[
  {"id":"C01","time":"01:00–01:28","seconds":60,"title":"Enigma and the wartime computer chronology","claim":"The transcript connects Enigma, wartime codebreaking and a digital computer through Turing to 1950.","status":"Correction needed in the transcribed account","finding":"Enigma encrypted messages; the Bombe helped break Enigma ciphers. The first British Bombe operated in 1940. These events should be separated from Turing’s 1950 publication and from the broader development of digital computers.","sources":[1,2,3],"locator":"TNMOC opening historical description; Turing paper title and opening section.","action":"Use the corrected historical account in the companion article. Confirm audio before assigning the wording definitively to the speaker."},
  {"id":"C02","time":"01:28–01:54","seconds":88,"title":"What Turing demonstrated","claim":"The transcript describes a test establishing that machines could think like humans.","status":"Correction needed in the transcribed account","finding":"Turing proposed an imitation game and discussed objections and future possibilities. His paper does not report the claimed experimental proof of human-like thought.","sources":[3,4],"locator":"Turing sections 1 and 6; section 6 explicitly distinguishes beliefs and conjectures from proved facts. Dartmouth proposal opening paragraph.","action":"Describe the imitation game as a proposed approach to evaluating machine behaviour."},
  {"id":"C03","time":"02:19–03:23","seconds":139,"title":"AI and machine learning","claim":"AI is described primarily as software that learns through experience and data, in contrast to rule-based software.","status":"Clarification needed","finding":"Machine learning is an AI approach, not a complete definition of the field. The NIST glossary includes techniques that approximate cognitive tasks and identifies machine learning as one included technique. Avoid presenting all AI as continuously learning software.","sources":[5],"locator":"NIST CSRC glossary definitions and their linked source contexts.","action":"Define the broad field first, then explain learning-based systems as one approach."},
  {"id":"C04","time":"03:33–04:34","seconds":213,"title":"Automatic learning and future recall","claim":"More inputs are said to improve AI, with information supplied today remembered tomorrow.","status":"Unsupported as a general rule","finding":"Prompt-based task performance can occur without changing model parameters, as demonstrated in the GPT-3 paper. Temporary context, stored memory and later model training are distinct mechanisms. Their availability must be established for the named product.","sources":[6],"locator":"Brown et al., abstract and section 2 description of in-context learning without gradient updates.","action":"Replace the universal statement with a product-specific explanation; do not promise permanent recall."},
  {"id":"C05","time":"04:55–05:07","seconds":295,"title":"IBM and chess","claim":"An IBM AI system defeated chess masters.","status":"Supported at this scope","finding":"IBM records Deep Blue’s 1997 match victory over reigning world champion Garry Kasparov. This supports the historical example. It does not verify the adjacent statement about unspecified GPT systems winning chess.","sources":[7],"locator":"IBM opening account and 1997 rematch description.","action":"Name Deep Blue, Kasparov and the year. Keep claims about other systems separate."},
  {"id":"C06","time":"05:12–06:58","seconds":312,"title":"Generative AI and general intelligence","claim":"The transcript presents narrow, generative and general AI as a progression and links general AI to complete autonomy.","status":"Clarification needed","finding":"Content generation, breadth of capability and autonomy describe different properties. Morris et al. distinguish performance, generality and autonomy. The transcript also says ‘general AI’ at 06:44 in a context where the intended term is uncertain; it has not been silently changed.","sources":[8],"locator":"Morris et al., abstract and framework separating capability levels from autonomy.","action":"Explain these dimensions separately. Resolve the 06:44 wording against audio before correcting the transcript."},
  {"id":"C07","time":"05:39–05:50","seconds":339,"title":"Every language versus evaluated coverage","claim":"The transcript suggests a tool can work in whichever language the user speaks.","status":"Insufficient evidence for universal coverage","finding":"No named model, version or language benchmark is supplied for that universal claim. NLLB evaluates a specified multilingual scope; it does not establish all-language competence or the performance of the products named in this episode. No Namibian-language product test was conducted here.","sources":[9],"locator":"NLLB abstract and evaluation scope; source used as an example of bounded evaluation, not as proof about the episode’s products.","action":"State the language and task actually tested. Leave ‘Oshombo’ unresolved in the raw transcript until audio review."},
  {"id":"C08","time":"09:14–09:26","seconds":554,"title":"Offensive comments on Instagram","claim":"An insulting comment is described as not going through because AI filters it.","status":"Overstated as a guarantee","finding":"Meta describes an intervention that warns users and gives them an opportunity to reconsider. The existence of automated moderation does not support a guarantee that every insulting comment is blocked.","sources":[10],"locator":"Meta’s December 2019 announcement, discussion of comment and caption warnings.","action":"Describe detection and moderation as fallible interventions whose behaviour depends on the feature and context."},
  {"id":"C09","time":"09:28–09:48","seconds":568,"title":"Weather forecasting as an AI example","claim":"Weather forecasts are included in a list described collectively as artificial intelligence.","status":"Supported only with qualification","finding":"AI is used in operational forecasting. ECMWF’s 2025 announcement also documents a traditional physics-based system operating alongside its AI system. This prevents treating all weather forecasting as AI.","sources":[11],"locator":"ECMWF announcement dated 25 February 2025, opening paragraph.","action":"Say AI supports some forecasting systems. Do not generalise the architecture of every forecast service."},
  {"id":"C10","time":"11:24–11:47","seconds":684,"title":"Use of user data for model training","claim":"Most AI services are said to use user information to train their models.","status":"Insufficient evidence for the prevalence claim","finding":"The episode provides no defined service sample or policy comparison to support ‘most’. Anthropic’s dated policy announcement distinguishes consumer choices from commercial services. It illustrates variation but is not an industry census.","sources":[12],"locator":"Anthropic announcement dated 28 August 2025, opening explanation and FAQ ‘What’s changing?’.","action":"Check the exact service, account terms and settings. Keep data retention, training and disclosure to others conceptually separate."}
]