# Jakob Uszkoreit > ~1986– · Computer Scientist, Transformer Co-author > > **Recorded contribution:** Transformer co-author; co-founded Inceptive (RNA design) ## How to use this dossier Read for a causal chain, not a hero story: inherited problem → contribution → mechanism → downstream capability → limit. Then close the page and complete the reconstruction exercise from memory. ## 1. Historical orientation German computer scientist Jakob Uszkoreit was one of eight authors of Attention Is All You Need at Google and a researcher in neural language systems. He later co-founded Inceptive, applying machine learning to RNA design. Public accounts often connect him to the proposal to center the Transformer project on self-attention, but the published architecture and experiments remain a team result. Uszkoreit's earlier neural-language work and Transformer co-authorship sit within a longer effort to make sequence representations parallel and context-sensitive. At Inceptive, the abstraction moved from natural-language tokens to biological sequences, where proposed RNA designs must satisfy physical and experimental constraints rather than textual plausibility. ## 2. The problem inherited Recurrent and convolutional sequence models constrained parallelism or required many steps for distant interactions; biomolecular design later posed another sequence-to-function search problem. ## 3. The central contribution Uszkoreit co-created the Transformer architecture and later helped build an institution applying learned sequence design to RNA therapeutics. ## 4. Reconstruct the mechanism 1. Map sequence items into content and position representations. 2. Compute pairwise attention scores so distant items interact in one layer. 3. Use multiple heads to represent different learned relations and stack transformations. 4. For design, score candidate biological sequences against predicted functional objectives and experimental evidence. ## 5. What changed downstream - Self-attention became a general computational primitive across modalities. - His later work illustrates the transfer of sequence-model methods from language into biological design. - This trajectory illustrates how attention architectures became transferable infrastructure and how AI laboratories began coupling sequence generation to wet-lab validation in biotechnology. ## 6. Attribution, limits, and uncertainty - The Transformer is an eight-author result; retrospective anecdotes about who proposed which step should not override published joint attribution. - Biological-model predictions require experimental validation, and company strategy or therapeutic success is volatile and not established by architecture alone. - Public stories about who first proposed self-attention are retrospective and incomplete; RNA candidates also require safety, manufacturability, delivery, biological validation, and regulatory evidence beyond model scores. ## 7. Reconstruction lab Construct a four-token sequence where a distant dependency matters and compare one recurrent pass with one attention layer. Count path length and computation, not just output accuracy. Define a sequence objective with conflicting structure and manufacturability constraints, then require an experimental-validation stage that can reject the model's highest-scoring candidate. ## 8. Evidence trail - [Attention Is All You Need](https://arxiv.org/abs/1706.03762) — NeurIPS - [Inceptive](https://www.inceptive.life/) — Inceptive - [Transformer proceedings record](https://papers.nips.cc/paper/7181-attention-is-all-you-need) — NeurIPS --- *Research checked 2026-08-09. Dates, roles, and claims about living people are historical snapshots. Linked sources remain the authority; this dossier is original instructional synthesis.*