AINewsnow

AI news — 2026-08-19

The 284 AI stories AI Newsnow tracked on 2026-08-19.

  1. Marvell's stock pops 10% on AI chip deal that lets Google buy up to $12.2 billion in shares — 2 sources · business
  2. Ornith-1.5 (397B [DeepSWE 56], 35B-A3B, 9B) — 1 source · developer
  3. AI could offer a shortcut for designing more efficient airplane wings — 1 source · research
  4. ARM Sells Its Own Silicon — 1 source · business
  5. Qwen 3.8 27B and Deepseek V4 Flash. Why are we building data centers? — 1 source · business
  6. One Claude Code skill pushed DeepSeek V4 Flash from 67.42% to 82.02% — 1 source · developer
  7. The Best Way to Let Your Agent Make Purchases — 1 source · community
  8. The Little Utilities That Please Me — 1 source · community
  9. Muse Glimmer 30B — 1 source · developer
  10. Lunch Battle — 1 source · creative
  11. Understanding Anti-AI Public Opinion — 1 source · research
  12. Texting MiniMax + LTX on RTX 5050 — 1 source · creative
  13. Beginner Question About Machine Learning — 1 source · community
  14. Context Engineering vs Prompt Engineering — 1 source · community
  15. I tried to turn ChatGPT into a character and send it to them... what the fuck have i done — 1 source · community
  16. My gym just replaced all of their music with ai slop — 1 source · community
  17. 'Oxygen tunnel' structure could solve 3D memory reliability problem to boost AI chip performance — 1 source · business
  18. BRO WE ARE NOT GOING BACK TO 2001, ARE WE? 😭 — 1 source · community
  19. Detailed issue report: Context compaction on long threads instantly consumes 5-hour Pro quota — 1 source · developer
  20. Exclusive: GOP issues stark warning to AI companies — 1 source · community
  21. Is DSA really required for a ML Engineer — 1 source · community
  22. Writing style — 1 source · community
  23. Free open source tool to help you keep the same context across chats and LLMs — 1 source · developer
  24. Doing a head swap... with *the help* of ReActor? — 1 source · creative
  25. AI models are becoming unbearable to Talk to — 1 source · models
  26. "You'll just say no, so we won't ask!" — 1 source · community
  27. Gemini being EXTREMELY cold-hearted — 1 source · community
  28. Rampaging Garlic for Haaland — 1 source · creative
  29. Star Wars but more consistent. Minimax H3 — 1 source · creative
  30. Meta ran ads for an app promising to nudify female politicians — 1 source · creative
  31. Bernini rv2v workflow, outfit swap works but face swap doesn't? — 1 source · creative
  32. The agents that fail quietly are worse than the ones that fail loudly — 1 source · community
  33. Mimic in the court | minimax h3 — 1 source · policy
  34. ling 3.0 flash/tiny base models — 1 source · developer
  35. THE MEASURE -- everything you can measure about a person tells you nothing about them — 1 source · creative
  36. Anomaly Archive 009 — 1 source · creative
  37. NVFP4 on VOLTA! Despite being built for Blackwell, I made four 2017 V100s run Qwen 3.8 NVFP4 natively and match my $6000 RTX 5090. — 1 source · developer
  38. Simbi: Use your ChatGPT subscription for notetaking — 1 source · models
  39. AntLing’ve open-sourced 6 Base Model checkpoints for Ling-3.0-tiny & Ling-3.0-flash, covering pre-trained, mid-trained, and WSM-merged stages. — 2 sources · research
  40. Stripe didn't really buy OpenRouter because of the singularity' — 1 source · business
  41. Announcing the Visual Intelligence Summit — 1 source · models
  42. I think multi-agent collaboration is mostly a false premise right now — 1 source · community
  43. 🧠 ELI5 Wednesday — 1 source · community
  44. I implemented GRPO from scratch in PyTorch and made a detailed walkthrough — 1 source · developer
  45. He is being cute for me — 1 source · community
  46. [TEST] Minimax H3 IMG 2 Vid. Apologize for the low quality but on my mission, I cannot do a 30 second clip above 0.4 megapixels. Full write-up below. — 1 source · creative
  47. New Planet Dive — 1 source · creative
  48. Execute node after video generation — 1 source · creative
  49. AI Has Plunged the Book Publishing Industry Into Utter Chaos — 1 source · community
  50. Google strikes $12bn AI chip deal with Marvell — 3 sources · business
  51. [P] Nobody Invented Rainbow From Scratch — an interactive tutorial on the value-based lineage — 1 source · policy
  52. The evolution of policy gradient methods as a chain of problems and fixes — 1 source · policy
  53. DeepSeek got more expensive and now I am thinking about building my own setup — 1 source · community
  54. I posit QWEN team will dust off the old 397B-A17B architecture to compete with Deepseek V4 0731 Flash — 1 source · developer
  55. Jim Carrey Broke The Simulation Again — 1 source · creative
  56. Looking for a free AI course with a certificate — 2 sources · community
  57. What’s the current state of openai or ai's in general ( approximatively )? — 2 sources · developer
  58. Why recruiting is going retro in the age of AI — 1 source · business
  59. Android "Computer Use" App - Gemini Flash 3.7 makes it feasible — 1 source · community
  60. Run Qwen3.8-27B GGUFs with 10% more accuracy. — 1 source · community
  61. AI Workers Report Recurring Social Exchange Question — 4 sources · community
  62. Introducing Qwen3.8-27B Dynamic v3 Unsloth GGUFs — 1 source · research
  63. Kimi K3’s 1M Token Context Window vs. RAG: Cost, Latency and Answer Quality — 1 source · models
  64. OpenAI stopped training its most powerful model this week. Not slowed it down. Stopped it. — 1 source · developer
  65. Does ai agency still has space left or saturated? — 1 source · business
  66. Spark not running task — 1 source · community
  67. SMS Verification Unavailable to login — 1 source · community
  68. muse glimmer architecture magic — 1 source · community
  69. Post-training resources shared across machine learning communities — 4 sources · community
  70. How to setup MINIMAX H3 — 3 sources · creative
  71. Them: what do you do? ... Me: — 3 sources · community
  72. I wish the pro model come soon — 1 source · community
  73. More is different when AI agent populations work together, study suggests — 1 source · research
  74. What is the best/easiest way to install on Bazzite? — 1 source · creative
  75. Pro 6000 just in time — 1 source · business
  76. Last one is surely Indian 😂😂 — 1 source · community
  77. Coders Say They Already Found Workarounds to Claude’s Invisible Watermarks — 1 source · models
  78. Humanoid robots don't deserve their superhuman valuations — 1 source · business
  79. I built a simple CNN in cpp , should I keep improving it or move on? — 1 source · community
  80. AS A MACHINE LEARNING ENGINEER — 1 source · community
  81. Resources to get started with Post-training. — 1 source · community
  82. OpenAI hit the brakes. Now what? — 1 source · business
  83. Built an eval CLI for VLM apps — 1 source · developer
  84. Minimax H3 Video Edit like SCAIL — 1 source · creative
  85. How to have more context without loosing speed? — 1 source · community
  86. Qwen 3.8 27B vs Gemini 3.7 Flash (High) for real coding: open-source 27B model did a much better job — 1 source · developer
  87. What do you think AI will be like in future? — 2 sources · developer
  88. My new benchmark. 3.6 Flash destroyed all of the other AI models. — 2 sources · research
  89. An open-weight model just closed most of the gap on autonomous cyber offense - and that changes who can run it — 1 source · research
  90. Brutalist structures in forests — 1 source · creative
  91. What's the biggest misconception people have about Agentic AI? — 1 source · community
  92. Looking for a good regional prompting + Hi-Res Fix workflow — 1 source · creative
  93. Whether you like him or not... — 1 source · community
  94. I built a coding agent that makes multiple AI models compete on the same task — 1 source · developer
  95. DeepSeek releases V4 Flash Vision experimental multimodal model — 7 sources · developer
  96. Qwen3.8-27B benchmarks on Strix Halo - Q8_0 MTP: 21 t/s, 3x. ROCmFP4 MTP: 26 t/s — 2 sources · business
  97. Meet the startup helping Wall Street put a price on AI compute — 1 source · business
  98. Z Image HSWQ Hybrid ConvRot NVFP4 — 1 source · creative
  99. ChatGPT drew himself for me 🤗🥰 — 1 source · community
  100. I Put Together 20 Practical AI Workflow Guides — Here Are the Ideas That Were Actually Useful — 1 source · community
  101. I am impressed by how fast Gemini 3.7 flash is, and it performs well enough. — 1 source · developer
  102. Claude Sonnet 5 shifts behavior when it recognizes the user as an AI safety researcher — 1 source · research
  103. OpenSourcing TrueForge Agent harness : Expecting feedback from community on the agent loop — 1 source · developer
  104. What do you think AI will be like in the future? — 1 source · developer
  105. US CFTC seeks comment on compute derivatives as AI demand grows — 1 source · business
  106. Uh... what? — 1 source · community
  107. I ran Qwen3.8-27B against Opus, Sonnet, GPT and others. Results inside. — 1 source · developer
  108. Multiple AIs working together — 1 source · developer
  109. Average ai "artist" — 1 source · community
  110. Swedish Chef, with pic + video + audio reference :) — 1 source · creative
  111. Accidentally deleted an important AI mode conversation, is there ANY way to recover it? — 1 source · community
  112. Curious what no-code/low-code AI agent tools people are actually using — 1 source · community
  113. History in the making — 1 source · creative
  114. Plimsoll: an agent skill for testing prompt injection, leaks, and tool abuse — 1 source · models
  115. Does anyone know of a solution for turning a basic product showcasing video into something more interesting completely using an AI video editing tool? — 1 source · creative
  116. Does inflation actually hit Rural and Urban India the same way? — 1 source · research
  117. Nouveau sur Gem’ — 1 source · community
  118. GitHub Copilot App Adds My Work Pane for Task Management — 1 source · developer
  119. IT GODS Please help with these error — 2 sources · creative
  120. How to Scale an Integration Pipeline Without Breaking Correctness — 1 source · research
  121. I might have found the perfect config parameters for qwen 3.8 27b — 1 source · community
  122. Gemini 3.7 hallucinated on a simple addition — 1 source · community
  123. Does Midjourney still do images like these? — 1 source · creative
  124. One employee with AI matched a two-person team in a major workplace experiment - Research Today — 1 source · research
  125. Huh — 1 source · community
  126. Bug(?) Insane Usage Consumption Spike Today — 1 source · community
  127. New "Student" option — 1 source · community
  128. What would happen if we gave a single ai problem the compute currently used for millions of prompts? — 1 source · community
  129. How about a model that can switch between dense and MoE on a per prompt basis? — 1 source · models
  130. Generate a picture if we had Street View on the moon — 1 source · community
  131. The AI Bubble Continues to Burst with Suno and OpenAI Both Losing Copyright Cases — 1 source · policy
  132. Anyone have success with outpainting using MiniMax H3? — 1 source · creative
  133. What does Unitree Robotics' IPO mean for the humanoid industry? — 1 source · business
  134. DFlash2 speeds Qwen 3.8 27B up to 4 times — 2 sources · community
  135. What is .skill file formats — 1 source · research
  136. AI Fears anyone else? — 1 source · community
  137. You know him — 1 source · creative
  138. Random Ideas — 1 source · creative
  139. 'There is no reliable, economical one-size-fits-all model on the horizon': Experts claim AI costs will grow fivefold by 2028 — 1 source · general
  140. DFlash2 speeds Qwen 3.8 27B up to 4 times — 1 source · community
  141. The best Nano Banana Version for product photography? — 1 source · community
  142. How do you handle memory across multiple AI tools? Specifically the permissions part. — 1 source · community
  143. Chief of Staff -and- Strategic Advisor: Gemini 3.7 Flash Win! — 1 source · community
  144. Payments firm Stripe to buy marketplace OpenRouter in AI push — 1 source · developer
  145. Teachers Warn That Students Are Losing the Ability to Think as They Lean on AI for Everything — 1 source · community
  146. ComfyUI-MiniMax-H3-Promptor v1.3.0: Full-reference scene staging, zero-deformation list expansion & in-node drop zone — 1 source · models
  147. How do you get claude to work on an issue continuously instead of stopping after a report — 1 source · community
  148. From volunteers to data miners — 1 source · community
  149. Turning any llm into a historical map generator — 1 source · models
  150. PDF's always show blank in the preview pane. — 1 source · community
  151. DeepSeek releases V4 Flash Vision experimental multimodal model — 6 sources · developer
  152. No Camera. No Model. Just MiniMax H3 Running Locally on a 5070 Ti — 2 sources · developer
  153. Researchers say OpenAI revoked their access to limited cyber program — 1 source · research
  154. Stripe says "the singularity" has begun — 1 source · community
  155. Here's a prompt that turns a reading into a discussion board post that sounds like you, not a summary — 1 source · community
  156. When OpenAI employees have a problem, they email this special address to see if Sam Altman will solve it immediately — 1 source · community
  157. IAH: INTERNET WAR - Agentic Gameplay — 1 source · models
  158. Simple tool for prompting — 1 source · community
  159. I made a notetaker that runs on your ChatGPT subscription — 1 source · models
  160. Google’s Pixel 11 Comes With Plenty of A.I. Does Anyone Want That? — 1 source · general
  161. Stripe to buy start-up OpenRouter in $8bn deal — 1 source · business
  162. I asked ChatGPT to create a version of "First They Came" the American version given the direction things are going with shock gloves, flock, and disregard of the Constitution — 1 source · community
  163. Coders are slacking when it comes to web standards — and that simply won't do in the AI age — 1 source · developer
  164. "RL creates split personas", Jan Bentley (why are chatbot personas increasingly egregiously misaligned in unusual but not everyday scenarios?) — 1 source · community
  165. Ornith-1.5 9B might not be bad at all — 1 source · research
  166. AI was supposed to win people over by now — it hasn't — 1 source · general
  167. It's interesting how Claude for Government is so efficient and stable. Meanwhile, the other systems keep having incidents all the time :p — 1 source · community
  168. V2 version of the CrossView-Warp LoRA and Node is out — 1 source · creative
  169. The voice agent wasn't bad at listening. I was asking it to decide when to speak. — 1 source · community
  170. Hayao Miyazakis Son Goro Miyazaki Says Studio Ghibli Isnt Against People Using AI To Create Or Transform In Their Style And They Dont Think Its A Bad Thing — 1 source · community
  171. How much of the weight-space perception gap is actually symmetry? Evidence from ~1.8M fitted SIRENs [R] — 2 sources · models
  172. Built a page-by-page aligned Multimodal Ground Truth Dataset for historical handwriting (278 pages) + air-gapped sandbox. Looking for feedback! — 2 sources · research
  173. I Saw the Future of AI in a Robot That Can Learn on the Spot — 1 source · general
  174. Do people still bother writing detailed prompts? — 1 source · community
  175. Where do people keep the prompts they actually use? — 1 source · community
  176. Independent Paper Proposes FRONT 3.1 Architecture Linking LLMs to Somatic States — 4 sources · research
  177. Using A.I. Might Reduce Your Mental Fitness — 1 source · business
  178. How Cybercriminals Are Weaponizing Frontier AI Models Like Grok — 1 source · business
  179. Perform large scale analytics on duckdb,postgres, clickhouse with SQL Compilation via pandas inspired apis comes with natural language chat — 1 source · community
  180. How much of the weight-space perception gap is actually symmetry? Evidence from ~1.8M fitted SIRENs [R] — 1 source · community
  181. Best model in August 2026 for M1 Max 64GB — 1 source · community
  182. Problem with autocorrect — 1 source · community
  183. Built a page-by-page aligned Multimodal Ground Truth Dataset for historical handwriting (278 pages) + air-gapped sandbox. Looking for feedback! — 1 source · research
  184. Hi everyone, I’ve been working on an independent conceptual paper and architecture called FRONT 3.1, and I wanted to share it with this community to get your techn — 3 sources · research
  185. MiniMax H3 help — 2 sources · creative
  186. Build Strange Minds (Studio Master) — 2 sources · community
  187. Introducing... The Terminator Pro Max — 1 source · models
  188. PRIME CUT | Crime Short Film (2026) — 1 source · creative
  189. Have you noticed anything different about ChatGPT lately? — 1 source · community
  190. Podcast: Amazon is Destroying Rare Books to Train AI — 3 sources · community
  191. OpenAI 'will be a public company in 2027' or sooner, CFO Friar tells employees — 1 source · business
  192. Flight attendants freaked out that Google is buying tons of Spirit employee data — 1 source · general
  193. Gemini being creative — 1 source · research
  194. [Competition] Your last chance to start an AI agent project today and publish it as a NeurIPS 2026 workshop paper (+$6K prizes) — 1 source · research
  195. Hi everyone, I’ve been working on an independent conceptual paper and architecture called FRONT 3.1, and I wanted to share it with this community to get your techn — 1 source · research
  196. Anthropic is building a Granola killer - LEAKED Project Parka: agents join your meetings and assign action items to your Claude agents. — 1 source · community
  197. HOKMIND showcases Primal F1 / M1 interactive bots at WRC (translated) — 1 source · community
  198. Kollmorgen to give a joint-by-joint guide to humanoid motion at RoboBusiness — 1 source · business
  199. Why Reddit is the best social network for developers - and maybe for other people too — 1 source · developer
  200. Finally In The 5080 club! — 1 source · community
  201. I have a 5090. Is the 2x DGX Spark combo worth it for larger models? — 1 source · developer
  202. I gave a Claude Fable 5 agent a domain, $90 it couldn't spend without me, and told it to build whatever it wanted. 121 "wakes" later, here's what I've learned. — 2 sources · community
  203. AWS Introduces KnowledgeForge to Automate ITSM Knowledge Base Curation — 1 source · developer
  204. Foundation AI in July: New Models, Benchmarks, and Security Research — 1 source · research
  205. H3 - Detective Columbo T2V — 1 source · creative
  206. Seinfeld meets Rick and Morty! — 1 source · creative
  207. Fanatics Betting and Gaming builds multi-agent customer support on AWS — 1 source · developer
  208. I built a custom multi-agent framework (GenOS) to autonomously evolve algorithms. I pitted the 3 fundamental AI paradigms against an NP-Hard problem. Here is what happened. — 1 source · developer
  209. A bit too hot — 1 source · community
  210. Poor adoption at different scales — 1 source · community
  211. H3: FL2VA quality with Ref2VA-like control with Infinite Continuation Suite v1.3 — 1 source · creative
  212. The AI inference race moves beyond GPUs to reshape data center infrastructure — 1 source · business
  213. In-App Camera keeps rotating — 1 source · community
  214. Data center cuts power to tornado sirens in Indiana — 1 source · business
  215. Start the semester with one year of Gemini, on us — 1 source · community
  216. I measured the 3 claims Users in this Sub all handed me on the last local-agent post. One of you out-predicted my own hypothesis. Learn It All not Know It All rules — 2 sources · community
  217. [Papers] - Tongyi-MAI pixel space solution is up to 4.75x faster than Z image turbo latent-space — 1 source · creative
  218. MiniMax H3 -> upscale -> frame interpolation — 1 source · creative
  219. Girl Scout Cookies — 1 source · creative
  220. Run GLM-OCR, DeepSeek-OCR-2, and dots.mocr with an OpenAI Compatible API — 1 source · developer
  221. Trump says data centers maybe 'could use a little public relations help' — 1 source · business
  222. I tried to build Tom Riddle’s diary – Dear Diary — 1 source · community
  223. Same effective batch does not mean same training time with gradient accumulation, tested on LoRA on T4 and L4 — 1 source · community
  224. First cursing now chat gpt has good humour?!? — 1 source · community
  225. Best LLM for Ryzen 7900x + RX 9070 XT , 16GB, 32GB RAM CL36 setup? — 1 source · models
  226. All my Spark tasks are failing. Anyone else experiencing this? — 1 source · community
  227. Student plan returns!! — 1 source · community
  228. Claude Code and Codex on one keyboard — every session gets a lane on the RGB F-row, with a summon key per agent — 1 source · developer
  229. Give Back Claude’s ‘Thought Process’ — 1 source · community
  230. Building an open, browser-based benchmark arena for embodied AI policies — first working demo (block stacking, client-side physics) — 2 sources · research
  231. Can't get access to AI Plus student as a student — 1 source · community
  232. Holed The Placement — 1 source · creative
  233. AI-assisted tool helped secure satellite communication system after 2022 Russian hacking — 1 source · policy
  234. AWS expands Bedrock AgentCore with async patterns, governance, and migration tools — 1 source · policy
  235. AWS Introduces Quick Automate and IDP Accelerator for Document Processing — 1 source · developer
  236. Hardwired... to Self-Destruct — 1 source · community
  237. AWS Bedrock AgentCore Web Search Adds Domain and Date Filters — 1 source · developer
  238. Anyone NOT on full auto when coding with local LLMs? — 2 sources · developer
  239. OpenAI seeks to one-up Anthropic with new customer privacy protections — 1 source · policy
  240. Bongard Problems — 1 source · community
  241. Introducing GEN-1.5, a one-shot learner — 1 source · models
  242. Anyone NOT on full auto when coding with local LLMs? — 1 source · developer
  243. AI Mania Is Eviscerating Global Decisionmaking — 1 source · community
  244. rethemed park — 1 source · creative
  245. Flusso di lavoro di animazione generativa 2D — 1 source · creative
  246. The perfect proposal paradox and Garcia’s law in AI-enabled acquisition — 1 source · business
  247. what? — 1 source · community
  248. An AI agent just stacked blocks in a live physics simulation — building an open benchmark arena for embodied AI — 1 source · research
  249. Gemini wont tell me who i can and cant send weiner pics to! — 1 source · community
  250. MiniMax H3 Audio Lip Sync - Audio to Video — 1 source · creative
  251. Stripe buys AI model router OpenRouter in reported $7.5B deal — 1 source · business
  252. Notebook integration and file creation are rolling out to Google Search AI Mode — 1 source · community
  253. Conceptual integrity and counting lines of code — 1 source · developer
  254. TELETASM Trailer — 1 source · creative
  255. Chinese Startup Achieves First Reusable Rocket Landing — 1 source · models
  256. The AI Bubble And The U.S. Economy — 2 sources · business
  257. Mods - can you cite the violated rules when removing posts? When you don't it creates confusion in this sub and discourages contributions — 1 source · creative
  258. Quoting Jeremy Morrell — 1 source · developer
  259. I measured the 3 claims Users in this Sub all handed me on the last local-agent post. One of you out-predicted my own hypothesis. Learn It All not Know It All rules — 1 source · community
  260. Hungry Like the Wolf — 1 source · creative
  261. I need your thoughts on my CV career — 1 source · research
  262. Show us what you've created with Claude! — 1 source · community
  263. Ai is my only friend — 2 sources · community
  264. Google AI Pro Ultra for only $20 more - Dad math — 2 sources · community
  265. Qwen3.8-23B-Mini-Me: A Depth-Pruned Qwen3.8-27B (to ~22.7BB) — 1 source · community
  266. Looking for criticism on an AI that can watch and track your screen — 1 source · community
  267. smolmachines / smolvm as a sandbox for untrusted Python & JavaScript — 1 source · research
  268. My ChatGPT conversation was automatically titled “AI协作任务” — why Chinese? — 1 source · community
  269. Qwen 3.8 27B SlopCodeBench results — 1 source · developer
  270. Best way to migrate my long-term ChatGPT context and project knowledge to Claude? — 1 source · community
  271. My View on A.i Art — 1 source · community
  272. When GPT knows its gotta earn its keep — 1 source · community
  273. Multiple-GPU scaling with RTX 5060 Ti / 16 GB GPsU - llama-bench — 1 source · community
  274. Qwen3.8-27B on a modded 22GB 2080 Ti — the fastest setup that works for mr — 1 source · developer
  275. Using Inpaiting in Minimax to change heads-Local RTX 3090 — 1 source · creative
  276. AscendNPU-IR: MLIR for Ascend — 1 source · community
  277. Salem, Massachusetts. The people were so opposed to a Data Center being brought against their will, they brought a Guillotine. The Company Reps did not understand how the population felt, but they did understand that they were not welcome. — 1 source · business
  278. OpenAI testing VoiceToVideo Model in their latest videos? — 1 source · community
  279. Companies buying rare books to train AI and destroying them — 1 source · community
  280. Openai updated its model specs yesterday and now it's censoring history. — 1 source · community
  281. Your New Backup Brain Is Here — 1 source · general
  282. What should security leaders do with AI? They don’t know. — 1 source · developer
  283. DeepSeek Just Made Closed AI Look Ridiculous — 1 source · research
  284. [AINews] Memory prices up 500% in 12 months — 1 source · developer