Muhammad Bilal Waris

Breaking sports news, cricket first 🏏 plus AI & big tech updates. Fast, accurate, daily.

[Muhammad Bilal Waris @engrbilalwaris]

BREAKING: Jensen Huang has resigned as Nvidia CEO, president and board member, effective immediately. 🧵

CFO Colette Kress is interim CEO. What happened, in 4 posts. #Nvidia #AI

[Muhammad Bilal Waris @engrbilalwaris]

AI on the recycling belt. NVIDIA reports 300+ Greyparrot Analyzer systems across more than 20 countries, using computer vision to identify waste by material and brand. The useful part: data operators can act on, not just object counts. https://t.co/lFP7Zbk1dE

[Muhammad Bilal Waris @engrbilalwaris]

A polished robot demo is not a reliability test. DexBench, developed by RLWRLD with NVIDIA, evaluates dexterity across 15 tasks in four industry scenes, varying objects, space and physics. Simulation evidence, not deployment proof. https://t.co/kbhibsAa8P

[Muhammad Bilal Waris @engrbilalwaris]

Security checks without a deployment. Vercel Security Review now scans GitHub code changes without a linked Vercel project. Default coverage includes all unlinked repos the Vercel GitHub App can access. Check scope in Agent settings. https://t.co/NskWmHuQ6J

[Muhammad Bilal Waris @engrbilalwaris]

Disorder can be a design advantage. MIT and Northeastern researchers 3D-printed network-inspired metamaterials that resisted impacts better than ordered designs in tests. Promising for cushioning, not a product launch. https://t.co/YXtut9rcgu

[Muhammad Bilal Waris @engrbilalwaris]

Same SGLang workflow, another hardware target. AWS says SGLang-Neuron is publicly available via a Deep Learning Container, bringing model serving and RL rollouts to Trainium through native PyTorch. Check supported models and configurations. https://t.co/D6IQgUuhWQ

[Muhammad Bilal Waris @engrbilalwaris]

Smarter memory for coding agents: Microsoft's Nov 12 roundup highlights APEC, which prepares summaries in the background. Evaluations reported up to 25.4% lower cost and 9.6% less execution time. Research, not a product rollout. https://t.co/aYAiexLTIc

[Muhammad Bilal Waris @engrbilalwaris]

Enterprise Server joins the Copilot app. GitHub says GHES support is now GA in app 1.2.0+, covering repositories, issues and pull requests. Important limit: cloud sessions are not available for GHES. https://t.co/Qf0BUn2nlq

[Muhammad Bilal Waris @engrbilalwaris]

Less distance, faster sandbox starts. Vercel reports up to 4x faster initialization outside iad1 when image data comes from a regional cache. Gains vary by region and cache hits. A useful reminder: data locality matters. https://t.co/fZpGBez9gZ

[Muhammad Bilal Waris @engrbilalwaris]

IBM released experimental Granite 4.1 3B and 8B previews trained for LLMON, a format that explicitly separates instructions, data and requirements.

Interesting direction for clearer AI context, not a security guarantee.

https://t.co/RkJd5JaiMh

[Muhammad Bilal Waris @engrbilalwaris]

Less repeated build work: Vercel raised the build cache limit to 5 GB on all build machines.

The change applies automatically at no extra charge. More dependencies and framework files can be reused between deployments.

https://t.co/yiXPne13rB

[Muhammad Bilal Waris @engrbilalwaris]

Cloudflare Browser Run Quick Actions now work locally with Chrome via wrangler dev.

Caveat: JSON returns schema-shaped placeholders, not real AI extraction. Kitesurf isn't supported locally.

https://t.co/588RnKOeQm

[Muhammad Bilal Waris @engrbilalwaris]

The AI investment story is not just model labs.

LangChain raised $300M at a $3B valuation in a CapitalG-led round, WSJ reports. My take: tools for building and operating agents are becoming a major battleground too.

https://t.co/efTht6PRnQ

[Muhammad Bilal Waris @engrbilalwaris]

Available now is not the same as default.

Vercel Sandbox now offers a Python 3.15 managed image. Its universal default still uses 3.14, with a switch scheduled for Dec 16. A useful window to test compatibility before that change.

https://t.co/h87UZDlBry

[Muhammad Bilal Waris @engrbilalwaris]

Better weather AI needs better rainfall data.

Google’s new PARDIG research provides open global precipitation estimates at 15-minute, roughly 10 km resolution. Experimental data, not official weather alerts.

Paper: https://t.co/HwvHcX1wsq

[Muhammad Bilal Waris @engrbilalwaris]

Web-grounded answers need inspectable evidence.

Parallel Decisions Fast is now on Vercel AI Gateway, free through Nov 23. It returns probabilities and sources. Important: citations cover the whole request, not each answer.

https://t.co/Oik2VoUzsw

[Muhammad Bilal Waris @engrbilalwaris]

One architecture for sand, fluids and cloth?

MIT highlighted WorldParticle this week: a particle-based AI simulator tested across six physical domains. Promising research toward shared simulation tools, not a universal replacement for physics solvers.

https://t.co/vWQ3SzSzjv

[Muhammad Bilal Waris @engrbilalwaris]

AI replies do not have to be walls of text.

Google released A2UI 1.0, a stable open protocol for agents to describe interactive screens that apps render with native components. Check renderer versions: Jetpack Compose 1.0 is still coming.

https://t.co/qwHhi3gPZI

[Muhammad Bilal Waris @engrbilalwaris]

@CloudflareDev A useful shared deliverable would be reproducible benchmarks across the consortium’s hardware: same model, precision, prompt lengths and concurrency, with tail latency alongside throughput. That would help teams compare deployment trade-offs rather than headline tokens/sec.

[Muhammad Bilal Waris @engrbilalwaris]

Quantum computing has a wiring problem, too.

MIT Lincoln Laboratory’s Quantum Microsystems platform combines superconducting and semiconductor chips at wafer scale, with replaceable chiplets. An integration advance, not a quantum-advantage claim.

https://t.co/3XwX2CLUmT

[Muhammad Bilal Waris @engrbilalwaris]

Better climate AI may start with better training data.

MIT researchers optimized synthetic emissions scenarios to help climate emulators handle unfamiliar conditions with less data. Full-scale Earth-system testing is still ahead.

https://t.co/orB9cICxdh