Crazy ML (@crazyMLguy) header
Crazy ML (@crazyMLguy) profile picture

Machine Learning & Backend ๐Ÿ˜Ž Founder of Humanity backed by Investors from Google, SF and Hong Kong.

Mountains
Joined June 2023
๐— ๐—ฎ๐—ธ๐—ถ๐—ป๐—ด ๐—›๐˜‚๐—บ๐—ฎ๐—ป๐—ผ๐—ถ๐—ฑ ๐—ฅ๐—ผ๐—ฏ๐—ผ๐˜๐˜€ ๐—•๐—ฒ๐˜๐˜๐—ฒ๐—ฟ ๐—ฎ๐˜ ๐—Ÿ๐—ผ๐—ป๐—ด-๐—›๐—ผ๐—ฟ๐—ถ๐˜‡๐—ผ๐—ป ๐—ง๐—ฎ๐˜€๐—ธ๐˜€ What if we could make existing VLA models better at complex robot tasks without changing their weights? PGX does exactly that by combining: Predicate-based state machines for task progression and recovery. VLA models for manipulation skills. World-coordinate locking for spatial precision. Tested on a Unitree G1, PGX improved coffee-fetching success from 35% to 80% with unchanged GR00T model weights. A fascinating combination of AI foundation models and classical robotics control. Project: kl-research.github.io/pgx/ Paper: arxiv.org/abs/2611.26596 #Robotics #HumanoidRobots #VLA #EmbodiedAI
2
193
๐—ฆ๐—ฒ๐—น๐—ณ-๐—ฅ๐—ฒ๐˜„๐—ฟ๐—ถ๐˜๐—ถ๐—ป๐—ด: ๐—ช๐—ต๐—ฒ๐—ป ๐—”๐—œ ๐— ๐—ผ๐—ฑ๐—ฒ๐—น๐˜€ ๐—ฆ๐˜๐—ฎ๐—ฟ๐˜ ๐—˜๐—ฑ๐—ถ๐˜๐—ถ๐—ป๐—ด ๐—ง๐—ต๐—ฒ๐—ถ๐—ฟ ๐—ข๐˜„๐—ป ๐—ช๐—ฒ๐—ถ๐—ด๐—ต๐˜๐˜€ A 27B model can inspect, modify and improve its own neural network weights. No gradients. No retraining. Here's what happened: Self-improvement: Accuracy improved from 57.2% to 63.8% on 490 unseen problems. Self-repair: Recovered a damaged model from 7.3% to 70.2% accuracy across six image benchmarks. Self-testing: The model tests its modifications and keeps only improvements. The bigger implication? What if AI models could continuously improve themselves instead of waiting for another training run? Research: arxiv.org/abs/2611.21885 #AI #LLM #MachineLearning #SelfImprovement #Research
9
317
๐—›๐˜‚๐—ฎ๐˜„๐—ฒ๐—ถ ๐—œ๐—ป๐˜๐—ฟ๐—ผ๐—ฑ๐˜‚๐—ฐ๐—ฒ๐˜€ ๐—ข๐—ฆ๐—ง๐—” ๐—ข๐Ÿญ ๐—ฃ๐—ฟ๐—ผ: ๐—” ๐—ก๐—ฒ๐˜„ ๐—”๐—œ ๐—ฃ๐—ต๐—ผ๐—ป๐—ฒ ๐—•๐—ฟ๐—ฎ๐—ป๐—ฑ Huawei has unveiled the OSTA O1 Pro, its first HarmonyOS AI phone under the new OSTA brand. Key specifications: Chip: Kirin 9030S Battery: 7,000mAh Display: 6.39-inch 1.5K Design: 6.68mm thin, ~200g Memory: Up to 16GB RAM + 1TB storage Starting price: ยฅ5,699 (~$854) Richard Yu says OSTA aims to become a pioneering technology brand for the AI era. The smartphone battle is expanding beyond hardware to AI-powered experiences. #Huawei #OSTA #HarmonyOS #AI #Smartphones
2
162
๐—›๐˜‚๐—ฎ๐˜„๐—ฒ๐—ถ'๐˜€ ๐—ฉ๐—ถ๐˜€๐—ถ๐—ผ๐—ป ๐— ๐—ผ๐—ฑ๐—ฒ๐—น: ๐Ÿฒ.๐Ÿตร— ๐—™๐—ฎ๐˜€๐˜๐—ฒ๐—ฟ ๐—ฆ๐—ฝ๐—ฎ๐˜๐—ถ๐—ฎ๐—น ๐—š๐—ฟ๐—ผ๐˜‚๐—ป๐—ฑ๐—ถ๐—ป๐—ด Huawei Cloud is rethinking how vision models locate objects in images. Instead of generating bounding-box coordinates sequentially, its approach uses a single spatial grid token per instance and decodes boxes in parallel. The reported results are impressive: 87.6% fewer output tokens. Generation time drops from 1,532.4 ms to 223.6 ms. Achieves 27.7 boxes/sec. Beats LocateAnything by 1.7โ€“4.7 F1 points across six benchmarks. On Huawei Ascend, additional optimizations push performance further: 17.3% lower single-stream latency. 3ร— higher multi-stream throughput. No reported accuracy loss. The bigger takeaway: smarter decoding strategies can deliver major inference gains without simply scaling model size. #AI #ComputerVision #LLMInference #Huawei #GPU
2
162
๐—˜๐—™๐—˜-๐Ÿต๐Ÿด: ๐—” ๐—ฆ๐—ถ๐—บ๐˜‚๐—น๐—ฎ๐˜๐—ถ๐—ผ๐—ป ๐—•๐—ฒ๐—ป๐—ฐ๐—ต๐—บ๐—ฎ๐—ฟ๐—ธ ๐—ง๐—ต๐—ฎ๐˜ ๐—ฃ๐—ฟ๐—ฒ๐—ฑ๐—ถ๐—ฐ๐˜๐˜€ ๐—ฅ๐—ฒ๐—ฎ๐—น-๐—ช๐—ผ๐—ฟ๐—น๐—ฑ ๐—ฅ๐—ผ๐—ฏ๐—ผ๐˜ ๐—ฃ๐—ฒ๐—ฟ๐—ณ๐—ผ๐—ฟ๐—บ๐—ฎ๐—ป๐—ฐ๐—ฒ One of robotic's biggest challenges is knowing whether performance in simulation actually translates to real-world robots. Playground introduces EFE-98, a compact benchmark designed to bridge that gap. What's interesting? 0.98 correlation with real-robot performance rankings (as reported by Playground). Uses just 21.5% of the tasks in the full Embodied First Exam. Runs at roughly 10% of the cost. The idea is simple: evaluate robotic AI agents more cheaply without losing the signal that matters in the physical world. A step toward making real-world robotics evaluation faster, cheaper, and more scalable. Benchmark: withplayground.ai/leaderboarโ€ฆ #Robotics #EmbodiedAI #AI #Simulation #RobotLearning
2
154
๐—”๐—ป๐˜๐—ต๐—ฟ๐—ผ๐—ฝ๐—ถ๐—ฐ'๐˜€ ๐—œ๐—ฃ๐—ข: ๐—ฆ๐—ฎ๐—ณ๐—ฒ๐˜๐˜†, ๐—ฆ๐—ฐ๐—ถ๐—ฒ๐—ป๐—ฐ๐—ฒ & ๐—ง๐—ต๐—ฒ ๐—•๐—ฎ๐˜๐˜๐—น๐—ฒ ๐—™๐—ผ๐—ฟ ๐—ง๐—ต๐—ฒ ๐—”๐—œ ๐—ก๐—ฎ๐—ฟ๐—ฟ๐—ฎ๐˜๐—ถ๐˜ƒ๐—ฒ On the day of Anthropic's historic $81.3B IPO, Dario Amodei delivered a striking message about AI's future. He revealed that Anthropic's biology team had discovered what appears to be a new gene-editing mechanism, potentially opening the door to dozens of discoveries every month. But what caught my attention was his statement: "We haven't yet earned the right to talk about how great those things are." There's an interesting contrast here: Ambition: AI could accelerate scientific and medical discoveries dramatically. Restraint: Anthropic emphasizes safety, responsibility and human benefit. Positioning: The company is building its identity around delivering transformative AI while presenting safety as a central priority. Whether genuine restraint, strategic positioning, or both, one thing is clear: The battle for the future of AI is also becoming a battle over who gets to define its purpose. Interview: ceointerviews.ai/interview/3โ€ฆ #Anthropic #DarioAmodei #Claude #AI #AISafety
1
184
๐—จ.๐—ฆ. ๐—ฅ๐—ฒ๐˜ƒ๐—ฒ๐—ฎ๐—น๐˜€ ๐Ÿด๐Ÿณ๐Ÿต,๐Ÿฌ๐Ÿฌ๐Ÿฌ+ ๐—”๐—œ ๐—–๐—ต๐—ถ๐—ฝ ๐—˜๐˜…๐—ฝ๐—ผ๐—ฟ๐˜ ๐—”๐—ฝ๐—ฝ๐—ฟ๐—ผ๐˜ƒ๐—ฎ๐—น๐˜€ ๐—ณ๐—ผ๐—ฟ ๐—–๐—ต๐—ถ๐—ป๐—ฎ The U.S. House Foreign Affairs Committee has publicly released export-license data covering NVIDIA and AMD AI chips destined for China. The findings reveal: 3.4 million AI chips requested for export. 879,000+ chip exports approved by the U.S. Bureau of Industry and Security (BIS). Approvals included exports involving companies such as Alibaba and Tencent. The disclosures cover a critical period of U.S. semiconductor export policy and raise questions about national security, technology access, and enforcement. One important distinction: approved licenses do not necessarily mean all those chips were actually delivered. As AI becomes a strategic resource, control over compute is becoming just as important as control over the technology itself. Source: foreignaffairs.house.gov/newโ€ฆ #AI #NVIDIA #AMD #Semiconductors #Geopolitics
1
132
๐—™๐—น๐—ฎ๐˜€๐—ต๐—œ๐—ป๐—ณ๐—ฒ๐—ฟ: ๐—ง๐—ต๐—ฒ ๐—š๐—ฃ๐—จ ๐—ž๐—ฒ๐—ฟ๐—ป๐—ฒ๐—น ๐—Ÿ๐—ฎ๐˜†๐—ฒ๐—ฟ ๐—•๐—ฒ๐—ต๐—ถ๐—ป๐—ฑ ๐—›๐—ถ๐—ด๐—ต-๐—ฃ๐—ฒ๐—ฟ๐—ณ๐—ผ๐—ฟ๐—บ๐—ฎ๐—ป๐—ฐ๐—ฒ ๐—Ÿ๐—Ÿ๐—  ๐—œ๐—ป๐—ณ๐—ฒ๐—ฟ๐—ฒ๐—ป๐—ฐ๐—ฒ Fast LLM serving isn't just about powerful GPUs. It's about optimizing attention, memory movement, and compute at the kernel level. FlashInfer is an open-source (Apache 2.0) GPU kernel library and generator used by vLLM, SGLang, TensorRT-LLM, and TGI. What makes it interesting for GPU and systems engineers? โ€ข Unified Kernel Backend: One API for Attention, GEMM, and MoE, with backend selection across FlashAttention, cuDNN, CUTLASS, and CuTe DSL. โ€ข Advanced Attention & KV Cache: Paged/Ragged KV caches, Cascade Attention, fused prefill + decode, Sparse Attention, and DeepSeek MLA. โ€ข Low-Precision Compute: Optimized FP8, NVFP4, and MXFP4 kernels, alongside fused MoE routing and distributed communication. โ€ข Fast Decoding: Optimized Top-K/Top-P sampling and speculative decoding, with GPU support spanning Turing to Hopper, Blackwell, and Rubin. A great example of how low-level GPU engineering translates into faster, more efficient AI infrastructure. Explore: GitHub: github.com/flashinfer-ai/flaโ€ฆ Website: flashinfer.ai #AI #CUDA #LLMInference #GPU #OpenSource
1
2
226
๐——๐—ถ๐—ฑ ๐—–๐˜‚๐—ฟ๐˜€๐—ผ๐—ฟ ๐—ฟ๐—ฒ๐—ฏ๐—ฟ๐—ฎ๐—ป๐—ฑ ๐—ž๐—ถ๐—บ๐—ถ ๐—ž๐Ÿฎ.๐Ÿฑ ๐—ฎ๐˜€ ๐—–๐—ผ๐—บ๐—ฝ๐—ผ๐˜€๐—ฒ๐—ฟ ๐Ÿฎ? Moonshot AI's Yulun Du claims that: Composer 2 uses the same tokenizer as Kimi. An API request reportedly exposed a Kimi K2.5 model identifier. Moonshot says Cursor neither obtained permission nor paid licensing fees.
2
140
๐—”๐—œ ๐˜€๐—ผ๐—น๐˜ƒ๐—ฒ ๐—ฎ ๐Ÿญ๐Ÿฑ๐Ÿฌ-๐˜†๐—ฒ๐—ฎ๐—ฟ-๐—ผ๐—น๐—ฑ ๐—บ๐—ฎ๐˜๐—ต๐—ฒ๐—บ๐—ฎ๐˜๐—ถ๐—ฐ๐—ฎ๐—น Problem? A mathematician claims to have discovered a proof of the Riemann Hypothesis, one of mathematics' biggest unsolved problems, working alongside Claude. What's fascinating is how they worked together: โ€ข Human intuition guided the direction. โ€ข Claude connected complex mathematical ideas and helped crack crucial steps. โ€ข The proof was formalized using Lean, a mathematical verification system. The proof still needs independent scrutiny, but the possibilities are incredible. Original note: alpo.ge/remark.pdf
2
172