[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-a-fix-for-miscalibrated-ai-inference-at-test-time":10,"sections":34},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":29,"feedback":33,"feedback_at":22,"cost_usd":33,"total_tokens":33},3343,"a-fix-for-miscalibrated-ai-inference-at-test-time","A Fix for Miscalibrated AI Inference at Test Time","Researchers found that common test-time guidance methods skew results away from true Bayesian posteriors, and built estimators that correct the drift.","Diffusion models get steered wrong at inference time, and a new paper explains why — then fixes it.\n\nA research team publishing on arXiv identified a structural flaw in how test-time guidance works for pretrained diffusion models. These models are commonly nudged at inference toward outcomes a reward function favors, but the paper shows that approach does not recover the correct Bayesian posterior distribution. The authors traced the problem to specific mathematical approximations baked into standard guidance methods, then built alternative estimators designed to sample from the true posterior instead of just chasing high reward scores. On a benchmark set of Bayesian inference tasks, their approach outperformed prior methods and set a new state-of-the-art peak signal-to-noise ratio on black hole image reconstruction.\n\nThe gap between \"maximize reward\" and \"sample correctly from the posterior\" sounds academic but has real consequences. A model that chases reward scores rather than the true probability distribution will produce confident-looking outputs that are statistically miscalibrated — a problem that compounds in scientific applications where accuracy matters more than aesthetics. Black hole imaging is a high-stakes example: the wrong posterior can make a blurry reconstruction look sharper than the data actually supports.\n\nDiffusion models have become the backbone of image generation and scientific reconstruction alike, so calibration bugs at inference time are not a niche concern. If the fix holds up under broader scrutiny, it could quietly reshape how practitioners tune these models — without a product launch or a funding round in sight.","[\"ai\",\"machine-learning\",\"diffusion-models\",\"bayesian-inference\"]","2026-07-02T04:00:00.000Z","2026-07-02T07:45:10.650Z","2026-07-02T07:45:13.559Z","published",null,[],"ai",[24,26,27,28],"machine-learning","diffusion-models","bayesian-inference",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2602.22428",0,{"sections":35},[36,40,45,50,55,60,65,70,75,80,85,89,94,99],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",2590,"2026-07-16T04:00:00.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":44},"Security","security",294,"2026-07-15T19:59:48.000Z",{"name":46,"slug":47,"count":48,"latest_published_at":49},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Policy","policy",158,"2026-07-16T00:02:48.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",122,"2026-07-14T19:46:26.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":64},"Consumer Tech","consumer-tech",93,"2026-07-13T13:20:48.000Z",{"name":66,"slug":67,"count":68,"latest_published_at":69},"Software","software",70,"2026-07-13T19:52:25.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":74},"Science","science",66,"2026-07-10T10:29:37.000Z",{"name":76,"slug":77,"count":78,"latest_published_at":79},"Dev Tools","dev-tools",59,"2026-07-07T04:00:00.000Z",{"name":81,"slug":82,"count":83,"latest_published_at":84},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":86,"slug":87,"count":83,"latest_published_at":88},"Startups","startups","2026-06-29T20:55:50.000Z",{"name":90,"slug":91,"count":92,"latest_published_at":93},"General","general",29,"2026-07-10T22:28:58.000Z",{"name":95,"slug":96,"count":97,"latest_published_at":98},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":100,"slug":101,"count":102,"latest_published_at":103},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]