Navanee (@navanee2u) profile picture
Joined April 2012
Photo by Navanee (@navanee2u)Photo by Navanee (@navanee2u)Photo by Navanee (@navanee2u)Photo by Navanee (@navanee2u)Photo by Navanee (@navanee2u)Photo by Navanee (@navanee2u)Photo by Navanee (@navanee2u)Photo by Navanee (@navanee2u)Photo by Navanee (@navanee2u)Photo by Navanee (@navanee2u)
Navanee retweeted
ICYMI - our recent launch of Z1
2
7
5
122
12,757
Meanwhile, in Bangkok, Thailand’s prime minister sang “Let It Be” and “Blowin’ in the Wind” as Singapore’s prime minister played guitar. #Thailand #Singapore
315
2,062
458
9,626
1,228,223
Fascinating how we are naturally intuit, classify and assign personalities - over-thinker, boring, condescending - to a string of matrix multiply operations :-) Deep down, is human personality also just a complex math function?
Qwen 3.8 27B in its default reasoning settings in LM Studio of "extra high" is a chronic over-thinker and I kind of love it
39
Humanoid robots that can barely grab a cup after 7th attempt. Meanwhile: specialized robots that have been dominating their specific task since 2008. ~~ ♻️ Join the weekly robotics newsletter, and never miss any news → ziegler.substack.com
13
22
175
15,514
$0.07 for M3, $3.39 for Opus. Both caught 13 of 17 bugs. Really interesting breakdown from @kilocode Definitely worth the read
We gave the same code audit to Claude Opus 4.8 and MiniMax M3. Same codebase. Same prompt. 17 known bugs planted in advance. MiniMax M3 caught 13 of them for $0.07. The cheapest Claude run caught the same 13 for $1.30. Here's the breakdown. 🧵
76
90
21
1,670
138,122
"Talk to a human" - Shouldn't we standardize on this and require that every AI Agent support this feature? Kinda like there is an unsubscribe section on every mass mailer. #aiagent #talktohuman #ai #chatbot
23
Fascinating.. It might seem tempting to focus on "AI" fears and see skynet here. But most of the improvements in Electronic, Mechanical, Software and AI subsystems is powered by human ingenuity. AI tools might have been used for some programming. But the real magic is hard work!
2025&2026 Chinese Spring Festival Gala
1
1
37
🚀 vLLM Talent Pool is Open! As LLM adoption accelerates, vLLM has become the mainstream inference engine used across major cloud providers (AWS, Google Cloud, Azure, Alibaba Cloud, ByteDance, Tencent, Baidu…) and leading model labs (DeepSeek, Moonshot, Qwen…). To meet the strong demand from top companies, the vLLM community is now collecting resumes year-round and helping with referrals (internships & full-time). If you have experience in any of the following areas, we’d love to hear from you: • RL frameworks & algorithms for LLMs • Tool calling, MCP, Harmony format, OpenAI/Anthropic API • Structured output / constraint decoding • High-performance kernels: attention, GEMM, sampling, sorting • CUTLASS / CUTE DSL / TileLang • Distributed systems: Ray, multiprocessing • vLLM + Kubernetes • Tensor / expert / context parallelism • NCCL, DeepEP, NVSHMEM, RDMA, NVLink • Prefill/Decode separation, KV-cache transport • Speculative decoding (Eagle, MTP, …) • MoE optimization • KV-cache memory management (hybrid models, prefix caching) • Multimodal inference (audio/image/video/text) • LoRA • Rust / Go / C++ / Python serving stacks • Attention mechanisms (MLA, MQA, SWA, linear attention) • Position encodings (RoPE, mRoPE) • Model architectures (DeepSeek, Qwen, etc.) • Embedding model support • torch.compile integration …or any other LLM inference engineering experience. Bonus points if you have: • Implemented core features in vLLM • Contributed to vLLM integrations (verl, OpenRLHF, Unsloth, LlamaFactory…) • Written widely-shared technical blogs on vLLM 💰 Compensation: Highly competitive, with no upper limit for exceptional inference engineers. 📍 Locations: Major cities in the US (SF Bay Area, etc.) Major cities in China (Beijing / Shanghai / Shenzhen / Guangzhou / Chengdu…) 📨 Apply: Send your resume to [email protected] (Sending your resume means you agree to share it with partner companies.) 🌱 Join the vLLM community: Slack: Apply at slack.vllm.ai Chinese community (WeChat): Add vllm_project with your name & affiliation Let’s build easy, fast, and cheap LLM serving for everyone — together! ⚡️
11
47
5
431
73,949
Anyone on X see a problem with this image output from Nano Banana Pro? Hint: see the final Da Vinci style self-portrait in the comic strip. :-) In the current political environment in the US, this kind of example could trigger a massive backlash. g.co/gemini/share/e6b7fbea59…
1
71
The technological breakthrough making this possible is stunning and the real world practical use cases are plenty. But the human element is likely trigger something like the classic "Uncanny Valley" chasm between reality and conspiracy. #ai #nanobanana #dystopian #genai #AIart
35
I crossed an interesting threshold yesterday, which I think many other mathematicians have been crossing recently as well. In the middle of trying to prove a result, I identified a statement that looked true and that would, if true, be useful to me. 1/3
61
295
98
2,466
894,791
Real data seems to point to GPT-5 as preferred model... The noise requesting older model seems a bit overblown. gptblindvoting.vercel.app/bl…
29
Click-bait title "You are going to die". But definitely worth reading and thinking about... Oliver Burkeman wants to remind his readers of their mortality, @HillaryKelly writes. How the writer became a self-help guru for self-help skeptics: theatlantic.com/magazine/arc…
15
Fascinating juxtaposition of two different viewpoints on the latest OpenAI o1 preview model... The jury is still out on this one.
33
Replying to @sciencegirl
63 seconds of fireworks in Japan - at Mount Fuji.
6
36
10
164
13,413
Another reason to quickly come up with standards to validate image, video & text content online... @pranaykotas @acorn "ChatGPT is making up fake Guardian articles. Here’s how we’re responding" | Chris Moran theguardian.com/commentisfre…
25
More specific recommendation - it is straightforward to mandate that every web page/online app/tool should provide a verifiable source-trail of images, videos & text content. Like the "Secure" HTTPS Certificate seen for all webpages. livemint.com/opinion/columns…
1
1
33
It'll be a slight distraction for the tech industry. It'll even create a mini market opportunity for some companies to deploy these technologies - like Y2k, RoHS compliance, GDPR compliance, etc... But in a few years' time, this can be accomplished.
1
1
17
Like @acorn says, some investment is needed in Public Policy to work out the full details and mandate a technology solution to this problem. This is a problem for all countries - US or Tanzania, Russia or Guatemala, China or India. Good project for countries to work together :-)
39