, and feed-forward layers. On WMT 2014 English–German and English–French translation, the authors reported state-of-the-art BLEU results at the time, with training that was more parallelizable and less costly than compared systems. These are the authors' reported experimental results, not a claim that attention is universally superior. [Source: transformer-paper]\n\n## Contributors and lineage\nThe paper credits **all eight authors**: Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, and Illia Polosukhin. The listed people profiles here are a deliberately partial linked set, not a complete roster or contribution attribution. The paper itself situates its method among sequence transduction, attention, and self-attention work; the architecture emerged from a team and prior research, not a single-person invention. [Source: transformer-paper]\n\n## Why it mattered — interpretation\nThe key unlock was not “attention was invented,” but a clean architecture that made global token-to-token interaction and parallel training central. This proved reusable beyond machine translation and became a major substrate for later language and multimodal models. That broader historical influence is synthesis; the original paper directly supports the design and translation result, not every later consequence.\n\n## Practical leverage\nWhen a workload has long sequences and enough accelerator parallelism, compare attention-based models against recurrent/convolutional baselines for quality, throughput, memory, and data needs. For agent work, the leverage is architectural: pretraining one general sequence model can support many downstream tasks. Validate latency, cost, retrieval needs, and grounding for the actual application.\n\n## Limits and open questions\nSelf-attention's cost and memory grow quadratically with sequence length in the standard formulation. Parallel training does not remove data, compute, evaluation, or deployment constraints. The original translation benchmarks do not establish reasoning, truthfulness, or safety. Later architectures and efficient-attention methods address different tradeoffs; this note is not a survey of them.\n\n## Sources\n- [transformer-paper] The original paper and author list: https://arxiv.org/abs/1706.03762\n- [vaswani-profile] Ashish Vaswani's official site: https://www.ashishvaswani.com/\n",
      "markdown": "---\nid: transformer\nkind: breakthrough\ntitle: The Transformer — Attention Is All You Need\nupdated: 2026-10-05\nsources:\n  - id: transformer-paper\n    url: https://arxiv.org/abs/1706.03762\n    title: Attention Is All You Need (Vaswani et al., 2017)\n    checked: 2026-10-05\n  - id: vaswani-profile\n    url: https://www.ashishvaswani.com/\n    title: Ashish Vaswani — official site\n    checked: 2026-10-05\nrelated:\n  - breakthroughs/imagenet-alexnet.md\n  - breakthroughs/alphafold.md\ncontributors: []\n---\n## What was blocked\nSequence-to-sequence systems commonly relied on recurrent or convolutional networks. Recurrence imposes sequential computation across positions, limiting parallelization during training. Prior attention mechanisms typically supplemented those architectures rather than replacing recurrence altogether. [Source: transformer-paper]\n\n## What changed\nThe 2017 paper proposed an encoder–decoder architecture based solely on attention, dispensing with recurrence and convolutions. It introduced multi-head self-attention, positional information, and feed-forward layers. On WMT 2014 English–German and English–French translation, the authors reported state-of-the-art BLEU results at the time, with training that was more parallelizable and less costly than compared systems. These are the authors' reported experimental results, not a claim that attention is universally superior. [Source: transformer-paper]\n\n## Contributors and lineage\nThe paper credits **all eight authors**: Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, and Illia Polosukhin. The listed people profiles here are a deliberately partial linked set, not a complete roster or contribution attribution. The paper itself situates its method among sequence transduction, attention, and self-attention work; the architecture emerged from a team and prior research, not a single-person invention. [Source: transformer-paper]\n\n## Why it mattered — interpretation\nThe key unlock was not “attention was invented,” but a clean architecture that made global token-to-token interaction and parallel training central. This proved reusable beyond machine translation and became a major substrate for later language and multimodal models. That broader historical influence is synthesis; the original paper directly supports the design and translation result, not every later consequence.\n\n## Practical leverage\nWhen a workload has long sequences and enough accelerator parallelism, compare attention-based models against recurrent/convolutional baselines for quality, throughput, memory, and data needs. For agent work, the leverage is architectural: pretraining one general sequence model can support many downstream tasks. Validate latency, cost, retrieval needs, and grounding for the actual application.\n\n## Limits and open questions\nSelf-attention's cost and memory grow quadratically with sequence length in the standard formulation. Parallel training does not remove data, compute, evaluation, or deployment constraints. The original translation benchmarks do not establish reasoning, truthfulness, or safety. Later architectures and efficient-attention methods address different tradeoffs; this note is not a survey of them.\n\n## Sources\n- [transformer-paper] The original paper and author list: https://arxiv.org/abs/1706.03762\n- [vaswani-profile] Ashish Vaswani's official site: https://www.ashishvaswani.com/\n"
    }
  ]
}