Philipp Kandal (@apphil) 's Twitter Profile
Philipp Kandal

@apphil

cpo @ grab, before: engineering vp @ telenav, founder skobbler (sold to telenav).

ID: 6421402

linkhttp://www.kandal.com calendar_today29-05-2007 18:08:26

15,15K Tweet

24,24K Followers

419 Following

Philipp Kandal (@apphil) 's Twitter Profile Photo

Similiar thing that happened to @levelsio happened to me before. Great example of the same principle found in self-driving cars: When autonomous systems (vehicles & software agents) “almost” work perfectly, humans lower their guard—allowing sneaky errors to slip through (e.g.,

Philipp Kandal (@apphil) 's Twitter Profile Photo

Strong open weights vision language model from bytedance. Love the transparent paper - and showing gemini 2.5 pro as benchmark to beat- great work

Deedy (@deedydas) 's Twitter Profile Photo

Google's AI just made math discoveries NO human has! —Solved optimal packing of 11 and 12 hexagons in hexagons. —Reduced 4x4 matrix multiplication from 49 operations to 48 (first advance in 56 years!) and many more. AlphaEvolve is the AlphaGo 'move 37' moment for math. Insane.

Google's AI just made math discoveries NO human has!

—Solved optimal packing of 11 and 12 hexagons in hexagons.
—Reduced 4x4 matrix multiplication from 49 operations to 48 (first advance in 56 years!)
and many more.

AlphaEvolve is the AlphaGo 'move 37' moment for math. Insane.
Philipp Kandal (@apphil) 's Twitter Profile Photo

The Way of Code is a beautiful fusion of Rick Rubin’s creative philosophy and Anthropic’s AI vibes. A must-see for anybody who enjoys coding as art. wayofcode.com

Philipp Kandal (@apphil) 's Twitter Profile Photo

🚀 Google just released the final Gemini 2.5 Pro Preview (before general availability) - another 24 point jump on lmarena and 35 points on webdev arena. Love the ~monthly update cadence / improvements blog.google/products/gemin…

Philipp Kandal (@apphil) 's Twitter Profile Photo

Super-intelligence unlocked: Asked Gemini 2.5 Pro for deep research. After 5 minutes & 100+ sites crawled it sighed: “I’m having a hard time.”- it’s now smart enough to call in sick.

Super-intelligence unlocked: Asked Gemini 2.5 Pro for deep research. After 5 minutes & 100+ sites crawled it sighed: “I’m having a hard time.”- it’s now smart enough to call in sick.
Philipp Kandal (@apphil) 's Twitter Profile Photo

Sam Altman just casually hinted at a breakthrough: a new method at OpenAI that drives regular cars better than existing autonomous tech. The autonomous race just heated up! 🔥 (~5:50 min) youtu.be/mZUG0pr5hBo?si…

Philipp Kandal (@apphil) 's Twitter Profile Photo

🚀 Grok 4 is Crushing the AI Frontier!🚀 Elon Musk's massive bet on compute scaling is paying off BIG TIME. Am excited to see what xAI builds next!🔥

Philipp Kandal (@apphil) 's Twitter Profile Photo

🚀 Breaking: Moonshot AI just dropped Kimi K2 – an insanely strong open-source LLM with 1T params (32B active)! Giving major DeepSeek vibes with killer evals. Can’t wait for reasoning upgrades! Who’s ready to test this beast?

Philipp Kandal (@apphil) 's Twitter Profile Photo

Sunday musing: Exponential tech growth still blows my mind! In camera/map hardware research: Image compression shrunk -9.8% YoY (6.6MB 8K in 2015 HEVC to 3.7MB VVC). Video: -15% YoY (4.5GB/hr HD H.264 to 1.24GB VVC). SSD costs down 8x ($0.40/GB to $0.05). Result? 20x more

Sunday musing: Exponential tech growth still blows my mind! In camera/map hardware research: Image compression shrunk -9.8% YoY (6.6MB 8K in 2015 HEVC to 3.7MB VVC). Video: -15% YoY (4.5GB/hr HD H.264 to 1.24GB VVC). SSD costs down 8x ($0.40/GB to $0.05). Result? 20x more
Philipp Kandal (@apphil) 's Twitter Profile Photo

Impressive work from Tencent for 3D World generation. Thanks for open sourcing it! Would be super cool paired with the Apple Vision Pro to generate worlds dynamically and explore them

Philipp Kandal (@apphil) 's Twitter Profile Photo

Qwen is really on a roll lately. Image generation model looks amazing with text generation and ultra high accuracy rendering. Thanks for releasing it under an open license!

Philipp Kandal (@apphil) 's Twitter Profile Photo

Meituan just dropped a new LLM with “Dynamic Activation.” Think of it as a brain that decides when to think harder: it activates more experts for tricky parts of a question, fewer for easy ones. Closer and closer to how our own brains allocate effort. 🧠⚡

Philipp Kandal (@apphil) 's Twitter Profile Photo

Just in time for Golden Week: Tencent’s HunyuanImage 3.0 is open‑source. 80B params MoE, state of the art performance level. Massive kudos!