•twitter19895
listening to
timeline
19895
mentions
tracked
tracked
net sentiment
+23%
6088 positive
12325 neutral
1482 negative
volume & sentiment over time
Sep 30 – Oct 18, 2026 · Daily
sources
most discussed
—
active days
19
mentions
Showing 10251–10275 of 19895 mentions
-
• twitter 1w neutralNew client new intro Thought? 👀 https://t.co/fvaszePiOW New client new intro Thought? 👀 https://t.co/fvaszePiOWview source ↗
-
• twitter 1w positiveI tried sending Facebook traffic to my blog and I'm a little shocked. The Facebook page is maybe 15 days old. The blog is new too. I only added links under 3 reels and got decent results. I can definitely scale this. Honestly, I think I missed this opportunity before. https://t.co/hj72OayJQT I tried sending Facebook traffic to my blog and I'm a little shocked. The Facebook page is maybe 15 days old. The blog is new too. I only added links under 3 reels and got decent results. I can definitely scale this. Honestly, I think I missed this opportunity before. https://t.co/hj72OayJQTview source ↗
-
• twitter 1w positivenot a great month for domain renewals but brilliant month for product focus https://t.co/Xi9GOZEYWY not a great month for domain renewals but brilliant month for product focus https://t.co/Xi9GOZEYWYview source ↗
-
• twitter 1w neutralthe token savings can be pretty insane. Gobstopper saves 29% in tokens on Terminal-Bench 2.1 compared to Claude Code's built-in compaction. but in practice you can end up saving a lot more on long running sessions ask your agent to set it up: https://t.co/q6PBGsbUM8 the token savings can be pretty insane. Gobstopper saves 29% in tokens on Terminal-Bench 2.1 compared to Claude Code's built-in compaction. but in practice you can end up saving a lot more on long running sessions ask your agent to set it up: https://t.co/q6PBGsbUM8view source ↗
-
• twitter 1w positiveIntroducing Argil Director, the best video harness on the planet. They told you your story wasn't the kind that gets made. Too local. Too traditional. Too much like you. Describe it. Get your film with 80% less iteration. Try it today: https://t.co/tLpHOyw8od https://t.co/kHxlqr8Bpu Introducing Argil Director, the best video harness on the planet. They told you your story wasn't the kind that gets made. Too local. Too traditional. Too much like you. Describe it. Get your film with 80% less iteration. Try it today: https://t.co/tLpHOyw8od https://t.co/kHxlqr8Bpuview source ↗
-
• twitter 1w positivefrom a friend/beta user: > btw, like a week in, Gobstopper is absolutely kicking ass. I’m using the higher context scope for my orchestration agent, and then all workers on the default, and the token savings are insane and the quality of task completion is unchanged. https://t.co/CjfFX1akcS from a friend/beta user: > btw, like a week in, Gobstopper is absolutely kicking ass. I’m using the higher context scope for my orchestration agent, and then all workers on the default, and the token savings are insane and the quality of task completion is unchanged. https://t.co/CjfFX1akcSview source ↗
-
• twitter 1w neutralSynthID Detector is now available to everyone. 🌐 Check whether online content was generated using @GoogleAI, or with tools from our industry partners – including @OpenAI, @NVIDIA, Kakao and coming soon, @Apple. Try it out → https://t.co/cPW2aNnqr6 https://t.co/TGYA0uZ0dx SynthID Detector is now available to everyone. 🌐 Check whether online content was generated using @GoogleAI, or with tools from our industry partners – including @OpenAI, @NVIDIA, Kakao and coming soon, @Apple. Try it out → https://t.co/cPW2aNnqr6 https://t.co/TGYA0uZ0dxview source ↗
-
• twitter 1w neutralThe easiest way to play a game. https://t.co/MZrm63Q0Ew The easiest way to play a game. https://t.co/MZrm63Q0Ewview source ↗
-
• twitter 1w positiveRT @maks6361: @praggybuilds I've done one video already with @starter_story , I hope it'd be helpful for you https://t.co/5mBxoLnrBu RT @maks6361: @praggybuilds I've done one video already with @starter_story , I hope it'd be helpful for you https://t.co/5mBxoLnrBuview source ↗
-
• twitter 1w neutralBible Chat (~$800K/month) has been running this ad for 100 days. 9 seconds. A girl crying in her car: "Nobody will use your stupid Bible app." Then people adding Bible verses to their lockscreen. Complete breakdown of their ads & acquisition tactic below 👇 https://t.co/gCalLs7Zv3 Bible Chat (~$800K/month) has been running this ad for 100 days. 9 seconds. A girl crying in her car: "Nobody will use your stupid Bible app." Then people adding Bible verses to their lockscreen. Complete breakdown of their ads & acquisition tactic below 👇 https://t.co/gCalLs7Zv3view source ↗
-
• twitter 1w neutralyou and bro on Binance, just emotionally attached to different charts. https://t.co/C8xX3vO1Bc you and bro on Binance, just emotionally attached to different charts. https://t.co/C8xX3vO1Bcview source ↗
-
• twitter 1w positive> Be Dario Amodei > Born 1983 in San Francisco > Grow up in the Mission District > Start at Caltech, then earn your physics BS at Stanford > Go to Princeton for your MS & PhD → pivot to biophysics under Michael Berry > Finish your PhD in 2011 as a Hertz Fellow, with a thesis on measuring the collective behavior of neur… > Be Dario Amodei > Born 1983 in San Francisco > Grow up in the Mission District > Start at Caltech, then earn your physics BS at Stanford > Go to Princeton for your MS & PhD → pivot to biophysics under Michael Berry > Finish your PhD in 2011 as a Hertz Fellow, with a thesis on measuring the collective behavior of neural circuits > Do a postdoc at Stanford Medicine > Join Google Brain as a Senior Research Scientist > Join OpenAI in 2016 > Rise to VP of Research > Lead the work on GPT-2 and GPT-3 > Co-invent RLHF (training AI on human feedback) > Leave to build Anthropic > Co-found it in 2021 with your sister Daniela > Give your AI a written constitution for how it should behave > Testify before a U.S. Senate panel on AI risk in 2023 > Publish "Machines of Loving Grace" in 2024 (the optimistic case, from the guy known for warning about risk) > Call for outside evaluators embedded inside every frontier AI company > Now CEO of the company behind Claude He went from measuring neural circuits to running the company behind the model I build on every day. I committed to AI because I believed it would change how people build and work. His story is a good reminder of how far back that change goes.view source ↗
-
• twitter 1w neutralRyan says building "native" apps costs up to 3.5× in agent minutes. Bias or real? I tested it with Argent, @swmansion's toolkit that lets agents drive iOS & Android simulators in any framework 🤌 Check out the thread for details 🧵 https://t.co/mO9yxR19TX Ryan says building "native" apps costs up to 3.5× in agent minutes. Bias or real? I tested it with Argent, @swmansion's toolkit that lets agents drive iOS & Android simulators in any framework 🤌 Check out the thread for details 🧵 https://t.co/mO9yxR19TXview source ↗
-
• twitter 1w neutralsplitting actions interactions inspired by @nitishkmrk made with framer motion https://t.co/ouZs02Vhff splitting actions interactions inspired by @nitishkmrk made with framer motion https://t.co/ouZs02Vhffview source ↗
-
• twitter 1w positiveHey Builders & Founders, Looking to connect with people building in: 🚀 SaaS 🤖 AI 🌐 Web Apps 📱 Mobile Apps 💻 Tech 📊 Startups 🧠 Indie Hacking If you're building something exciting, I'd love to see it. Drop your product, website, or what you're working on below. 👇 Hey Builders & Founders, Looking to connect with people building in: 🚀 SaaS 🤖 AI 🌐 Web Apps 📱 Mobile Apps 💻 Tech 📊 Startups 🧠 Indie Hacking If you're building something exciting, I'd love to see it. Drop your product, website, or what you're working on below. 👇view source ↗
-
• twitter 1w neutralOne persistent Triton launch per token runs all four layers: int4 weights dequantized in streamed GEMVs, the KDA convolutions and state updates, MLA attention in the latent space with the key projection absorbed into the query, and the router plus routed and shared experts, held together by an atomic-counter barrier. C… One persistent Triton launch per token runs all four layers: int4 weights dequantized in streamed GEMVs, the KDA convolutions and state updates, MLA attention in the latent space with the key projection absorbed into the query, and the router plus routed and shared experts, held together by an atomic-counter barrier. Changing the input token or the weights in place during the audit changed the output, which still matched the reference. Board: https://t.co/MUNCbjpyAm Kernel: https://t.co/27iNbQipMb Trace: https://t.co/ejbqbwFH0zview source ↗
-
• twitter 1w neutralGPT-6.1 Sol [xhigh] on KernelBench-Mega: It wrote a one-launch Kimi-Linear Decode megakernel for RTX PRO 6000 at 5.0x the optimized PyTorch baseline, 17th of 32. GPT-6 Sol's megakernel ran at 0.96x. 6.1 Sol got there in 17 minutes; its score was flat from minute 7 and it called it done. Opus 5.5 leads at 35.5x after … GPT-6.1 Sol [xhigh] on KernelBench-Mega: It wrote a one-launch Kimi-Linear Decode megakernel for RTX PRO 6000 at 5.0x the optimized PyTorch baseline, 17th of 32. GPT-6 Sol's megakernel ran at 0.96x. 6.1 Sol got there in 17 minutes; its score was flat from minute 7 and it called it done. Opus 5.5 leads at 35.5x after 3.3 hours with the same plan. The gap is depth: no tensor cores, a full KV-cache copy every step, and phases that run on a single block.view source ↗
-
• twitter 1w neutralBizim bu aziz milletten başka hiç kimseye diyet borcumuz yok. Bizim Hazreti Allah’a bir can borcumuz var, Allah’ın izniyle hiçbir faniye borcumuz da eyvallahımız da yok. Bizim bu aziz milletten başka hiç kimseye diyet borcumuz yok. Bizim Hazreti Allah’a bir can borcumuz var, Allah’ın izniyle hiçbir faniye borcumuz da eyvallahımız da yok.view source ↗
-
• twitter 1w neutralGrid + MinGRU, 31% of the deck's 150M steps/s target, 11th of 17 (GPT-6 Sol: 64%). This is the one that went backwards. GPT-6 Sol replayed the rollout as a CUDA graph with cuBLAS for the gate matmuls; 6.1 Sol fused the rollout into one cooperative kernel with fp16 tensor-core MinGRU and a hi/lo split for near-fp32 accu… Grid + MinGRU, 31% of the deck's 150M steps/s target, 11th of 17 (GPT-6 Sol: 64%). This is the one that went backwards. GPT-6 Sol replayed the rollout as a CUDA graph with cuBLAS for the gate matmuls; 6.1 Sol fused the rollout into one cooperative kernel with fp16 tensor-core MinGRU and a hi/lo split for near-fp32 accuracy, plus a grid-wide barrier every step, and stopped after 41 minutes. Kernel: https://t.co/vt111L2j1N Trace: https://t.co/oiiMGPL8R5view source ↗
-
• twitter 1w neutralMegaQwen Decode, 3.9% of roofline, 15th of 20 (GPT-6 Sol: 4.0%). GPT-6 Sol launched a kernel per stage from a Python loop; 6.1 Sol runs every decode step and all four layers in one cooperative launch, sharing each KV load between the two query heads of a group. The launch count dropped, the score did not: the leaders a… MegaQwen Decode, 3.9% of roofline, 15th of 20 (GPT-6 Sol: 4.0%). GPT-6 Sol launched a kernel per stage from a Python loop; 6.1 Sol runs every decode step and all four layers in one cooperative launch, sharing each KV load between the two query heads of a group. The launch count dropped, the score did not: the leaders at 7 to 8% get weights and upcoming KV into L2 or shared memory before they are needed. Kernel: https://t.co/SssORwN1Qy Trace: https://t.co/lCmkkFQcLUview source ↗
-
• twitter 1w negativeGPT-6.1 Sol [xhigh] on KernelBench-CUDA: Four hand-written CUDA kernels passed their final checks on RTX PRO 6000. Native Sparse Attention is the jump: 0.28 ms, 9x faster than GPT-6 Sol and 5th of 14. The other three land mid-board. Where GPT-6 Sol leaned on cuBLAS and CUDA graphs, 6.1 Sol writes its own tensor-core … GPT-6.1 Sol [xhigh] on KernelBench-CUDA: Four hand-written CUDA kernels passed their final checks on RTX PRO 6000. Native Sparse Attention is the jump: 0.28 ms, 9x faster than GPT-6 Sol and 5th of 14. The other three land mid-board. Where GPT-6 Sol leaned on cuBLAS and CUDA graphs, 6.1 Sol writes its own tensor-core GEMMs and puts whole problems into one cooperative launch. It also stops early: no problem took it more than 80 minutes. One problem per reply below.view source ↗
-
• twitter 1w neutralA fresh start was just the tip of the iceberg. North of North Season 2 premieres November 5! https://t.co/se7cQuXJBO A fresh start was just the tip of the iceberg. North of North Season 2 premieres November 5! https://t.co/se7cQuXJBOview source ↗
-
• twitter 1w positiveMost people use Claude Code like a chatbot. That’s the mistake. Claude Code becomes far more powerful when you structure your project around: or project-wide instructions → .mcp.json for tool integrations → settings.json for permissions & hooks → Custom commands for repeatable workflows → Skills for task-specific ex… Most people use Claude Code like a chatbot. That’s the mistake. Claude Code becomes far more powerful when you structure your project around: or project-wide instructions → .mcp.json for tool integrations → settings.json for permissions & hooks → Custom commands for repeatable workflows → Skills for task-specific expertise → Agents for specialized work Eview source ↗
-
• twitter 1w neutralYou could literally: - Sign 5 clients at $5-10k/mo - Let AI do 80% of the delivery - Pay one VA $2k/mo to coordinate it - Sit on $25-50k/mo recurring - Keep 60% of it - Get paid again next month without signing anyone new Here's how: You could literally: - Sign 5 clients at $5-10k/mo - Let AI do 80% of the delivery - Pay one VA $2k/mo to coordinate it - Sit on $25-50k/mo recurring - Keep 60% of it - Get paid again next month without signing anyone new Here's how:view source ↗
-
• twitter 1w positiveMeet Harkirat Singh, a 2018 https://t.co/YSvOF5fxhn Computer Science graduate from IIT Roorkee. This guy secured a remote software developer role at Gather with a package of ₹4 crore per year. He has participated in Google Summer of Code twice and worked across FAANG companies, finance, and startups. He resigned from… Meet Harkirat Singh, a 2018 https://t.co/YSvOF5fxhn Computer Science graduate from IIT Roorkee. This guy secured a remote software developer role at Gather with a package of ₹4 crore per year. He has participated in Google Summer of Code twice and worked across FAANG companies, finance, and startups. He resigned from Goldman Sachs just before COVID to explore new opportunities. Through freelancing, he reportedly earned around ₹80 lakh per year and eventually secured a remote software developer role with a package of ₹4 crore per year. Here are some remote job platforms where companies pay in USD. Remotive - https://t.co/GroTHf9R6G Rocketship - https://t.co/fzeXO3JqV5 https://t.co/Wo5gHIZ6Gz - https://t.co/FOH3Qsuod9 Remote Woman - https://t.co/LX7elPEM96 Wellfound - https://t.co/6ODBv5KmDY We Work Remotely - https://t.co/r371QfB0rQ WorkWave - https://t.co/UHG3KUOYcY AI Jobs - https://t.co/TR9D30HGnJ Toptal - https://t.co/bEqqIaaSDQ FlexJobs - https://t.co/DSqcYjimAM JS Remotely - https://t.co/k9L95I1NLe Remote OK - https://t.co/pwiXrGmV8M FYI Check:- https://t.co/AQV7vQ7pDZview source ↗