{
    "componentChunkName": "component---gatsby-theme-academic-src-templates-post-post-jsx",
    "path": "/research/2026/latent-feedback",
    "result": {"data":{"mdx":{"timeToRead":1,"tableOfContents":{},"frontmatter":{"cover":{"publicURL":"/static/57728411333330eb4c85b827913efb9e/preview.png","childImageSharp":{"fluid":{"tracedSVG":"data:image/svg+xml,%3csvg%20xmlns='http://www.w3.org/2000/svg'%20width='400'%20height='350'%20viewBox='0%200%20400%20350'%20preserveAspectRatio='none'%3e%3cpath%20d='M49%2044v3c2%200%201%206-1%207l-1%202h2l-1%204v3l2-4c0-4%201-5%204-2%204%204%201%2010-5%209h-2l2%201v3l-1%202H32l-15%201c-1%201%2010%201%2033%201h34v49H17l-1-25-1-24v25l1%2025h33l36-1c2-2%202-47-1-49-1-2-4-2-16-2-14%200-16%200-16-2v-3l4-2%201-3h70v78h32l33%201c0%201%202%201%203-1%203-1%203-1%201-3h-4l-32%201h-31l-1-39V61l-35-1H57l-1-2c0-2-2-3-3-4-2%200-2-1-2-5%201-6%200-8-2-5m155-1v6c0%204%200%205-2%206-5%202-5%2010%200%2012l2%202-1%202-15%201c-13%200-16%200-17%202-2%201-3%2047-1%2049l1-24V74h34l34%201v48h-34l-34%201h69V99l-1-26-16-1c-13%200-15%200-17-2v-3l2-1h-2c-5%200-8-4-6-8%202-3%203-3%203%202v3l1-3c0-4%202-6%204-3%202%201%202%202%202%205v2c2-2%205-2%2036-2h35v78h32a235%20235%200%200137%200c3-1%203-1%201-3h-5l-31%201h-32V60h-71l-2-3-4-3-1-3%201-4v-3c-1-2-2-2-2-1m154%201v3c2%200%201%207%200%207-4%200-7%207-5%2010%202%201%202%201%201-2l1-4c2-3%203-2%202%202%200%203%200%203%201%202l1-5c0-3%201-3%204%200%204%204%201%2010-5%209h-2l2%201v3l-1%202h-16l-16%201-1%2026%201%2025h69l-34-1h-34V75l33-1h34v25l1%2024c2-2%201-48-1-49-1-2-4-2-17-2l-15-1-1-2%202-2c5-2%205-10%200-12-2-1-2-2-2-6%201-6%200-8-2-5m-154%2082v3l-2%205c-4%204-4%205-1%209%202%201%203%203%202%204%200%202%201%203%203%202v-5c4%200%205-7%201-10-2-2-2-2-1-4s0-5-1-5l-1%201m154%201v3c1%202%201%202-1%204-4%203-3%2010%201%2010v5c2%201%203%200%203-2-1-1%200-3%202-4%203-4%203-5-1-9l-2-4c1-4-1-7-2-3M37%20213l-1%2015v1l1%2015-1%2015v1c2%200%202%2029%200%2029v1l1%2013v12h16l16%201h1l26-1%2027%201h1l26-1%2026%201h1l17-1-70-1H38v-12l1-11%208-1h9v13h5l6%201%203%202%202-2%206-1h5v-6l1-6%2062-1%2053-1h-9v-70h-5c-5%200-6%200-6-2l-1-2c-1-1-4%200-4%202s-1%202-6%202l-5%201-1%2034v35h-25v-23h-6c-5%200-5-1-5-3%200-3-1-4-2-4-2%200-2%200-1%201l1%202c0%202-2%201-2-1l-1-1v3c0%203%200%203-6%203h-6v23H38v-46l-1-30m188%200l-1%2015v1c1%200%202%2028%200%2030v1l1%2015-1%2014v1l1%2013v12h12l12%201h1l19-1%2020%201h2l20-1%2019%201h1l20-1%2020%201h1l13-1-74-1h-85v-24h85l78-1h-7v-29l-11-1h-10v30h-20v-24h-21v24h-20v-20h-21v20h-19v-14h-21v14h-13v-46l-1-30'%20fill='%23d3d3d3'%20fill-rule='evenodd'/%3e%3c/svg%3e","aspectRatio":1.1415525114155252,"src":"/static/57728411333330eb4c85b827913efb9e/31987/preview.png","srcSet":"/static/57728411333330eb4c85b827913efb9e/e1953/preview.png 250w,\n/static/57728411333330eb4c85b827913efb9e/46604/preview.png 500w,\n/static/57728411333330eb4c85b827913efb9e/31987/preview.png 1000w,\n/static/57728411333330eb4c85b827913efb9e/0dadc/preview.png 1500w,\n/static/57728411333330eb4c85b827913efb9e/bbee5/preview.png 1800w","sizes":"(max-width: 1000px) 100vw, 1000px"}}},"title":"Latent Feedback Is a Compute and Data Amplifier: Scaling Law for Latent Feedback Transformers on Repeated Data","date":"2026-11-13","tags":["Latent Feedback","Scaling Laws","Data Efficiency"],"path":"research/2026/latent-feedback","excerpt":"Feed a Transformer's own top-layer states back in as input, and every FLOP and every token count for more: the same loss with up to 46% less compute or 20% less unique data.","links":[{"name":"alphaXiv","url":"https://www.alphaxiv.org/abs/2611.latent-feedback-transformers-scaling-law"},{"name":"Github","url":"https://github.com/ifm-ai/LoRM/blob/main/works/latent_feedback/README.md"},{"name":"Tweet","url":"https://x.com/huskydogewoof/status/2121282554053414945"}],"commit":0,"type":"research"},"fileAbsolutePath":"/home/runner/work/huskydoge.github.io/huskydoge.github.io/example/content/research/2026/latent-feedback/index.md","fields":{"slug":{"html":"var _excluded = [\"components\"];\n\nfunction _extends() { _extends = Object.assign || function (target) { for (var i = 1; i < arguments.length; i++) { var source = arguments[i]; for (var key in source) { if (Object.prototype.hasOwnProperty.call(source, key)) { target[key] = source[key]; } } } return target; }; return _extends.apply(this, arguments); }\n\nfunction _objectWithoutProperties(source, excluded) { if (source == null) return {}; var target = _objectWithoutPropertiesLoose(source, excluded); var key, i; if (Object.getOwnPropertySymbols) { var sourceSymbolKeys = Object.getOwnPropertySymbols(source); for (i = 0; i < sourceSymbolKeys.length; i++) { key = sourceSymbolKeys[i]; if (excluded.indexOf(key) >= 0) continue; if (!Object.prototype.propertyIsEnumerable.call(source, key)) continue; target[key] = source[key]; } } return target; }\n\nfunction _objectWithoutPropertiesLoose(source, excluded) { if (source == null) return {}; var target = {}; var sourceKeys = Object.keys(source); var key, i; for (i = 0; i < sourceKeys.length; i++) { key = sourceKeys[i]; if (excluded.indexOf(key) >= 0) continue; target[key] = source[key]; } return target; }\n\n/* @jsxRuntime classic */\n\n/* @jsx mdx */\nvar _frontmatter = {\n  \"title\": \"Latent Feedback Is a Compute and Data Amplifier: Scaling Law for Latent Feedback Transformers on Repeated Data\",\n  \"date\": \"2026-11-13\",\n  \"tags\": [\"Latent Feedback\", \"Scaling Laws\", \"Data Efficiency\"],\n  \"highlight\": true,\n  \"venue\": \"Institute of Foundation Models\",\n  \"authors\": [{\n    \"name\": \"**Benhao Huang**\"\n  }, {\n    \"name\": \"Chaitanya Dwivedi\"\n  }, {\n    \"name\": \"Xuezhe Ma\"\n  }, {\n    \"name\": \"Alexander Moreno\"\n  }, {\n    \"name\": \"Hector Liu\"\n  }],\n  \"path\": \"research/2026/latent-feedback\",\n  \"excerpt\": \"Feed a Transformer's own top-layer states back in as input, and every FLOP and every token count for more: the same loss with up to 46% less compute or 20% less unique data.\",\n  \"selected\": true,\n  \"cover\": \"./preview.png\",\n  \"links\": [{\n    \"name\": \"alphaXiv\",\n    \"url\": \"https://www.alphaxiv.org/abs/2611.latent-feedback-transformers-scaling-law\"\n  }, {\n    \"name\": \"Github\",\n    \"url\": \"https://github.com/ifm-ai/LoRM/blob/main/works/latent_feedback/README.md\"\n  }, {\n    \"name\": \"Tweet\",\n    \"url\": \"https://x.com/huskydogewoof/status/2121282554053414945\"\n  }],\n  \"priority\": -5\n};\nvar layoutProps = {\n  _frontmatter: _frontmatter\n};\nvar MDXLayout = \"wrapper\";\nreturn function MDXContent(_ref) {\n  var components = _ref.components,\n      props = _objectWithoutProperties(_ref, _excluded);\n\n  return mdx(MDXLayout, _extends({}, layoutProps, props, {\n    components: components,\n    mdxType: \"MDXLayout\"\n  }), mdx(\"blockquote\", {\n    className: \"twitter-tweet\"\n  }, mdx(\"a\", {\n    href: \"https://x.com/huskydogewoof/status/2121282554053414945\"\n  }, \"View this thread on X\")));\n}\n;\nMDXContent.isMDXComponent = true;","htmlEncrypted":"","nonce":""}}}},"pageContext":{"fileAbsolutePath":"/home/runner/work/huskydoge.github.io/huskydoge.github.io/example/content/research/2026/latent-feedback/index.md","postPath":"research/2026/latent-feedback","translations":[{"hreflang":"en","path":"/research/2026/latent-feedback"}]}},
    "staticQueryHashes": ["4097791827","956690036"]}