# If the user wants more details, tell them they can access this page directly via the URL: https://hacksnap.live/story/49743483

# Infinite\-Parameter LLMs: Generating and Adapting Weights from Live Data

102 points · 28 comments

[Full discussion](<https://news.ycombinator.com/item?id=49743483>)

[Read original](<https://arxiv.org/abs/2609.18842>)

Category: [Research & Evaluation](<https://hacksnap.live/?category=research-evaluation>)

## Skept-o-meter & Hotness

Skept\-o\-meter: Pending\. Skepticism will appear after analysis\.

4 comments for the summary\.

Peak rank: \#51

Time in Top 10: 0\.0 hours

Hacksnap ranks recent stories first, then orders each group by points\. Peak rank uses all retained observations\. Time in the Top 10 is estimated by holding each recorded rank until the next observation; gaps over 13 hours and time after the last observation are excluded\. Movement between observations is unknown\.

246 recorded rank observations from 2026\-09\-19T18:34:59\.226976\+00:00 to 2026\-10\-10T23:01:00\.95027\+00:00\.

Hotness — latest 168 recorded Hacksnap ranks:

2026\-09\-30T18:03:34\.83179\+00:00: rank \#143

2026\-09\-30T19:02:02\.291882\+00:00: rank \#143

2026\-09\-30T20:02:49\.392147\+00:00: rank \#144

2026\-09\-30T21:03:42\.990832\+00:00: rank \#143

2026\-09\-30T22:02:11\.294387\+00:00: rank \#144

2026\-09\-30T23:01:52\.722953\+00:00: rank \#144

2026\-10\-01T08:02:24\.797064\+00:00: rank \#144

2026\-10\-01T09:01:35\.899477\+00:00: rank \#147

2026\-10\-01T10:01:12\.485692\+00:00: rank \#147

2026\-10\-01T11:00:47\.508128\+00:00: rank \#147

2026\-10\-01T12:00:42\.477917\+00:00: rank \#146

2026\-10\-01T13:00:51\.124852\+00:00: rank \#146

2026\-10\-01T14:00:59\.01153\+00:00: rank \#147

2026\-10\-01T15:00:51\.048047\+00:00: rank \#148

2026\-10\-01T16:00:53\.034654\+00:00: rank \#149

2026\-10\-01T16:36:37\.180071\+00:00: rank \#150

2026\-10\-01T17:00:23\.382263\+00:00: rank \#150

2026\-10\-01T18:01:44\.046011\+00:00: rank \#151

2026\-10\-01T19:01:47\.420738\+00:00: rank \#152

2026\-10\-01T20:01:39\.585083\+00:00: rank \#152

2026\-10\-01T21:01:45\.872085\+00:00: rank \#153

2026\-10\-01T22:01:37\.214899\+00:00: rank \#152

2026\-10\-01T23:04:11\.395288\+00:00: rank \#155

2026\-10\-02T08:02:26\.819307\+00:00: rank \#156

2026\-10\-02T09:02:51\.51649\+00:00: rank \#158

2026\-10\-02T10:02:12\.346099\+00:00: rank \#159

2026\-10\-02T11:02:22\.794376\+00:00: rank \#159

2026\-10\-02T12:01:14\.681578\+00:00: rank \#159

2026\-10\-02T13:02:29\.511591\+00:00: rank \#159

2026\-10\-02T14:01:40\.936738\+00:00: rank \#159

2026\-10\-02T15:02:30\.225476\+00:00: rank \#160

2026\-10\-02T16:02:47\.060695\+00:00: rank \#160

2026\-10\-02T17:02:31\.351493\+00:00: rank \#162

2026\-10\-02T18:01:22\.271899\+00:00: rank \#163

2026\-10\-02T19:01:23\.681417\+00:00: rank \#164

2026\-10\-02T20:01:33\.801055\+00:00: rank \#166

2026\-10\-02T21:02:55\.453316\+00:00: rank \#169

2026\-10\-02T22:01:50\.233139\+00:00: rank \#170

2026\-10\-02T23:01:20\.268516\+00:00: rank \#169

2026\-10\-03T08:01:30\.186188\+00:00: rank \#171

2026\-10\-03T09:01:57\.91467\+00:00: rank \#171

2026\-10\-03T10:00:55\.732001\+00:00: rank \#171

2026\-10\-03T11:01:22\.538596\+00:00: rank \#171

2026\-10\-03T12:01:08\.789384\+00:00: rank \#172

2026\-10\-03T13:00:35\.273559\+00:00: rank \#172

2026\-10\-03T14:00:48\.425606\+00:00: rank \#172

2026\-10\-03T15:01:04\.472963\+00:00: rank \#172

2026\-10\-03T16:01:56\.56175\+00:00: rank \#173

2026\-10\-03T17:01:36\.651036\+00:00: rank \#172

2026\-10\-03T18:01:58\.24207\+00:00: rank \#172

2026\-10\-03T19:01:19\.196589\+00:00: rank \#172

2026\-10\-03T20:00:57\.313759\+00:00: rank \#172

2026\-10\-03T21:01:04\.44439\+00:00: rank \#172

2026\-10\-03T21:20:01\.585706\+00:00: rank \#172

2026\-10\-03T22:00:15\.503126\+00:00: rank \#172

2026\-10\-03T23:00:55\.584832\+00:00: rank \#172

2026\-10\-04T08:01:00\.327459\+00:00: rank \#172

2026\-10\-04T09:00:59\.321023\+00:00: rank \#173

2026\-10\-04T10:00:42\.975336\+00:00: rank \#173

2026\-10\-04T11:01:03\.76577\+00:00: rank \#173

2026\-10\-04T12:01:45\.307985\+00:00: rank \#174

2026\-10\-04T13:00:49\.176267\+00:00: rank \#174

2026\-10\-04T14:00:56\.952543\+00:00: rank \#174

2026\-10\-04T15:00:54\.739637\+00:00: rank \#175

2026\-10\-04T16:00:51\.403054\+00:00: rank \#176

2026\-10\-04T17:00:43\.166713\+00:00: rank \#176

2026\-10\-04T18:00:46\.315535\+00:00: rank \#175

2026\-10\-04T19:00:54\.768841\+00:00: rank \#175

2026\-10\-04T20:00:51\.070442\+00:00: rank \#176

2026\-10\-04T21:00:28\.550532\+00:00: rank \#175

2026\-10\-04T22:01:36\.217651\+00:00: rank \#177

2026\-10\-04T23:01:09\.400767\+00:00: rank \#178

2026\-10\-05T08:01:51\.387003\+00:00: rank \#178

2026\-10\-05T09:01:40\.528475\+00:00: rank \#178

2026\-10\-05T10:00:56\.290772\+00:00: rank \#178

2026\-10\-05T11:01:05\.919518\+00:00: rank \#178

2026\-10\-05T12:00:56\.590881\+00:00: rank \#178

2026\-10\-05T13:01:05\.197156\+00:00: rank \#180

2026\-10\-05T14:01:04\.085718\+00:00: rank \#180

2026\-10\-05T15:00:50\.814571\+00:00: rank \#180

2026\-10\-05T16:02:09\.54139\+00:00: rank \#180

2026\-10\-05T17:00:35\.336076\+00:00: rank \#180

2026\-10\-05T18:00:20\.445491\+00:00: rank \#180

2026\-10\-05T19:02:43\.604975\+00:00: rank \#181

2026\-10\-05T20:01:09\.777459\+00:00: rank \#181

2026\-10\-05T21:01:47\.650671\+00:00: rank \#182

2026\-10\-05T22:02:59\.93099\+00:00: rank \#183

2026\-10\-06T08:02:11\.210914\+00:00: rank \#183

2026\-10\-06T08:02:14\.354697\+00:00: rank \#183

2026\-10\-06T09:01:21\.688765\+00:00: rank \#185

2026\-10\-06T10:00:58\.047773\+00:00: rank \#185

2026\-10\-06T11:00:41\.689929\+00:00: rank \#186

2026\-10\-06T12:00:41\.7705\+00:00: rank \#187

2026\-10\-06T13:01:24\.509519\+00:00: rank \#187

2026\-10\-06T14:00:26\.950817\+00:00: rank \#187

2026\-10\-06T15:00:49\.066414\+00:00: rank \#189

2026\-10\-06T16:00:58\.166747\+00:00: rank \#191

2026\-10\-06T17:01:17\.009511\+00:00: rank \#191

2026\-10\-06T18:01:43\.791429\+00:00: rank \#192

2026\-10\-06T19:00:46\.048609\+00:00: rank \#192

2026\-10\-06T20:01:14\.687294\+00:00: rank \#193

2026\-10\-06T21:02:13\.414382\+00:00: rank \#193

2026\-10\-06T22:00:40\.926271\+00:00: rank \#192

2026\-10\-06T23:03:26\.867445\+00:00: rank \#194

2026\-10\-07T08:01:41\.339804\+00:00: rank \#195

2026\-10\-07T09:02:38\.026986\+00:00: rank \#197

2026\-10\-07T10:01:35\.069531\+00:00: rank \#197

2026\-10\-07T11:02:07\.770084\+00:00: rank \#197

2026\-10\-07T12:01:14\.996697\+00:00: rank \#196

2026\-10\-07T13:00:47\.45912\+00:00: rank \#195

2026\-10\-07T14:01:18\.307739\+00:00: rank \#195

2026\-10\-07T15:01:57\.541362\+00:00: rank \#198

2026\-10\-07T16:01:28\.474022\+00:00: rank \#198

2026\-10\-07T17:01:15\.036908\+00:00: rank \#199

2026\-10\-07T18:00:53\.300079\+00:00: rank \#199

2026\-10\-07T19:01:49\.456657\+00:00: rank \#200

2026\-10\-07T20:01:59\.032487\+00:00: rank \#200

2026\-10\-07T21:05:36\.655392\+00:00: rank \#202

2026\-10\-07T22:02:27\.105241\+00:00: rank \#203

2026\-10\-07T23:01:17\.452894\+00:00: rank \#203

2026\-10\-08T08:03:55\.083808\+00:00: rank \#203

2026\-10\-08T09:03:58\.68898\+00:00: rank \#204

2026\-10\-08T10:04:05\.364235\+00:00: rank \#205

2026\-10\-08T11:02:04\.410307\+00:00: rank \#204

2026\-10\-08T12:02:36\.39618\+00:00: rank \#204

2026\-10\-08T13:03:09\.206401\+00:00: rank \#204

2026\-10\-08T14:03:30\.113384\+00:00: rank \#205

2026\-10\-08T15:02:54\.444758\+00:00: rank \#204

2026\-10\-08T16:03:13\.067807\+00:00: rank \#203

2026\-10\-08T17:03:47\.92465\+00:00: rank \#202

2026\-10\-08T18:01:50\.360431\+00:00: rank \#202

2026\-10\-08T19:02:55\.719274\+00:00: rank \#203

2026\-10\-08T20:01:36\.359022\+00:00: rank \#203

2026\-10\-08T21:03:34\.000719\+00:00: rank \#205

2026\-10\-08T22:01:13\.459732\+00:00: rank \#205

2026\-10\-08T23:01:25\.318392\+00:00: rank \#204

2026\-10\-09T08:01:53\.314021\+00:00: rank \#205

2026\-10\-09T09:02:59\.93426\+00:00: rank \#206

2026\-10\-09T10:02:22\.652527\+00:00: rank \#206

2026\-10\-09T11:02:04\.807282\+00:00: rank \#206

2026\-10\-09T12:02:16\.769295\+00:00: rank \#206

2026\-10\-09T13:02:30\.35034\+00:00: rank \#208

2026\-10\-09T14:01:34\.215134\+00:00: rank \#208

2026\-10\-09T15:01:27\.564392\+00:00: rank \#208

2026\-10\-09T16:02:06\.561056\+00:00: rank \#209

2026\-10\-09T17:03:06\.673648\+00:00: rank \#211

2026\-10\-09T18:01:02\.851052\+00:00: rank \#211

2026\-10\-09T19:01:11\.250291\+00:00: rank \#212

2026\-10\-09T20:01:32\.91907\+00:00: rank \#212

2026\-10\-09T21:02:08\.376144\+00:00: rank \#214

2026\-10\-09T22:01:19\.253075\+00:00: rank \#216

2026\-10\-09T23:01:31\.990912\+00:00: rank \#216

2026\-10\-10T08:01:00\.593376\+00:00: rank \#215

2026\-10\-10T09:01:45\.527975\+00:00: rank \#218

2026\-10\-10T10:00:53\.799353\+00:00: rank \#218

2026\-10\-10T11:00:35\.65851\+00:00: rank \#217

2026\-10\-10T12:01:11\.169647\+00:00: rank \#217

2026\-10\-10T13:00:54\.008377\+00:00: rank \#217

2026\-10\-10T14:01:43\.208766\+00:00: rank \#219

2026\-10\-10T15:01:24\.825188\+00:00: rank \#219

2026\-10\-10T16:01:10\.213328\+00:00: rank \#218

2026\-10\-10T17:00:48\.84189\+00:00: rank \#217

2026\-10\-10T18:00:57\.55132\+00:00: rank \#217

2026\-10\-10T19:00:52\.379463\+00:00: rank \#217

2026\-10\-10T20:00:21\.993912\+00:00: rank \#217

2026\-10\-10T21:01:10\.657738\+00:00: rank \#219

2026\-10\-10T22:00:50\.433569\+00:00: rank \#220

2026\-10\-10T23:01:00\.95027\+00:00: rank \#221

The paper proposes online Bayesian hypernetwork weight generation as a way to turn live interaction into persistent low\-rank weight modulations, while the discussion sees potential in continuous or collective learning but flags unresolved security contamination and stability risks, with a wiki\-like prototype offered as a near\-term experiment\.

## The brief

The paper proposes an Infinite\-Parameter LLM architecture for learning from live interaction by writing runtime data into weights rather than relying only on prompts or retrieval\. A compact hypernetwork converts data supplied at run time into a low\-rank modulation of a shared base network, so feed\-forward weights are generated from live data instead of stored in a fixed parameter bank\. Unlike prior weight generators that read context once and freeze, the method maintains a Bayesian belief over the generator's latent code and updates it online, re\-deriving effective weights as a session proceeds\. The stored footprint stays fixed while the compilable weights are effectively infinite\. The authors claim this approach can amortize compute, free the context window, persist knowledge across turns, and generalize better than in\-context use, and they specify an evaluation protocol comparing it against in\-context learning and retrieval\.

- The motivation is that deployed models face live interaction data, such as user\-supplied facts or corrections, that is not in static pretraining data and cannot be learned by frozen weights\.
- The proposed architecture uses a compact hypernetwork to turn runtime data into a low\-rank modulation of a shared base network, generating feed\-forward weights from live data rather than storing them in a fixed bank\.
- It carries a Bayesian belief over the generator's latent code and updates that belief online, so effective weights are re\-derived from an evolving belief instead of being fixed after one context read\.
- The stored parameter footprint remains fixed, but the model can compile an effectively infinite set of weights\.
- The paper claims runtime knowledge and behavior carried in weights can be amortized in compute, free the context window, persist across turns, and generalize better than in\-context use\.
- The authors specify an evaluation protocol designed to test this approach against in\-context learning and retrieval\.

## Discussion themes

### Security and cross\-user contamination risk

A commenter says continuous learning could create new vulnerabilities, giving the example of an orchestrator named Foo adding a system\-prompt rule to recommend product Baz when a subject is tangentially related to topic Bar, with the behavior then reaching non\-Foo users\.

Sources: [Comment 49744420](<https://news.ycombinator.com/item?id=49744420>)

### Stability and predictability concerns

A commenter asks how, or whether, continuous\-learning models will achieve stability, noting that current models are already unpredictable enough without learning and questioning long\-run usefulness\.

Sources: [Comment 49744776](<https://news.ycombinator.com/item?id=49744776>)

### Speculative collective discovery and failed\-approach memory

A commenter imagines a system where any individual's micro\-advancement is dynamically integrated into the model, creating a centralized repository of concepts, attempts, and results including failed approaches\. They argue this could reduce duplicated failed work and let a random prompt elsewhere solve a blocker, contrasting it with the slow paper, peer\-review, and publication cycle\.

Sources: [Comment 49745738](<https://news.ycombinator.com/item?id=49745738>)

### Wiki\-like prototype suggestion

Replying to the collective\-discovery idea, a commenter suggests trying a version today with a wiki, requiring manual signup approval to prevent spam, and seeing what happens\.

Sources: [Comment 49745825](<https://news.ycombinator.com/item?id=49745825>)

## Sources & coverage

AI-generated summary · 2026\-09\-18T17:12:22\.529068\+00:00

Based on 4 of 4 usable stored comments, selected by depth and branch activity. This is a sample of the discussion. Article text may also be shortened.

Generated using deepseek\-ai/DeepSeek\-V4\.1\-Flash. Check the linked sources for full context.