Panca

best way to get started with hardware from scratch

Meta engineer lives with the absolute basics.... and he's getting backlash for it.

Clearly, Google Omni has been wildly underrated. Here it's turning a normal human hand into a live anatomy demo! Letting you see the muscles, tendons, and cartilage as if the skin were removed. Brilliant for educational purposes!

I built myself a chrome extension called Pose Any clothing model of any store becomes me. Each brand still conveys their brand aesthetic, but I can quickly understand how something would look on me.

Hardware engineers spend 80% of their time reading datasheets and 20% actually building. We're flipping that. Introducing blueprint.am, Claude Code but for Hardware.

煙のアノテーションとかマジ無理！ そう思う方はSAM3で「smoke」で検出してましょう。 カメラだけで新しい火災検知機とか作れると思います。 #SAM3 #ultralytics

🚨APPLE SPENT BILLIONS. ONE DEVELOPER USED $8. An $8 DIY project reportedly recreated a similar version of one of Apple Vision Pro’s most hyped AI features. Apple used an M2 chip, R1 processor, 12 cameras, LiDAR, and years of R&D. A Japanese developer used an ESP32, a tiny camera, and open-source MediaPipe. Open-source AI is collapsing the cost of innovation.

This Chinese engineer's AI-powered bin tracks your trash mid-throw and moves to catch it:

Excited to share our new work R³: 3D Reconstruction via Relative Regression. Only 372M params (~⅓ of recent 1B-class baselines), trained on 6×48G GPUs, but competitive on streaming reconstruction. Runs at 30+ FPS. Project: kevinxu02.github.io/r3-site/ Paper: arxiv.org/abs/2605.26519

Human in the loop 🤖 Made in @ComfyUI with a laundry list of tools: @ltx_model 2.3 + @thesystms FLW LoRA, @Alibaba_Wan 2.2 I2V + T2V, Florence2, @Meta Sapiens2, @OpenAI GPT Image 2.0, @suno , @AdobeAE

YC and a16z are closed. These are open right now: 1. @yzilabs Easy Residency S4 - 5 weeks online + 5 weeks in person, up to $500K. Web3, AI, Biotech. Deadline: June 21 2. @fdotinc Off Season II - 6 weeks in SF, 99 fellow student founders, up to $250K on demo day. Deadline: June 7 3. @alliance ALL18 - up to $500K, pitch @paradigm, @a16z, @foundersfund at Demo Day. Crypto, Web3, AI. Deadline: May 27 4. @solanalabs x @incubator Cohort 5 - 3 months in NYC starting September 2026. Web3. Deadline: June 5. 5. @XFounders_camp - rolling membership, Seed through Series A, AI, Fintech, Web3. Best programs don't always have the biggest names. P.S. If you're applying to any of these, your data room needs to be ready before they ask. Build yours today in under 10 minutes → @ThePageform

every day someone vibe codes a new app that has never existed before

From a single prompt to a playable arcade game. 👾 Live from the #GoogleIO Arcade, see how Antigravity 2.0 uses AI subagents to move from prompt to playable code on physical hardware instantly. Get all things AI here: goo.gle/3PMl8ga

だいぶ具合がよくなった もうコーディングしなくていいな そもそもプログラミング苦手だし #Codex #倒立振子

Omni feels like magic 🤩

SAM3の凄いところは「黒い犬」と指定すると黒い犬だけを検出できるところ。 #SAM3 #ultralytics

I Went From $3,000/Month on Claude to $5/Week on DeepSeek And honestly? 80% of my work is identical. For the past two months, I was burning $3-5K monthly on Claude Code. Every idea from design to development to testing - full end-to-end automation, even simulating users to test my products and provide feedback. Extremely token-intensive. But Claude's caching sucked, making it insanely expensive. Then I discovered DeepSeek V4. The numbers: • Claude: $5 input, $25 output per million tokens • DeepSeek: $0.28 input, <$1 output (with their current discount) • DeepSeek cached: $0.0002 - literally less than a penny The caching optimization is game-changing. Once DeepSeek has seen content, it basically stops charging tokens. My result: $5/week vs $1,000/week for the same workload. What works exactly the same: • UX modifications • Product development • Competitor research • Content writing • Code reviews Where Claude still wins: • Complex architectural decisions • Extremely nuanced problems But here's the thing - Claude has been getting dumber recently. It often says "done!" when it's clearly not done. Then apologizes but still doesn't finish the work. My current stack: • DeepSeek V4 Flash/Pro for 80% of daily work • Codex 5.5 for the hardest problems (more reliable execution) • Claude Code occasionally (because I already paid for it 🤷‍♂️) DeepSeek is also 3x faster than Claude. For tasks like "compare these repos" or "read this long document," DeepSeek finishes instantly while Claude takes 3+ minutes. Fun fact: I heard DeepSeek's speed comes from both optimization and gradually switching to Chinese chips (Huawei). If that's true, we might see even better performance later this year. Everyone's betting on Anthropic's rising valuation in secondary markets. But when 80% of daily dev work can be done faster, better, and cheaper by open-source models... Is Anthropic guaranteed to be the final winner? I think it's too early to call.

土地を数秒でスキャンし最も効率的な駐車場配置を計算してくれるAIが素敵

You should read this thread. It used to take about 25 seconds to generate a 5-second video on 8 Blackwell GPUs. The legends at @haoailab brought that down to just 4.2 seconds on a single Blackwell GPU… and then open sourced the tech behind it.

We are cooked. AI can fake crowd size now 💀 Nothing is real anymore.