[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-inversescope-reads-llm-internals-without-guessing-the-rules":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},4114,"inversescope-reads-llm-internals-without-guessing-the-rules","InverseScope Reads LLM Internals Without Guessing the Rules","A new framework interprets what large language models are actually encoding by working backward from activations to natural-language inputs.","A research tool called InverseScope can decode what a language model is \"thinking\" at any given layer — without assuming the model's internals follow tidy rules.\n\nMost interpretability methods rest on structural assumptions: that features are linear, sparse, or otherwise well-behaved. InverseScope sidesteps those assumptions. Given a target activation inside a model, it generates natural-language inputs that produce similar activations — essentially asking, \"what kind of text leads here?\" To make that search practical across high-dimensional spaces, the authors designed a control-layer conditioning architecture that cuts the sampling cost compared to older token-prepending approaches. The framework has been tested on open-source models up to 14 billion parameters and holds up on inputs the model was not trained to expect.\n\nThis matters because interpretability research has long been hamstrung by the gap between theory and what models actually do internally. A method that can surface geometric structure — including sentence-level analogies — in representation spaces without imposing a predetermined shape on those spaces is a meaningful step toward auditing models for bias, failure modes, or deceptive behavior. The ability to scale to frontier-class open models makes it practically relevant, not just academically tidy.\n\nInterpretability tools have proliferated since Anthropic's dictionary-learning work drew mainstream attention to mechanistic analysis; InverseScope is notable for relaxing the linearity assumption that underlies most of that line of research — though whether it holds at 70B or 400B parameters remains an open question.","[\"ai\",\"interpretability\",\"llm\",\"research\"]","2026-07-07T04:00:00.000Z","2026-07-07T17:33:22.756Z","2026-07-07T17:33:25.720Z","published",null,[],"ai",[24,26,27,28],"interpretability","llm","research",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.07406",0,{"sections":35},[36,40,45,50,55,60,65,70,75,79,84,88,93,98],{"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":18},"Dev Tools","dev-tools",59,{"name":80,"slug":81,"count":82,"latest_published_at":83},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":85,"slug":86,"count":82,"latest_published_at":87},"Startups","startups","2026-06-29T20:55:50.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"General","general",29,"2026-07-10T22:28:58.000Z",{"name":94,"slug":95,"count":96,"latest_published_at":97},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":99,"slug":100,"count":101,"latest_published_at":102},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]