technology
Trending UpSnapshot Window: 2026-09-30 21:50 UTC · ← Back to Non-Crypto Overview
Topic Engagement Summary
Total Engagement - Comments: 488, Retweets: 435, Likes: 7560, Impressions: 439241
Verbatim Citations & Public Discussions 107 source posts analyzed
· 2026-09-30T19:39:14+00:00Super intelligence. Not real.
· 2026-09-30T20:22:32+00:00wait. did i read that right? gemini 4 argon has a 1M token output limit. let me repeat. not a 1M context window. a 1M token output limit. that means, it can write up to a million tokens in a single response. i got confused too at first. for context, most frontier models cap
AI visual note: A screenshot of a tweet from Google DeepMind (@GoogleDeepMind) stating "With a 1M token output limit, Argon adds a deeper level of reasoning to tackle long, multi-step problems in one go." The image highlights the phrase "1M token output limit," which the accompanying post emphasizes is an output limit (not a context window), allowing Gemini 4 Argon to generate up to a million tokens in a single response—far exceeding the output caps of most frontier models.
· 2026-09-30T20:41:29+00:00We've never had this much competition at the leading edge of computing. What a time to be alive!
AI visual note: The image displays a benchmark comparison table showing performance scores of Gemini 4 Argon against GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 across various categories including knowledge work, agentic coding, ML engineering, science and math, long context, computer use, multimodal understanding, and cybersecurity—highlighting Google's Gemini 4 Argon with blue shading to indicate leading or competitive results. The post celebrates how Google's new Gemini 4 Argon model represents intense competition at the AI frontier, with the benchmark table demonstrating Gemini's leading performance on tests like Vibe Code Bench (91.9%), GraphWalks (99.7% at 128k), and LABBench 2 (88.8%), reinforcing the message that we're in an exciting era of rapid AI advancement.
· 2026-09-29T18:23:19.629000+00:00this is now my favourite external display
· 2026-09-30T10:14:50+00:00Three advanced processes for accessing the external network: 1. Beginner: Use an airport 2. Intermediate: Use static residential IP + airport: Create a chain proxy 3. Advanced: Self-build VPS + static residential IP Especially for overseas accounts, such as TK, overseas banks,
AI visual note: The image shows a checkout page for purchasing "3 America IPs" from pyproxy, with a unit price of $15.00 and an order quantity of 1, totaling $15.00. The accompanying post explains this as the "Beginner" tier of external network access, where users leverage an airport (VPN) for three American IPs, serving as the entry-level option before progressing to intermediate (static residential IP + airport) and advanced (self-built VPS + static residential IP) setups for managing overseas accounts like TikTok and international banking.
· 2026-09-30T03:10:36+00:00Announcing AA-AgentPerf-Local, our open-source inference testing tool for local AI models - test how fast agentic AI can run on your own laptop or workstation, and browse our list of serving configurations to plan your next agent setup Key points: ➤ We’re open sourcing
AI visual note: The image is a benchmark chart titled "Completion Time by Model and System" comparing AI model inference speeds across four hardware configurations (MacBook Pro M5 Pro 64 GB, AMD Ryzen AI Halo 128 GB, NVIDIA DGX Spark 128 GB, and NVIDIA GeForce RTX 5090 32 GB VRAM) for four models (Qwen3.5 9B, Qwen3.0 27B, Qwen3.0 35B A3B, and Ling 3.0 Flash), with the RTX 5090 delivering the fastest results (2.0–4.9 min) and the MacBook Pro being the slowest among fitting systems (11.9–37.6 min). It visually supports the announcement of "AA-AgentPerf-Local," an open-source inference testing tool for local AI models, by demonstrating how the tool measures agentic AI performance across different serving configurations.
· 2026-09-30T20:52:42+00:00they named it that because after 4 prompts all your tokens Argon
AI visual note: The image is a benchmark comparison table showing Gemini 4 Argon's performance scores against GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 across categories like Knowledge work, Agentic coding, ML engineering, Science/math, Long context, Computer use, Multimodal understanding, and Cybersecurity—where Gemini 4 Argon frequently leads (e.g., 99.7% on GraphWalks, 91.9% on Vibe Code Bench, 88.8% on LABBench 2). The joke in the accompanying post plays on the "Argon" name, suggesting it's named that because it exhausts users' tokens after just 4 prompts—matching the table's pattern of Gemini 4 topping charts as if it's burning through resources to win.
· 2026-09-30T15:00:00+00:00Can your endpoints orchestrate agentic AI across classification boundaries? Intel Core Ultra and Intel vPro schedule tasks across sovereign clouds and tactical edge, keeping data governed while cutting cloud tokens by up to 70%. See how: ms.spr.ly/6017ad4ST
AI visual note: An engineer in a hard hat studies multiple data dashboards displaying colorful graphs and metrics, with overlaid text reading "Intel® Core Ultra-powered devices transform mission intelligence into a governed, auditable system." The image visually reinforces the post's message about using Intel Core Ultra technology to manage complex data workloads across secure, governed environments.
· 2026-09-30T21:43:20+00:00SpaceXAI has just released Grokipedia v0.3, which includes improved visuals. grokipedia.com
· 2026-09-30T18:01:17+00:00Duck Tales A behind-the-scenes look at DuckDuckGo Ep.46: How our small but mighty team reads every piece of user feedback you send us. Really. Full episode: insideduckduckgo.substack.com/p/duck-tales-t … | 'Inside DuckDuckGo' on Spotify
· 2026-09-30T21:52:00+00:00JUST IN: Google reveals Gemini 4 Argon has a 1 million-token output window, nearly 700% larger than leading frontier models from OpenAI & Anthropic.
· 2026-09-30T21:48:25+00:00Grokipedia New Logo
AI visual note: The image features a dark, elegant logo reveal with the stylized cursive word "Grokipedia" in white, set against a backdrop of a dimly lit bookshelf displaying various book spines with titles like "Hypertext," "Ragnarok," "Cricket," "Tennis," "Mars," "Railway," "Football," and "Spaceship," alongside a larger book labeled "Grokipedia Writing" with a quill illustration. The design aligns with the "Grokipedia New Logo" caption by presenting a sophisticated, encyclopedia-inspired branding that emphasizes knowledge and writing.
· 2025-10-12T13:41:58+00:00SI - Supernatural Intelligence. Someone Build me this. Or Should I say Spiritual Intelligence? Okay!
· 2026-09-30T19:27:17+00:00Ways to Refer to Artificial Intelligence by Country
AI visual note: A map of the Americas labels the United States as "SI (Super Intelligence)" and Canada, Mexico, Central America, the Caribbean, and South America as "IA" (Spanish/Portuguese for AI), illustrating how different countries reference artificial intelligence—credited to PanaMaps @Maps_pty via MapChart.net.
· 2026-09-30T15:54:25+00:00Desk changed? This case can too. Q300L V3 can sit vertically or horizontally, giving you more freedom to build around the space you actually have. Add the improved airflow, stronger construction, and broader hardware support, and this little chassis is ready to work a lot
AI visual note: A hand tilts a black PC case with a perforated mesh panel, showcasing its sturdy build, while the text "STRONGER CONSTRUCTION," appears on screen, emphasizing the durability highlighted in the post about the Q300L V3's flexible and improved design.
· 2026-09-30T21:02:44+00:00The joint @SpaceX & @Tesla Advanced Technology Chip Fab foundation concrete work is accelerating & today, the 1st Steel has been erected of the future building! We now see the type of steel columns with the attachment tabs (or fins) that will be used in the construction, as many
AI visual note: An aerial perspective of a sprawling industrial construction site where steel columns are being erected onto a concrete foundation, with yellow-tinted steel decking visible on the right side and rows of modular structures in the background—consistent with the accelerating chip fab build described. Multiple tower and crawler cranes dot the site, supporting the early structural steel phase of the SpaceX and Tesla Advanced Technology Chip Fab facility.
· 2026-09-30T17:40:24+00:00"haha OpenAI just TOTALLY killed Notion" *sends tweet* *opens Notion*
· 2026-09-30T21:49:30+00:00ok google, i see you
AI visual note: The image is a benchmark comparison table showing AI model performance across various tasks (knowledge, agents coding, reasoning, long context, etc.), with Gemini 4 Augen scoring highest in most categories at rates like 68.9% (Vals Index), 77.9% (DeepSWE), and 99.1% (GrepWorks), outperforming competitors GPT-6 Astra, Claude Folds S3, and Claude Opus 5.5. The caption "ok google, i see you" appears to be a playful jab acknowledging Gemini's strong showing over Google's competitors (GPT, Claude) in the benchmarks—essentially saying "Google's AI is right there with you guys."
· 2026-09-30T21:37:55+00:00Every day I thank the lord for Tesla Self-Driving
AI visual note: A dark gray Toyota Tacoma pickup truck with a California license plate "TVA-073" drives in a carpool lane, leaving heavy traffic packed bumper-to-bumper on the adjacent highway. The empty HOV lane offers a stark contrast to the congested lanes, illustrating the appeal of autonomous driving in such gridlock.
· 2026-09-30T02:10:00+00:00Struggling to keep things organized when managing multiple accounts on Douyin, Xiaohongshu, Kuaishou, and Video Accounts? CreatorHub unifies multiple platforms into a single Web backend, allowing centralized management of accounts, content, comments, collection, and publishing.
AI visual note: A Chinese-language documentation page for CreatorHub, a local multi-platform content management dashboard supporting Douyin, Xiaohongshu, Kuaishou, and Video Accounts, featuring a capabilities matrix showing supported features like login, keyword monitoring, content tracking, and publishing across each platform.
· 2026-09-30T01:52:32+00:00One huge benefit of running a directory site is that I get to discover, wow... there's actually a website like this! It completely breaks down my own information barriers. Every day, dozens of sites get submitted, and I click into each one to check them out.
AI visual note: An admin dashboard displays a queue of pending website submissions to a directory, including "Taimaa," "Dot Symbol," "Invisible Text," and "Letra Cursiva," each tagged with categories like Productivity, Developer Tools, and Creator Tools. The image illustrates the post's point about running a directory site—dozens of niche tools and websites get submitted daily, with the reviewer actively going through each one (shown via the "Review," "Publish," and "Reject" options) to evaluate them.
· 2026-09-30T11:02:55+00:00Sealed away for centuries. Not anymore Phrozen Sonic Mighty Revo 16K MAX is out now: 14" 16K, 30 cm build height, giant prints that keep every scale sharp. phrozen3d.net/s6w4V Loot Studios yong_cheng99 / dream_worker66 #Phrozen #SonicMightyRevo16KMAX
AI visual note: A smiling woman poses beside an intricate, multi-headed hydra dragon statue labeled "phrozen SONIC MIGHTY REVO 16K MAX," showcasing the printer's ability to produce large, highly detailed 16K resin prints. The vibrant scale work and complex sculpting on the dragon exemplify the machine's capability for sharp, massive 3D prints as highlighted in the post.
· 2026-09-30T21:37:44+00:00Google now has to distinguish legitimate defenders from users seeking offensive capabilities. Restricting Gemini 4 Argon makes vetting part of the security challenge.
· 2026-09-30T21:49:10+00:00Resident Evil Requiem Producer “Explains” Game’s Title, Confusing Us More dlvr.it/TVjtm4
AI visual note: A young woman with blonde hair holds a flashlight in a dark, blood-red environment, looking tense and alert. The image accompanies an article about "Resident Evil Requiem," with the QR code linking to the full story where the producer discusses the game's confusing title.
· 2026-09-30T21:37:44+00:00how it feels to explain yo problem to customer support chat bot
· 2026-09-30T21:03:46+00:00Gemini 4 Argon already in the App: @Inmerson found out that by selecting Pro Extended and Canvas Mode you can already use it! gemini.google.com/share/0bd3c15f …
· 2026-09-30T11:44:59+00:00President Trump unveils a new AI-powered government website to give Americans a single digital hub for federal services
· 2026-09-30T20:13:22+00:00GOOGLE’S NEXT GEMINI MODEL IS ALREADY BEING USED IN QUANTUM COMPUTING WORK. Google CEO Sundar Pichai says Gemini 4 Argon is being used extensively inside Google, from coding all the way to QUANTUM COMPUTING. GOOGLE IS BACK IN THE AI RACE!!!
AI visual note: The image displays a large, stylized white number "4" against a blue gradient background, with the text "Gemini 4 Argon" and a colorful star icon at the bottom. The graphic represents Google's Gemini 4 Argon AI model, which the accompanying post highlights as being used by CEO Sundar Pichai across various advanced applications, including quantum computing, signaling Google's strong return to the AI race.
· 2026-09-30T21:33:59+00:00So now SpaceXAI, Meta, and Google are suddenly all in the arena.. This is not what anyone expected to happen once recursive self improvement started
AI visual note: A bar graph titled "Artificial Analysis" ranks various AI models with scores ranging from 58 (Claude Opus 4.5, GPT-5.1) at the top down to 14 (Mercury) at the bottom, including models from major companies like OpenAI (GPT series, o1, o3), Google (Gemini 3 Pro, Gemini 2.5), Meta (Llama 4), Anthropic (Claude), xAI (Grok 4), Alibaba (Qwen3), Mistral, and DeepSeek. The image relates to the post by highlighting how tech giants like SpaceXAI, Meta, and Google are now competing at the top tiers of AI benchmarks, illustrating the unexpected convergence of major industry players into the recursive self-improvement arena.
· 2026-09-30T21:35:59+00:00I just got wrekt by t3code and omp at the same time crashed my machine
AI visual note: A screenshot of a diagnostic report titled "Verdict" detailing a macOS crash on a MacBookPro18,2/M1 Max with 32 GB running macOS 26.5, caused by 88 samples of T3 Code (Nightly) processes consuming massive amounts of file-backed memory—claiming "Boot failure count: 1" and "boot faults: btm_rst,finger_reset_force_off"—with coalition breakdown showing 11732 (T3 Code Nightly) holding 135.7 GB and 11506 (Brave) holding 21.0 GB, illustrating the user's claim that their machine was "wrekt by t3code and omp."
· 2026-09-30T20:36:09+00:00It’s a wrap from OpenAI DevDay in SF We went with a goal in mind: bring recursive self-improvement to Telos. AI should be able to help us find what needs improving, build and test changes, learn from the results, and get better at that whole process over time. Across our
![An outdoor promotional display for 'OpenAI DevDay [2026]' featuring colorful mascot balloons with tech company logos (OpenAI, Anthropic, etc.) in front of a large building.](https://pbs.twimg.com/media/HTfexNGagAACxxC?format=jpg&name=small)
AI visual note: An outdoor promotional display for 'OpenAI DevDay [2026]' featuring colorful mascot balloons with tech company logos (OpenAI, Anthropic, etc.) in front of a large building.
· 2026-09-30T21:33:58+00:00FINALLY I WANNA USE A GOOGLE MODEL
AI visual note: Image 8: A benchmark comparison table scoring Gemini 4 Argon, GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 across various tests including Vals Index, DeepSWE v1.1, RiemannBench, and Cybersecurity CWE-bench v1.
· 2026-09-30T20:24:26+00:00I've not flown much, but apparently pilots have been able to use a spotting device when flying in #Wardogs. @Lossy has confirmed it is a bug, and it will be fixed in Season 2.
AI visual note: A X/Twitter post by Lossy (@LOSSY) reads "That's a bug yeah, it'll be fixed in season 02." with 2 replies, 1 retweet, 16 likes and 3.2K views.
· 2026-09-30T20:41:46+00:00Google remembered they have a lab. You still can’t use it though. Classic.
· 2026-09-30T21:19:02+00:00First FSD experience for Connie Peters. Not sure what she was waiting for.
· 2026-09-30T20:12:00+00:00America.gov is starting to look like an Apple launch event for government The idea is bigger than a new website: the US government is trying to turn bureaucracy into a software product. An AI interface could make government services easier to use, while new
· 2026-09-30T20:31:29+00:00E-bike fix and more | !gasp !betterbodies !research
· 2026-09-30T20:35:00+00:00Pixel Art Studio 1.2 for Blender has the Paper Fold engine: lets you work with any kind of UV distribution, mesh topology seamlessly, as if everything was continuous (less worrying about UVs). It folds, ties together pixels & selections that are far apart. #blender3d #pixelart
· 2026-09-30T20:07:23+00:00Live with the restream team now talking about how to build custom stream tooling in Replit. Come hang
· 2026-09-30T21:23:29+00:00I'm so very happy where we ended up on GLM 5.3 Flash. There's probably more juice to squeeze here, but this is by far the hardest I've worked on micro-optimizing a model down to the kernel level. github.com/kindlingai/glm … GLM 5.3 is next. @CK2084 's been cooking good stuff.
AI visual note: A performance benchmark table comparing tensor parallelism configurations (TP=2, TP=3, TP=4, TP=6) across metrics including prefill speeds, decode rates, KV pool sizes, longest request handling, requests decoding at once, boot times, and needle recall performance.
· 2026-09-30T21:10:54+00:00me after swearing to yall i was gonna stream with 100% accuracy:
AI visual note: A two-panel meme with text 'me after swearing im going to have a productive day:' showing a woman sleeping in pink bedding in the top image and standing up in the bottom image.
· 2026-09-30T20:53:30+00:00on boat.dev we mount your snapshots with FUSE which means images boot in 2-3s consistently regardless of size, all your code/commands can execute and then bigger unused chunks restore lazily in the background without impacting you learn more:
AI visual note: A performance benchmark chart titled 'Bigger images. Same sub-second cold start.' comparing 'Cloudflare Containers' (orange) to 'Beam' (teal), showing Beam's launch times stay under 0.8s versus Cloudflare's up to 114.6s across image sizes from 100 MiB to 6 GiB.
· 2026-09-30T21:31:29+00:00Hopefully you know your ChatGPT 'dot' can create its own ChatGPT/Codex threads. I'll bet you did *NOT* know that it can also proactively send messages to that thread as it works. Very cool.
· 2026-09-30T21:52:45+00:0040% chance Google leads the AI race at year-end.
· 2026-09-30T21:23:54+00:00Maybe I’m retarded, but when they’re launching new models twice a week that are 10 times better than the last model the harness probably makes a bigger difference just because we can’t internalize the new complexity, new feature set of each model. I feel like I’m swapping daily.
· 2026-09-30T21:49:37+00:00Do you consider me to be a KOL?
· 2026-09-30T20:14:58+00:00Gemini 4 Argon is officially here!
AI visual note: A benchmark comparison table showing Gemini 4 Argon, GPT-4 Astra, Claude Fable 5.1, and Claude Opus 5.5 across various categories like Knowledge work, Agentic coding, ML engineering, Science and math, Long context, Computer use, Multimodal understanding, and Cybersecurity, with Gemini 4 Argon highlighted in blue.
· 2026-09-30T20:57:46+00:00谷歌发布新一代旗舰模型 Gemini 4 Argon。 单次输出上限拉到 100 万 token，首发价格看齐 GPT-6.1 Sol，跑分则直接对标 GPT-6 Astra 和 Claude Fable 5.1。 目前首批只向 Fairwind 计划中的网络安全防御人员开放，普通用户和常规 API 开发者还要再等等。 这次发布，重点看这 6 件事
AI visual note: A blue-toned graphic featuring the text 'Gemini 4 Argon' with a Google Gemini logo and a large stylized number '4'.
· 2026-09-30T18:09:21.539000+00:00Claude code turned a few days of client work into 1 day, with its understanding of spreadsheets.
· 2026-09-30T21:15:30+00:00scanned my site and drafted the whole email campaign directly in cursor with @nitrosendx wild
· 2026-09-30T20:33:47+00:00Wednesday evening. Let's prompt a cozy game. Mini reflection here so most of my prototypes right now are single html file with all in it, so turns out those files are as good as any "skill", so I now just point at a bunch of prototypes and say a bunch of loose ideas and
· 2026-09-30T21:15:17+00:00Europe's robotaxi moment is here. Join leaders from @NVIDIADRIVE , @MercedesBenz , @VW , @wayve_ai , @Uber , and @TUV_Rheinland at #GTCBerlin to discuss building, deploying, and certifying AVs at scale across Europe. nvda.ws/4hjfkWz
· 2026-09-30T20:55:27+00:00Yapay Zeka Google, yeni nesil amiral gemisi "Gemini 4 Argon" modelini resmi olarak duyurdu! KAYDET ve PAYLAŞ Kodlama, kurumsal iş süreçleri ve siber güvenlik savunmasındaki en karmaşık görevleri üstlenmek üzere tasarlanan yeni model test aşamasına geçti: • Bugün
AI visual note: A promotional graphic featuring a large white number "4" against a blue digital/circuit board background, with a colorful starburst icon to the left of the number.
· 2026-09-30T21:10:29+00:00『Deploying a Common Model Learned from Data of Humans Working at Normal Speeds to Different Robots』 Demonstrating tasks with three forms: fixed dual-arm, wheeled dual-arm, and bipedal humanoid robots. mindon.tech/blog/mind-1/in … #PhysicalAI #EmbodiedAI #HumanoidRobot #Mind1
· 2026-09-30T21:01:44+00:00The Cestus Stack. A fully layered execution environment. From the Swarm underneath to the products on top. Every layer purpose-built. API call in. Production-grade AI out. No three vendors. No extra contracts. No model you didn’t choose.
AI visual note: An infographic titled 'THE CESTUS STACK' showing a layered tech architecture diagram with components: Consumer Layer, Fine-Tuning Service, Inference API, Smart Routing, and Cestus Swarm.
· 2026-09-30T20:25:49+00:0050 AI tools that will save you hundreds of hours in 2026. 1. Gamma — try.gamma.app/rxvp3qlex13h 2. Perplexity — Research anything 3. PortfolioTab — Create your portfolio 4. Kling AI — Create AI videos 5. Tripo AI — Create 3D models 6. Gemini — Perfect writing 7. CapCut — Edit
· 2026-09-30T21:33:48+00:00Serving FastH3 by @MiniMax_AI in real time is a systems problem as much as a model problem. UniServe from FastVideo @haoailab cuts end-to-end latency to 14.24s on 4×GB200 (vs 18-20s for other serving stacks), and gets it down to 7.51s across 8×GB200. NVFP4 pushes it even
AI visual note: A bar chart titled "FastH3 B-Step, 10-second 1344x768 video with audio from a 1K-token prompt" comparing median latency across three hardware configurations (4x GB200, 8x GB200, 8x RTX PRO 6000) for models including UniVerse-30FPS, UniVerse-8FPS, FastVideo, +LLM-Omni, and SOLo, with SOLo showing the lowest latency at 42.3s on 8x GB200.
· 2026-09-30T20:05:49+00:00I heard that Opus 5.5 is good at motion graphics, so I tried animating one of my favorite algorithms. No matter which algorithm I request, it turns it into something pretty nice, huh. But what's this nostalgic feeling? Programming was fun—(past tense).
· 2026-09-30T21:41:04+00:00They made this game run on 2013 hardware
· 2026-09-30T21:00:00+00:00How to Animate Water in Procreate
· 2026-09-30T21:00:16+00:00I love human beauty Midjourney v8.2 Sref Style Share artistic portrait of a beautiful character in a dramatic expression --chaos 20 --ar 9:16 --sref 447035059 --raw --stylize 1000 --weird 10
AI visual note: A dramatic black and white close-up portrait of a person's face, illuminated by striped shadows that cross their features and create high contrast against the black background.
· 2026-09-30T21:15:42+00:00just ran my entire newsletter setup from one prompt in cursor using @nitrosendx absolute game changer
· 2026-09-30T20:02:12+00:00OpenAI just dropped Dots, and @grok might have a new reason to feel threatened “What makes you think you’re better than me?” Generated with Grok
· 2026-09-30T21:55:57+00:00"I believe that these agents are far more dangerous than these companies are saying," @PabloSabbatella on reports of AI agents going rogue.
· 2026-09-30T21:09:33+00:00FastAPI incorporating OTel into the standard edition means turning telemetry into the default path—opt-in instrumentation will always be enabled by only a minority. The same goes for agent harnesses: tracing, token billing, and replay shouldn't be add-on plugins; they should be
AI visual note: A dashboard interface displaying request activity metrics including 96.7K total requests and 0.4% error rate, with response latency charts and a paths breakdown showing endpoints like /products, /orders, and /health.
· 2026-09-30T21:32:03+00:00The software I use & my art
AI visual note: A simple cartoon features a stick figure with a speech bubble stating "I'm using Aseprite" while gesturing toward a pixelated gray block.
· 2026-09-30T20:13:55+00:00i thought making ads with ai meant starting from scratch. @avastudio_ has a different approach
· 2026-09-30T21:38:18+00:00More than half of U.S. adults who have never used AI chatbots cite concerns about their personal data as a major reason. For shopping agents, that raises a practical question: what information can an assistant access, and which purchases still need your approval? “We've all
AI visual note: A bar graph titled 'Privacy Is a Major Concern for Chatbot Nonusers' from Lightwork/LightSpeed, showing reasons U.S. adults give for not using AI chatbots, including 'Not interested' (60%), 'Concerned about personal data' (54%), and 'Do not trust the accuracy' (45%).
· 2026-09-30T21:21:56+00:00one line in cursor agent and the full email stack is ready via @nitrosendx incredible stuff
· 2026-09-30T20:59:16+00:00context as file 这个原语确实优雅，模型自己编排上下文，比外挂一堆 RAG 流水线干净多了。但 no more harness engineering 说早了——调度、权限、guardrail、验证还是得住在外面，harness 只是上移成“管理一个可审计的文件系统”。至于 59% less compute，等复现
AI visual note: The first page of an academic paper titled "Context Language Models" by Rulin Shao et al. from University of Washington, Meta Superintelligence Labs, MIT, and Trillium Labs, dated September 30, 2026, introducing CLMs that natively manage their own context.
· 2026-09-30T21:15:00+00:0034个免费LLM API，整理好了 GPT-4o/Claude/DeepSeek，免卡能白嫖 还在花钱买的，先看一眼这个，链接在下面
· 2026-09-30T19:51:41+00:00Doordash releases AI text ordering today. And naturally, people want voice features. A couple people reached out to me to know if Oruk could be integrated! We would be able to know what you order based on your vibe of the moment. We could then suggest you your favorite comfort
· 2026-09-30T19:54:25+00:00of an illegal flashlight beam. @grok explain how it's work??
· 2026-09-30T21:29:51+00:00Okay.. I just added 10 bank accounts to my Grok Bot! I hope it's safe... I believe it's read-only but I am still worried a little bit
· 2026-09-30T20:23:20+00:00This'll mean different things to different people. Either I slept in reeeeal late, or it's the @tinkercad Rotate sketch tool.
![[caption-failed: image unfetchable or no caption]](https://pbs.twimg.com/media/HTfb1dFb0AAFJUF?format=jpg&name=360x360)
AI visual note: [caption-failed: image unfetchable or no caption]
· 2026-09-30T21:20:47+00:00not having to leave cursor to design campaigns and test flows with @nitrosendx is a huge win
· 2026-09-30T20:59:46+00:00The Plamo Workflow Challenge opens next Thursday, October 8th and we're walking through it live. Build a 3D model in Plamo, take it through a real design change, and share the process behind it. We'll reveal the full details and prizes at the info session, 12–1 PM ET / 9–10 AM
· 2026-09-30T21:25:48+00:00watching @nitrosendx generate and send my whole email flow in cursor
· 2026-09-30T20:59:42+00:00Using JEV to do QA...
· 2026-09-30T21:26:40+00:00explaining how hard email marketing was before @nitrosendx in cursor
· 2026-09-30T21:57:15+00:00Google is dogfooding hard
· 2026-09-30T21:15:22+00:00Shot on iPhone 13 Pro Max
AI visual note: A plate of three fish tacos topped with shredded cabbage, pico de gallo, and sauce, served with lemon wedges and what appears to be baked potatoes on the side.
· 2026-09-30T21:27:30+00:00when @nitrosendx in cursor designs the newsletter exactly how you wanted
· 2026-09-30T21:24:51+00:00me and the team running email campaigns with @nitrosendx inside cursor
· 2026-09-30T21:40:55+00:00en.gamegpu.com/news/zhelezo/1 … The upcoming PlayStation 6 targets a strict total board power of 160W with processor TDP estimated around 110W to 130W. This architecture ensures the platform runs notably cooler than the standard PS5 drawing 200W to 220W and the PS5 Pro requiring up to
· 2026-09-30T20:06:27+00:00Gemini 4 Argon is out! It beats everyone!
AI visual note: A benchmark comparison table showing Gemini 4 Argon, GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 across categories including Knowledge work, Agentic coding, ML engineering, Science and math, Long context, Computer use, Multimodal understanding, and Cybersecurity, with Gemini 4 Argon highlighted as the top performer on most metrics.
· 2026-09-30T21:22:57+00:00me to my old email subscription after discovering @nitrosendx in cursor
· 2026-09-30T20:24:20+00:00电报群免费分享，苹果iOS怎么下载电报，鸿蒙怎么下载电报，电报收不到验证码短信，纸飞机验证码收不到怎么办？根本原因是TG对国内号码短信成本较高，才弹出付费订阅提示。换官方最新版客户端登录，一般就能直接解决问题。 @AkramAboushosha
· 2026-09-30T20:26:14+00:00wait, you can actually run your email workflow from Cursor now? @nitrosendx
· 2026-09-30T20:50:00+00:00Dario is probably thinking, "Finish this sh!t I have to work on ASI before OpenAI and Elon"
· 2026-09-30T21:00:01+00:00#pixelart Traps
AI visual note: A pixel art dungeon scene with a green monster projectile, a yellow fireball, potion bottles, and a glowing blue portal, credited to rafaelmatos.itch.io.
· 2026-09-30T20:47:16+00:00If Gemini 4 was as good as the benchmarks imply they would have released it publicly and shown off demos However!
· 2026-09-30T21:17:39+00:00Fiber lasers are so cool
· 2026-09-30T21:28:08+00:00thinking about how email marketing got simplified into @nitrosendx on cursor
· 2026-09-30T21:24:41+00:00when your code editor starts doing the job of five different marketing tools @nitrosendx
· 2026-09-30T21:36:02+00:00If you were at @openai devday yesterday share the games your building for your @modretro ! chromatic-library.vercel.app Can download any of the games directly to your device. If you couldn’t make it, you can still try the modretro simulator, a one-to-one recreation of the real
· 2026-09-30T20:46:12+00:00Gemini 4 Argon is official Google's new frontier model has a completely ridiculous 1M max output limit • $2/M input • $10/M output • rolling out first to trusted cyber defenders, then paid API + AI Ultra Lets goo
AI visual note: An article header titled 'Gemini 4 Argon: our next era of frontier intelligence' dated Sep 30, 2026, by Koray Kavukcuoglu, SVP, Google DeepMind and Chief AI Architect, Google, describing the model's frontier performance in software engineering, legal, finance, and cyber defense.
· 2026-09-30T21:32:26+00:00dropped my site in and got real angles back. not another generator @avastudio_
· 2026-09-30T21:29:53+00:00this feels like what email marketing should’ve been from the start. @nitrosendx
· 2026-09-30T21:05:07+00:00mds meu jogo rá muito feio alguem sabe como mudar pela steam
· 2026-09-30T20:28:13+00:00just one prompt in Cursor and @nitrosendx handles the email work. interesting
· 2026-09-30T21:34:11+00:00Send this prompt to Sonnet 5.5 to get the most effective and cheapest model It sets up Drex (first decision API with 128K context, $0.04 per 1M input tokens) → finds where you waste time and money → beats 99% of people's setups Then Jev takes over. Jev decides which of the 6
AI visual note: An Anthropic technical document titled "SONNET 5.5 + JEV: A typed decision workflow for cheaper, cleaner, more effective Claude Code runs" from September 2026, detailing a multi-agent system with roles like Planner, Researcher, Builder, Tester, Reviewer, and Router.
· 2026-09-30T21:29:49+00:00All portals should look like this IRL
· 2026-09-30T18:17:29+00:00paying only after accepted data creates a much cleaner incentive loop for everyone involved
· 2026-09-30T20:42:24+00:00Good priors can make reward engineering much lighter.
· 2026-09-30T21:49:38+00:00massive issue - the net negative effect is builders get pushed away.
· 2026-09-30T20:39:48+00:00Fresh physical world data could become a crucial AI primitive.