{"resourceId":"kaohsiung-ai-lighthouse-accuracy-cost-dispute-knews-20260601","versions":[{"version":"external-2905dc78745c0d26fbce8f3f91603306fdddb12ab64922d3846b05994aa00d77","resource":{"id":"kaohsiung-ai-lighthouse-accuracy-cost-dispute-knews-20260601","title":"Kaohsiung's AI Lighthouse: councilor cites 2025 benchmark showing weaker recognition at night, in rain and on blurry images and estimates NT$720 million a year in compute; city calls the estimate far too high","organization":"Kaohsiung City Council; Kaohsiung Research, Development and Evaluation Commission","sector":"Local government oversight","geography":"Kaohsiung, Taiwan","publishedAt":"2026-06-01 (background; newly relevant to the Oct 29 results)","publicationDate":"2026-06-01","eventDate":null,"sourceName":"Knews (知新聞)","sourceLabel":"高雄砸2.5億推智慧燈塔計畫！遭疑AI恐誤判、算力恐年花7.2億 市府回應了","sourceUrl":"https://www.knews.com.tw/news/8E939558D4D1B84565CF341E6D7995D3","evidenceClass":"independent-reporting","outcomeClass":"cautionary","topics":["infrastructure","governance-procurement","operating-model"],"finding":"Citing the city's 2025 VLM benchmark report, a councilor said recognition degraded at night, in rain and on blurry images, similar accident types could be confused, and school drop-off zones were misread as illegal parking. She estimated that applying VLM recognition to about 50,000 cameras could cost NT$720 million a year in compute, bandwidth and storage, using Google Vertex AI Vision pricing. The city said the benchmark reflected July 2025 early testing, that models met thresholds by end of 2025 with 6.45 million hours of training video, and that dynamic compute allocation makes the estimate far too high, while the actual cost depends on usage.","sledRelevance":"Gives the accuracy and cost questions behind the Oct 29 headline results. U.S. cities considering VLM camera analytics on NVIDIA infrastructure will face the same issues: degraded conditions, confusable events and recurring compute cost.","evidence":"Newspaper report of council questioning and the city's written response; neither the benchmark report nor current accuracy data is published in the article.","architectureImplications":"Dynamic allocation of compute by season and event type is the city's answer to cost; it requires scheduling and priority policies across scenarios.","governanceImplications":"Accuracy thresholds and the current benchmark results should be published to let oversight bodies check claims.","securityPrivacyImplications":"Scaling to 50,000 cameras expands surveillance reach; retention and access controls become central.","caveats":"Published June 1, 2026, filed as background; cost estimate is the councilor's and disputed; current accuracy figures not disclosed.","streamIds":["nvidia"],"roles":{"sales":"Bring this to any camera-analytics discussion as the honest counterweight to headline results. The city's own early benchmark showed weaker recognition at night, in rain and on poor images, and a councilor estimated large recurring compute costs, which the city disputes. Offer U.S. customers a bounded evaluation that tests accuracy under local conditions and models recurring cost before scale-up. Do not repeat either the councilor's cost estimate or the city's rebuttal as settled.","engineering":"Engineering should build condition-specific test sets from local footage, including night, rain, glare and low-resolution cameras, and report precision and false alarms separately for each. Test confusable classes, such as types of collisions and permitted versus illegal stopping, with rules for school zones and events. Model compute cost per camera and per scenario at realistic detection frequencies, and design priority scheduling so flood detection can take capacity during storms.","delivery":"Delivery should make condition-specific accuracy a gate for each expansion stage and publish the thresholds. Track alert review time and false-alarm volume as operating costs. Maintain a cost dashboard showing compute, storage and bandwidth per scenario so budget questions can be answered with data. Assign a central operations owner to set seasonal compute priorities, and review the benchmark with oversight bodies before widening coverage."},"roleDetails":{"sales":{"customerProblem":"Cities need realistic accuracy and recurring-cost evidence before scaling camera analytics.","stakeholders":["CIO","Budget office","Transportation director","City council"],"discoveryQuestions":["How does accuracy change at night and in rain on your cameras?","What will recurring compute cost per camera be?"],"valueHypothesis":"Condition-specific testing avoids overpromising and surprise operating costs.","potentialEngagement":"Local accuracy evaluation and cost modelling.","unsupportedClaims":["The NT$720 million estimate","That models now meet thresholds in all conditions"]},"engineering":{"fit":"Relevant to VLM analytics on large municipal camera networks.","architectureIntegration":"Condition-specific evaluation and priority-based compute scheduling.","prerequisites":["Local labelled footage by condition","Cost telemetry"],"constraints":["Degraded accuracy in poor conditions","Confusable event classes"],"security":"Limit retention and log access to footage and alerts.","proofOfValue":["Proposed: per-condition precision report","Proposed: cost per camera per scenario"]},"delivery":{"work":"Gate expansion on condition-specific accuracy and cost data.","dependencies":["Camera owners","Operations center","Budget office"],"ownership":"Central operations owner.","skillsAdoption":"Train reviewers on known failure modes.","governanceCheckpoints":["Accuracy gate per expansion stage","Quarterly cost review"],"acceptanceCriteria":["Proposed: thresholds published and met"],"risks":["Hidden recurring costs","False alarms overwhelming staff"]}},"retrievedAt":"2026-10-31T02:56:40Z","enrichedAt":"2026-10-31T03:02:39Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Misreads such as school drop-off zones create work for staff reviewing alerts.","procurementImplications":"Require vendors to state accuracy by condition (day, night, rain) and to model recurring cost per camera and per scenario.","operatingModelImplications":"Seasonal reallocation of compute between flooding and traffic implies a central operations team setting priorities.","sourceVerification":{"openedUrl":"https://www.knews.com.tw/news/8E939558D4D1B84565CF341E6D7995D3","referenceExcerpt":"AI在夜間、雨天及模糊影像下辨識能力下降","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}