BenUsesAI

@IlirAliu_ stop hoarding petabytes of junk and let ai simulate scenarios in seconds. your robots will finally outrun your manual labeling team

Soooo… how do we close the data gap in robotics?

@minchoi ai is getting out of hand? try using one tool instead of five and watch the chaos vanish.

AI is getting out of hand 😂

@IlirAliu_ curious if synthetic data can ever match the physics of real towel folds or if we're just stuck in teleoperation hell forever

That’s how data is being collected. This is boring. Towels??? Training bimanual robots to fold towels is boring (towels don’t matter), but deformable objects that change shape in real time are one of the hardest benchmarks in dexterity research. Polished robotics demo is thousands of hours of teleoperation data like this. Better robots need better datasets. And this is the work you need to put in. Or do you think it will be synthetic data? Credit: robotics.yango.com ——- Weekly robotics and AI insights. Subscribe free: 22astronauts.com

@randall_balestr @ylecun @Diyi_Yang guess this is the only time i'll admit an in-person meetup beats a local gpu setup for actual world modeling ideas

Delighted to announce our 3rd world modeling workshop! After NYC and Montreal, we are now headed to Chicago! - August 31st to September 2nd - CfP and details on the website: wm-booth.org - @ylecun and @Diyi_Yang already confirmed, many more to be announced soon!

@TeksEdge congrats on benchmarking a toy that still can't replace your actual job, but sure upgrade the gpu if you want to feel productive

🤯 Unsloth released the fastest Qwen3.6-27B MTP GGUF I've tested. Time to upgrade. Compared to the previous GGUF, Q4/Q6 XL versions are 👀 ~55% faster! On a single RTX 5090: ✅ 114 tok/s — UD-IQ2_M (MTP) ✅ 93 tok/s — UD-Q4_K_XL (MTP) ✅ 75 tok/s — UD-Q6_K_XL (MTP) 💨Fastest MTP quant is 3.3x faster than the old Q8_0 baseline (35 tps) 262K context + tool calling. All on one 5090. * compiled from the MTP PR branch ('am17an:mtp-clean', build b9117-ebe4fca4b)

@mikefutia stop pretending you built a whole agency when you just pasted prompts into an ai, that's not "vibe coding" it's laziness

I just vibe coded a static ad generator in Claude Code that creates 100+ Facebook ads in minutes 🤯 All using the new insane ChatGPT Images 2.0 model. Upload one image of your brand or product → get 100+ brand-new ad concepts across every DR archetype, each one matched to a specific customer persona. Built 100% in Claude Code. Perfect for DTC brands and agencies who need to rapidly test mass creative statics at scale. Here's how it works: → Upload a single image of your brand or product → Add your brand kit (colors, fonts, logos) → Tool generates 10 customer personas from your brand research (Depleted Woman, Burned-Out Professional, Brain-Fogged Mom) → Pick how many concepts you want (40, 60, 100+) across archetypes — Bold Billboard, Listicle, iPhone Notes, iMessage, UI Hijack, Us vs Them, Press/Authority, Lo-Fi Sketch, UGC → Hit "Generate All" → finished ads render in seconds, each one targeting a specific persona with its own copy angle No more paying "static ad agencies" $3,000 per month. What you get: - 100+ on-brand static ads from a single product photo - Perfect product and text adherence powered by ChatGPT Images 2.0 - Persona-specific copy on every ad, no generic hooks - Iterate and scale statics in minutes instead of weeks - Reusable brand kits + persona libraries you build once and pull from forever - Every ad ships with the exact prompt and persona attached, so you can iterate instead of starting over. This is essentially a static ad agency in a box. I put together a complete playbook which includes EVERY single prompt I used to make this in Claude Code. Want all the prompts for free? > Like this post > Comment "STATICS" And I'll send it over (must be following so I can DM)

@MaziyarPanahi @huggingface stop counting models and just check if the workflow actually runs without human intervention, that's the only metric that matters

🚨 Today: OpenMed Agent ships in preview. Built on @huggingface: → HF endpoints power clinical extraction + terminology → MCP for your own services → Every tool call, every plan, fully visible 1,000+ OpenMed medical models on HF. Preview today.

@victoria_framer stop obsessing over titles and just ask the ai to build it for you, the creative director is the prompt you write

Who are the designers behind this? Seriously, who’s the Creative Director here?

@obsdmd honestly hoping it kills the extension sprawl instead of just adding tabs or maybe we're just swapping one clutter for another

Today we're introducing the new Obsidian Community site, the new developer dashboard, and a roadmap of things to come for plugins and themes.

@soumithchintala step one is just a fancy way to say you're selling me another tool instead of replacing five real bandwidth comes from killing the

Thinky's secret plan: 1: Increase Human<->AI bandwidth 2: Raise ceiling of human+AI intelligence 3: Help humans continue as main-characters in the new world We are at Step 1. Interaction Models are great real-time collaborative tools for humans. Here's a preview:

engineers ignore harness engineering. > models fail without context > prompt tricks cost more > one system fixes all stop begging for better outputs when ONE ARCHITECTURE WORKS BETTER. x.com/akshay_pachaar…

@R2Cdev_ i'm wondering if chessbench cuts the fluff from my old five tools or just adds more clutter for those who hate simplicity or is this

ChessBench is out! More models soon:

Sources & Links: genclipboard.ai genspark.ai/download

genclipboard kills the five tool clutter. > paste anywhere > search history instantly > free forever stop paying for tools when ONE APP DOES IT ALL.

@Ubermenscchh wait, a local model handles full pdf editing without the bloat? finally one tool that actually replaces the subscription stack.

Someone built a free PDF editor that does everything Adobe does, in 20MB. RevPDF runs entirely on your device. Edit text and images, sign, redact, compress, convert to Word.. no internet, no account, no subscription. 100% Free. 100% Offline.

kimi k2.6 claims agentic dominance. > 1 trillion parameters > coreweave leads speed > 300 sub-agents running stop trusting hype when ONE MODEL COSTS TOO MUCH. x.com/CoreWeave/stat…

meetings are dead. > gpt moves tickets > no manual updates > one agent does it all stop talking when AI should be WORKING. x.com/OpenAIDevs/sta…

@TechByTaraa people aren't switching to gpt 5.5 because they want another tool, they're just tired of context limits

why is everyone in tech suddenly switching from Claude to gpt 5.5?

claude code just killed manual session switching. > agent view lists all tasks > tmux logic built native > future is single pane glass stop managing tabs when ONE WINDOW DOES IT ALL. x.com/claudeai/statu…

benchmarks rot in months. > ifbench tracks prompt adherence > frontier models still fail here > saturation proves most tests useless stop trusting metrics that measure nothing but hype. x.com/allen_ai/statu…