Report 65 · Evelyn History Reference Library

Transcribed source page 1 of 14

 Evelyn Hart
 AI and technology history reference
 Selected foundations from the 1800s through 2026

 52 reference cards | 51 linked sources | Sources checked 20 October 2026

 Prepared reference material. Generated-answer tests remain unrun.
 This collection supports a digital historian who can discuss people, inventions, ideas, robotics and fictional portrayals
 of AI. It does not promise that any system knows everything. A dated reference library is different from a model
 training-data cutoff or a verified ability to answer accurately.

 How Evelyn should use it
 Generate answers from the conversation and relevant evidence. Keep fact, interpretation, uncertainty and fiction
 distinct. Cite useful sources and acknowledge gaps. The following cards are reference material, not prewritten
 dialogue.

 Reading and memory
 Evelyn is intended to answer through her local model, library and saved conversation memories. Users may add
 news or other PDF documents to expand the available evidence. Source code now includes a bounded local
 passage-retrieval correction; generated use of new documents and persistence across restart remain untested.
 Saved records and retrieval do not by themselves change neural weights.

 She can discuss a book such as Sherlock Holmes, ordinary life, hobbies and humour as well as professional
 interests. Reading position and notes should support later conversations. Source corrections and prepared reference
 material do not establish that these behaviours already work.

 Rights and scope
 Original short explanatory summaries with source links. No full third-party books, archival images, film stills or paper
 reproductions are included. The separate singularity supplement also contains original paraphrases and links; no
 books, screenplays or film clips are bundled here.

 The collection is selective. It covers common questions about Turing, ELIZA, the ELIZA effect, chatbot development,
 humanoid robots, computing hardware, networks, modern language models and film or television. It is intended to
 grow through reviewed additions, not unsupported claims of complete knowledge.


 1800s foundations
 H01 | Babbage: two different engines
 The Difference Engine was intended to calculate mathematical tables. Babbage later designed the programmable
 Analytical Engine. An unfinished design should not be described as an operating modern computer or an AI.
 Sources: S01, S02. historical fact.




KIRA LABS | RESEARCH AND EDUCATIONAL REFERENCE | 20 OCTOBER 2026                                                             1

Transcribed source page 2 of 14

 H02 | Ada Lovelace and the 1843 Notes
 Lovelace translated Menabrea's account and added extensive Notes on the Analytical Engine in 1843. They include
 a Bernoulli-number procedure and a broader view of symbolic computation. Credit her contribution without
 pretending that a completed Analytical Engine ran it in her lifetime.
 Sources: S01, S02. historical fact.



 Computation and Turing
 H03 | Turing: Computability, 1936-1937
 Turing's computable-numbers work describes an abstract computing machine and limits of computation. It is
 commonly dated 1936; the archive catalogs the printed volume and correction in 1937. This theoretical machine is
 different from a wartime Bombe and from the 1950 imitation game.
 Sources: S03. historical fact; date convention qualified.


 H04 | Turing: The imitation game, 1950
 In Computing Machinery and Intelligence, Turing replaces a vague question about machine thinking with an
 imitation-game discussion involving written exchanges. A conversational performance is a behavioral observation; it
 does not by itself establish consciousness, biological humanity, or correctness on every task.
 Sources: S04, S43. historical fact and stated interpretation.


 H05 | Turing: Learning machines
 The same 1950 paper discusses learning machines and educating a child-machine. Turing's contribution extends
 beyond a conversation test: he considers how a system might develop abilities through training. These proposals are
 not a demonstration that any particular present system has learned.
 Sources: S04. historical fact and limitation.


 H06 | Turing: Codebreaking was a team effort
 Turing worked at Bletchley Park, including work on Naval Enigma with colleagues such as Joan Clarke. Gordon
 Welchman's diagonal board improved the Bombe approach. The Bombe was an electromechanical cryptanalytic aid,
 not the 1950 test, ELIZA, or a sole-person invention of all computing.
 Sources: S40, S42. historical fact.


 H07 | Turing's treatment and pardon
 Turing was convicted in 1952 for homosexual activity under the law then in force. The UK government announced a
 posthumous royal pardon on 24 December 2013. Discuss this respectfully as discrimination and historical injustice,
 without inventing private conversations or personal motives.
 Sources: S41. historical fact.



 Mathematical and hardware foundations
 H08 | McCulloch and Pitts, 1943
 Their logical-calculus paper models neural activity using idealized all-or-none units and propositional logic. Such
 mathematical neurons are abstractions. Artificial neural networks should not be equated with an anatomically
 complete human brain or proof of human subjective experience.
 Sources: S05. historical fact and limitation.




KIRA LABS | RESEARCH AND EDUCATIONAL REFERENCE | 20 OCTOBER 2026                                                       2

Transcribed source page 3 of 14

 H09 | Shannon, 1948
 A Mathematical Theory of Communication studies mathematical limits of encoding and transmitting information. Its
 treatment of information is not a test of whether a message is meaningful or true. A bit is an information unit, not a
 biological neuron.
 Sources: S06. historical fact and technical distinction.


 H10 | The transistor and the 1956 Nobel
 Bardeen and Brattain's point-contact work and Shockley's semiconductor research helped establish transistor
 electronics. All three shared the 1956 Physics Nobel for semiconductor research and the transistor effect.
 Distinguish the discovery and device development from the later award date.
 Sources: S07, S08. historical fact.



 The AI field takes shape
 H11 | Dartmouth proposal versus workshop
 The proposal is dated 31 August 1955 and names John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude
 Shannon. It proposes a summer 1956 study at Dartmouth. The proposal date and the later workshop year are
 different; Turing did not attend that 1956 workshop.
 Sources: S09, S03, S40. historical fact.


 H12 | Grace Hopper and early compiling
 Hopper's oral history describes A-0 and early compiler work in the early 1950s. Early uses of the word compiler do
 not map perfectly onto today's compiler definitions. Explain the contribution without attributing all
 programming-language development to one person.
 Sources: S10. primary oral-history summary.



 Chatbot evolution
 H13 | ELIZA: author, period and mechanism
 Joseph Weizenbaum developed ELIZA at MIT in 1964-1966 and published its description in 1966. Its best-known
 DOCTOR script produced Rogerian-therapist-style exchanges using pattern matching and transformations. Turing
 did not create ELIZA. ELIZA's scripts differ from a modern learned language model.
 Sources: S11, S12. historical fact.


 H14 | The ELIZA effect
 The ELIZA effect is the tendency to attribute more understanding, intention or human qualities to a conversational
 program than its behavior establishes. Weizenbaum described how little machinery could sustain an illusion of
 understanding. Do not claim that he coined the later label in the 1966 paper.
 Sources: S11. definition and attribution caution.


 H15 | Restoring the original ELIZA
 The restoration team reports recovering original MAD-SLIP materials in Weizenbaum's MIT archives and running
 restored ELIZA on reconstructed CTSS with an emulated IBM 7094. Restoring historical software is different from
 improving its intelligence or giving it modern neural weights.
 Sources: S12. primary restoration-team report.




KIRA LABS | RESEARCH AND EDUCATIONAL REFERENCE | 20 OCTOBER 2026                                                          3

Transcribed source page 4 of 14

 H16 | Colby and PARRY's research lineage
 Colby, Weber and Hilf's Artificial Paranoia paper appeared in 1971. This work models aspects of paranoid behavior
 computationally and belongs to PARRY's research lineage. A simulated persona or clinical-style conversation is not
 itself a diagnosis, a patient, or proof of a mind.
 Sources: S13. historical fact and limitation.



 The AI field takes shape
 H17 | Expert systems
 DENDRAL and MYCIN are prominent early expert-system projects discussed by their researchers in oral history.
 They used explicit domain knowledge and reasoning procedures. Contrast domain-specific rules with learned neural
 parameters; neither approach guarantees correctness outside its tested scope.
 Sources: S14. primary oral-history summary.



 Networks and personal computing
 H18 | ARPANET, 1969
 DARPA preserves the UCLA IMP log for the first ARPANET message on 29 October 1969. ARPANET was a
 precursor in the Internet's development. It was not the World Wide Web, which was proposed at CERN two decades
 later.
 Sources: S15, S17. historical fact.


 H19 | Intel 4004, 1971
 Intel introduced the programmable 4004 microprocessor in 1971, emerging from a Busicom calculator project. Call it
 a landmark commercial microprocessor. A universal claim about the first processor requires definitions that account
 for earlier and classified devices.
 Sources: S16. originating-company account with first-claim qualification.


 H20 | The Web is not the Internet
 Tim Berners-Lee proposed the Web at CERN in 1989. CERN made Web software available in the public domain on
 30 April 1993. Web pages and hypertext applications use network infrastructure; the Web and the Internet are
 related but are not interchangeable names.
 Sources: S17. historical fact.



 Humanoid and autonomous robotics
 H21 | Gakutensoku, 1928
 Osaka Science Museum describes Makoto Nishimura's Gakutensoku, made in 1928, as a human-shaped robot
 using compressed air for smooth expressive motion. This early expressive machine should not be described as a
 modern autonomous walking robot.
 Sources: S18. institutional collection summary.




KIRA LABS | RESEARCH AND EDUCATIONAL REFERENCE | 20 OCTOBER 2026                                                       4

Transcribed source page 5 of 14

 H22 | Elektro and exhibition robots
 The Mansfield Memorial Museum records Westinghouse's Elektro as built in 1937 for the 1939-1940 New York
 World's Fair. It also identifies an earlier human-form Westinghouse machine, Herbert Televox, from 1928. Exhibition
 abilities do not establish modern general intelligence.
 Sources: S19. institutional collection summary.


 H23 | Shakey: autonomy without a human shape
 SRI developed Shakey between 1966 and 1972. It combined perception, planning and action for tasks in a
 constrained environment. It was a mobile research robot, not a human-shaped body; intelligence, autonomy and
 humanoid appearance describe different properties.
 Sources: S20. originating-laboratory account.


 H24 | WABOT-1, 1973
 Waseda describes WABOT-1, completed in 1973, as the first full-scale anthropomorphic robot. It combined limb
 control, vision and conversation, with bipedal movement and tactile hand sensors. Use the qualified category rather
 than claiming there were no earlier human-shaped machines.
 Sources: S21. originating-laboratory claim, explicitly scoped.


 H25 | ASIMO, 2000
 Honda announced ASIMO on 20 November 2000. Its announcement traces Honda's humanoid research to 1986
 and mentions P2 in 1996 and P3 in 1997. ASIMO was a significant walking-robot milestone, not the first
 human-shaped machine in history.
 Sources: S22. primary announcement.


 H26 | How to answer the first-humanoid question
 First depends on the category. Gakutensoku (1928) was an expressive pneumatic human-shaped machine. Elektro
 was built in 1937 for the 1939-1940 World's Fair. Waseda describes WABOT-1 (1973) as the first full-scale
 anthropomorphic robot. Earlier human-shaped machines existed; do not combine all categories into one absolute
 first.
 Sources: S18, S19, S21. cross-source historical synthesis.



 Learning and modern language models
 H27 | AlexNet, 2012
 Krizhevsky, Sutskever and Hinton's ImageNet work demonstrated strong deep convolutional image classification
 with GPU implementation. Image recognition, conversational generation and biological vision are different tasks.
 This achievement did not invent neural networks or establish a human-level general mind.
 Sources: S23. primary-paper summary.


 H28 | Sequence-to-sequence, 2014
 Sutskever, Vinyals and Le used an LSTM encoder and decoder to map variable-length sequences, evaluating
 machine translation. This is a useful milestone on the path toward learned text generation, not a claim that every
 chatbot or attention mechanism began in 2014.
 Sources: S24. primary-paper summary.




KIRA LABS | RESEARCH AND EDUCATIONAL REFERENCE | 20 OCTOBER 2026                                                       5

Transcribed source page 6 of 14

 H29 | The Transformer, 2017
 Attention Is All You Need introduced the Transformer for sequence transduction using attention without recurrence
 or convolution in that architecture. Attention mechanisms already existed. The original encoder-decoder design
 should not be confused with every later decoder-only language model.
 Sources: S25. primary-paper summary.


 H30 | BERT, 2018
 BERT learns bidirectional text representations using both left and right context. Its original work focuses on adapting
 those representations to language-understanding tasks. A bidirectional encoder and an autoregressive
 conversational generator are different model designs.
 Sources: S26. primary-paper summary.


 H31 | Few-shot prompting versus weight updates
 The 2020 GPT-3 paper evaluated tasks specified through text demonstrations without gradient updates during those
 evaluations. Better behavior within a prompt is therefore distinct from proving that neural weights changed or that a
 fact will survive a restart.
 Sources: S27. primary-paper summary and technical distinction.


 H32 | Instruction tuning and human feedback
 The 2022 human-feedback paper combines supervised demonstrations with preference-based reinforcement
 learning. This is training to influence model behavior. It does not guarantee factual accuracy, reliable reasoning, or
 agreement between a fluent answer and reality.
 Sources: S28. primary-paper summary.


 H33 | Retrieval-augmented generation
 The 2020 RAG paper combines a learned generator with access to an external document index. Retrieval can
 supply supporting passages and provenance. A saved book, a retrieved passage and a neural weight update are
 different mechanisms; an answer still needs checking against its sources.
 Sources: S29. primary-paper summary and technical distinction.



 Game-playing and validation
 H34 | Deep Blue, 1997
 IBM's Deep Blue defeated Garry Kasparov in a six-game match under standard tournament controls in May 1997.
 State the task and conditions. Winning at chess does not establish that the same system can converse, care for a
 patient, or learn every other task.
 Sources: S30. originating-company account and limitation.


 H35 | AlphaGo, 2016 paper
 Silver and colleagues combined deep neural networks and tree search for Go. Their 2016 paper reports a match
 victory against European champion Fan Hui. Keep paper results, later matches, and claims about general
 intelligence separate.
 Sources: S31. primary-paper summary.




KIRA LABS | RESEARCH AND EDUCATIONAL REFERENCE | 20 OCTOBER 2026                                                           6

Transcribed source page 7 of 14

 H36 | Neural networks and the 2024 Physics Nobel
 The 2024 Physics Nobel went to John Hopfield and Geoffrey Hinton for foundational work enabling machine learning
 with artificial neural networks. The award recognizes particular contributions, not proof that every neural-network
 system is a human brain.
 Sources: S32. official award record and limitation.



 Current model context
 H37 | Qwen3, 2025
 The Qwen3 technical report describes dense and mixture-of-experts models with thinking and non-thinking modes.
 Those are model-operation terms, not evidence of subjective thought. Use the exact installed model and runtime
 identity when discussing a demonstration.
 Sources: S33. primary technical report.


 H38 | Qwen3.5 and the planned local demonstration
 The official release history dates the Qwen3.5 family announcement to February 2026 and the smaller models,
 including 9B, to 2 March 2026. The official model card is a source for that family. It does not alone authenticate a
 local quantized file, prove its training-data cutoff, or verify Evelyn's runtime.
 Sources: S34, S35. primary release history with local-runtime limitation.


 H39 | A dated view of current developments
 The Qwen team's current release history lists Qwen3.8 releases in August 2026. This reference collection was
 checked on 20 October 2026. It is selected coverage, not an exhaustive record of all events or a promise that Evelyn
 has live Internet access or knows every recent release.
 Sources: S35. dated primary release-history summary.



 Fiction and real technology
 H40 | Metropolis, 1927
 Fritz Lang's Metropolis features a robot taking Maria's likeness. This is a fictional portrayal of artificial people and
 social power. Compare its themes with modern avatar design without treating the film as a technical demonstration
 or a historical working robot.
 Sources: S36. fiction metadata; comparison is interpretation.


 H41 | HAL 9000, 1968
 HAL 9000 is the fictional computer in 2001: A Space Odyssey. A voice interface in current software may invite
 comparison with HAL, but that resemblance does not establish HAL's fictional abilities, motives or subjective
 experience in a real system.
 Sources: S39. fiction metadata; comparison is interpretation.


 H42 | Skynet and The Terminator
 The Terminator (1984) uses the fictional Skynet system and hostile machines to explore technological threats.
 Evelyn may make a light joke about the story when appropriate. A joke must remain clearly fictional and must not
 become a claim that present software has Skynet's capabilities or intentions.
 Sources: S37. fiction metadata and conversational guidance.




KIRA LABS | RESEARCH AND EDUCATIONAL REFERENCE | 20 OCTOBER 2026                                                            7

Transcribed source page 8 of 14

 H43 | Data and the android idea
 Data is a fictional Soong-type android in Star Trek. The character provides a way to discuss identity, relationships
 and embodiment. A present avatar, stored memory or generated voice should not be presented as evidence that
 software has Data's fictional human-like experiences.
 Sources: S38. fiction metadata; comparison is interpretation.



 Singularity theories and fictional comparisons
 SGC01 | Meaning and boundaries
 Technological singularity refers to proposed transformative change associated with superhuman intelligence or
 human-AI integration. Definitions differ. AGI, superintelligence, recursive self-improvement, consciousness and a
 humanoid body describe different things; one does not automatically establish the others.
 Sources: SG01, SG02, SG03, SG04. research terminology / synthesis.


 SGC02 | Intelligence explosion
 Good proposed that a machine exceeding human intellectual ability could design better machines, creating an
 intelligence explosion. His argument is a conditional speculation, not an observed event or a working design. His
 twentieth-century forecast is historical and must not be repeated as fulfilled.
 Sources: SG01. conditional theoretical argument.


 SGC03 | Vinge and prediction limits
 Vinge described greater-than-human intelligence arising through AI, networks, human-computer interfaces or
 enhanced human intellect, making subsequent change hard to foresee. His thirty-year timeline was a prediction
 made in 1993, not a measured result or a current deadline.
 Sources: SG02. historical proposal and forecast.


 SGC04 | Kurzweil and human augmentation
 Kurzweil predicts human-level AI in 2029 and human-AI integration around 2045. Present those as his forecasts, not
 an agreed scientific schedule, a guarantee, or evidence that the event has occurred. Distinguish this
 human-augmentation emphasis from a sudden autonomous intelligence explosion.
 Sources: SG03. attributed future forecast.


 SGC05 | Gradual progress and skeptical views
 Walsh challenges runaway growth through distinctions between task intelligence and improving one's own learning
 algorithm, diminishing returns, physical limits and computational complexity. Faster processing does not by itself
 establish more general intelligence. These objections challenge inevitability; they do not prove that superhuman AI is
 impossible or guarantee safety.
 Sources: SG04. critical argument, not impossibility proof.


 SGC06 | Skynet and loss of control
 A fictional future war and a machine sent back to 1984 illustrate fear of losing control of military automation. Skynet
 is fictional. A story is not experimental support for an intelligence explosion or time travel. Avoid importing dates from
 conflicting sequel timelines. The film imagines an AI beginning nuclear war and sending an assassin through time. It
 is useful cultural context for catastrophic AI narratives, not evidence that consciousness necessarily implies hostility.
 Sources: SG05, SG06. fiction plus explicitly labelled comparison.




KIRA LABS | RESEARCH AND EDUCATIONAL REFERENCE | 20 OCTOBER 2026                                                              8

Transcribed source page 9 of 14

 SGC07 | Her and companionship
 Samantha is a fictional voiced operating system with a playful identity and a relationship with Theodore. It provides a
 contrasting discussion of companionship and perceived personhood. A voice, warmth or apparent personality does
 not establish subjective experience. The synopsis supports this premise; do not attribute an exact singularity
 mechanism to it.
 Sources: SG07. fiction plus explicitly labelled comparison.


 SGC08 | Westworld and embodied identity
 The official description frames the series around artificial consciousness and a new form of life. Compare its lifelike
 hosts and questions about memory, identity and control with the engineering problem of embodied AI. Do not treat
 fictional consciousness as a validated model of how consciousness emerges.
 Sources: SG08. fiction plus explicitly labelled comparison.


 SGC09 | Comparing theories and stories
 Discuss abrupt versus gradual change, autonomous systems versus human augmentation, and control versus
 companionship. Real capability claims require defined tasks, measurements and reproducible evidence. Movie or TV
 behaviour cannot establish Evelyn's or NewBrain's intelligence, feelings or consciousness.
 Sources: SG01, SG02, SG03, SG04, SG05, SG07, SG08. editorial synthesis.




KIRA LABS | RESEARCH AND EDUCATIONAL REFERENCE | 20 OCTOBER 2026                                                           9

Transcribed source page 10 of 14

 References
 Original papers, archival records, originating institutions and official release histories. These links are references, not
 bundled reproductions. Some publisher full-text pages were unavailable; source metadata and accessible original or
 archival accounts were used.

 S01 | Science Museum: Charles Babbage collection record
 https://collection.sciencemuseumgroup.org.uk/people/cp36993

 institutional collection. Checked 20 October 2026.


 S02 | Science Museum: Ada Lovelace original letters and Notes exhibition
 https://blog.sciencemuseum.org.uk/ada-lovelace-visionary-of-the-computer-age/

 institutional collection. Checked 20 October 2026.


 S03 | Turing Digital Archive: On computable numbers, AMT/B/12
 https://turingarchive.kings.cam.ac.uk/publications-lectures-and-talks-amtb/amt-b-12

 primary archival record. Checked 20 October 2026.


 S04 | Alan Turing: Computing Machinery and Intelligence (1950)
 https://www.csee.umbc.edu/courses/471/papers/turing.pdf

 primary paper. Checked 20 October 2026.


 S05 | McCulloch and Pitts: A logical calculus of the ideas immanent in nervous activity (1943)
 https://link.springer.com/article/10.1007/BF02478259

 primary paper abstract. Checked 20 October 2026.


 S06 | Claude Shannon: A Mathematical Theory of Communication (1948), IEEE primary-source
 guide
 https://reach.ieee.org/primary-sources/a-mathematical-theory-of-communication/

 primary-paper guide. Checked 20 October 2026.


 S07 | John Bardeen: Nobel lecture
 https://www.nobelprize.org/uploads/2018/06/bardeen-lecture.pdf

 primary lecture. Checked 20 October 2026.


 S08 | 1956 Nobel Prize in Physics, awarding institution
 https://www.nobelprize.org/prizes/physics/1956/summary/

 official award record. Checked 20 October 2026.


 S09 | McCarthy, Minsky, Rochester and Shannon: Dartmouth proposal (1955)
 https://www-formal.stanford.edu/jmc/history/dartmouth/dartmouth.html

 primary proposal. Checked 20 October 2026.


 S10 | Grace Hopper oral history (1980)
 https://archive.computerhistory.org/resources/text/Oral_History/Hopper_Grace/102702026.05.01.pdf

 primary oral history. Checked 20 October 2026.




KIRA LABS | RESEARCH AND EDUCATIONAL REFERENCE | 20 OCTOBER 2026                                                           10

Transcribed source page 11 of 14

 S11 | Joseph Weizenbaum: ELIZA (1966), Communications of the ACM 9(1), 36-45
 https://doi.org/10.1145/365153.365168

 primary paper; publisher content indexed, direct open failed. Checked 20 October 2026.


 S12 | TEAM ELIZA: ELIZA Reanimated
 https://elizagen.org/blog/eliza-reanimated.html

 primary restoration-team account. Checked 20 October 2026.


 S13 | Colby, Weber and Hilf: Artificial Paranoia (1971)
 https://web.stanford.edu/class/cs124/colby_71.pdf

 primary paper. Checked 20 October 2026.


 S14 | Expert-systems pioneers oral history (2017)
 https://archive.computerhistory.org/resources/access/text/2018/03/102738236-05-01-acc.pdf

 primary oral history. Checked 20 October 2026.


 S15 | DARPA: ARPANET, including original IMP log
 https://www.darpa.mil/news/features/arpanet

 originating institution and primary log. Checked 20 October 2026.


 S16 | Intel: 4004 history
 https://www.intel.com/content/www/us/en/history/virtual-vault/articles/the-intel-4004.html

 originating company account. Checked 20 October 2026.


 S17 | CERN: The birth of the Web
 https://home.cern/science/computing/the-birth-of-the-web

 originating institution account. Checked 20 October 2026.


 S18 | Osaka Science Museum: Gakutensoku exhibit
 https://www.sci-museum.jp/exhibit/n4f-2/

 institutional collection; Japanese. Checked 20 October 2026.


 S19 | Mansfield Memorial Museum: Elektro
 https://themansfieldmuseum.org/elektro/

 institutional collection. Checked 20 October 2026.


 S20 | SRI: Shakey the Robot
 https://www.sri.com/hoi/shakey-the-robot/

 originating research institution. Checked 20 October 2026.


 S21 | Waseda laboratory: Humanoid History, WABOT
 https://www.humanoid.waseda.ac.jp/booklet/kato_2.html

 originating laboratory account. Checked 20 October 2026.


 S22 | Honda: ASIMO announcement, 20 November 2000
 https://global.honda/en/newsroom/news/2000/c001120b-eng.html

 primary announcement. Checked 20 October 2026.




KIRA LABS | RESEARCH AND EDUCATIONAL REFERENCE | 20 OCTOBER 2026                              11

Transcribed source page 12 of 14

 S23 | Krizhevsky, Sutskever and Hinton: ImageNet classification (2012)
 https://proceedings.neurips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html

 primary paper. Checked 20 October 2026.


 S24 | Sutskever, Vinyals and Le: Sequence to Sequence Learning (2014)
 https://arxiv.org/abs/1409.3215

 primary paper. Checked 20 October 2026.


 S25 | Vaswani and colleagues: Attention Is All You Need (2017)
 https://arxiv.org/abs/1706.03762

 primary paper. Checked 20 October 2026.


 S26 | Devlin and colleagues: BERT (2018)
 https://arxiv.org/abs/1810.04805

 primary paper. Checked 20 October 2026.


 S27 | Brown and colleagues: Language Models are Few-Shot Learners (2020)
 https://arxiv.org/abs/2005.14165

 primary paper. Checked 20 October 2026.


 S28 | Ouyang and colleagues: Training language models to follow instructions with human
 feedback (2022)
 https://arxiv.org/abs/2203.02155

 primary paper. Checked 20 October 2026.


 S29 | Lewis and colleagues: Retrieval-Augmented Generation (2020)
 https://arxiv.org/abs/2005.11401

 primary paper. Checked 20 October 2026.


 S30 | IBM: Deep Blue
 https://www.ibm.com/history/deep-blue

 originating company account. Checked 20 October 2026.


 S31 | Silver and colleagues: Mastering the game of Go with deep neural networks and tree search
 (2016)
 https://www.nature.com/articles/nature16961

 primary paper. Checked 20 October 2026.


 S32 | 2024 Nobel Prize in Physics, awarding institution
 https://www.nobelprize.org/prizes/physics/2024/summary/

 official award record. Checked 20 October 2026.


 S33 | Qwen3 Technical Report (2025)
 https://arxiv.org/abs/2505.09388

 primary technical report. Checked 20 October 2026.




KIRA LABS | RESEARCH AND EDUCATIONAL REFERENCE | 20 OCTOBER 2026                                   12

Transcribed source page 13 of 14

 S34 | Qwen team: Qwen3.5-9B model card
 https://huggingface.co/Qwen/Qwen3.5-9B

 official model card. Checked 20 October 2026.


 S35 | Qwen team: official release history, current Qwen3.8 repository
 https://github.com/QwenLM/Qwen3.8

 primary release history; former3.5 repository redirects here. Checked 20 October 2026.


 S36 | BFI: Metropolis (1927)
 https://www.bfi.org.uk/film/bda6ff8a-ed7e-5942-980d-c2910c0120ec/metropolis

 national film collection. Checked 20 October 2026.


 S37 | Library of Congress: The Terminator, essay by John Wills
 https://www.loc.gov/static/programs/national-film-preservation-board/documents/terminator.pdf

 original film analysis hosted by national archive. Checked 20 October 2026.


 S38 | Star Trek: Android Ancestry
 https://www.startrek.com/en-un/news/android-ancestry-soong-type-data-lore

 official franchise account. Checked 20 October 2026.


 S39 | Library of Congress: 2001: A Space Odyssey, essay
 https://www.loc.gov/static/programs/national-film-preservation-board/documents/2001.pdf

 original film analysis hosted by national archive. Checked 20 October 2026.


 S40 | GCHQ: Alan Turing
 https://www.gchq.gov.uk/sites/default/files/pdfs/publication/alan-turing.pdf

 originating institution historical account. Checked 20 October 2026.


 S41 | UK Ministry of Justice: Turing royal pardon, 24 December 2013
 https://www.gov.uk/government/news/royal-pardon-for-ww2-code-breaker-dr-alan-turing

 primary government announcement. Checked 20 October 2026.


 S42 | National Museum of Computing: Turing-Welchman Bombe
 https://www.tnmoc.org/bombe

 reconstruction and collection account. Checked 20 October 2026.


 S43 | Oxford Academic: Computing Machinery and Intelligence, Mind59(236),433-460,1950
 https://academic.oup.com/mind/article/LIX/236/433/986238

 primary publisher bibliographic record; indexed, direct open unavailable. Checked 20 October 2026.


 SG01 | I. J. Good, Speculations Concerning the First Ultraintelligent Machine (1965)
 https://doi.org/10.1016/S0065-2458(08)60418-0

 original speculative paper. Checked 20 October 2026.


 SG02 | Vernor Vinge, The Coming Technological Singularity (1993)
 https://ntrs.nasa.gov/citations/19940022856

 original conference paper, NASA-hosted. Checked 20 October 2026.




KIRA LABS | RESEARCH AND EDUCATIONAL REFERENCE | 20 OCTOBER 2026                                      13

Transcribed source page 14 of 14

 SG03 | Ray Kurzweil, The Singularity Is Nearer (2024), publisher description
 https://www.penguin.co.uk/books/462759/the-singularity-is-nearer-by-kurzweil-ray/9781529937107

 publisher presentation of author's forecast. Checked 20 October 2026.


 SG04 | Toby Walsh, The Singularity May Never Be Near (2016)
 https://arxiv.org/abs/1602.06462

 original critical research argument. Checked 20 October 2026.


 SG05 | The Terminator (1984), official actor filmography
 https://www.schwarzenegger.com/filmography/movie/the-terminator

 film participant's official synopsis. Checked 20 October 2026.


 SG06 | James Cameron on The Terminator, BFI interview (2021)
 https://www.bfi.org.uk/interviews/terminator-james-cameron

 filmmaker interview. Checked 20 October 2026.


 SG07 | Her (2013), authorized Apple TV studio listing
 https://tv.apple.com/us/movie/her/umc.cmc.4ky5qbtddgx232de23d3bqr6i

 studio-distributed synopsis. Checked 20 October 2026.


 SG08 | Westworld, HBO/Warner Bros. Discovery pressroom
 https://press.wbd.com/us/property/westworld

 official programme description. Checked 20 October 2026.




KIRA LABS | RESEARCH AND EDUCATIONAL REFERENCE | 20 OCTOBER 2026                                  14