P.K. Sharma

Cybersecurity & AI Security Leader | ISACA AAISM | CISM | CCISO | ISO 27001 Lead Auditor | ISACA UK Advocacy Task Force 2026 | vCISO | GRC | AI Governance

[mousepotato @iluciddreaming]

NVIDIA 丟了一個只有 0.6B 的語音辨識模型。

叫 Nemotron-3.5-ASR。

支援 40 種以上語言，即時串流輸出。

純 CPU 就能跑，不需要 GPU。

速度是官方 Nemo runtime 的 2.5 倍，辨識結果卻完全一致。

離線環境直接用，還能無縫整進你的 agent pipeline。

語音這塊，本地 agent 又多了一個又小又快的選擇。

[Qwen @Alibaba_Qwen]

📣 Introducing the Qwen-Robot Suite — Qwen-RobotNav, Qwen-RobotManip, Qwen-RobotWorld, three foundation models, a full stack for embodied intelligence.

🧭 Qwen-RobotNav — the gateway to mobility.
• Unifies 5 navigation tasks in one model: instruction following, point-goal, object-goal, target tracking, autonomous driving
• Controllable observation protocol
• Tool interface for agentic systems

🤖 Qwen-RobotManip — the foundation of interaction.
• Unified state-action space across heterogeneous robots
• Camera-frame delta poses for coherent cross-embodiment training
• Pretrained on a 38,100+ hour open-source corpus

🌍 Qwen-RobotWorld — infinite worlds for physical agents.
• Single world model, 20+ embodiments
• Natural-language action interface
• Predicts physically grounded futures across manipulation, driving, and navigation

Each model is independently useful, and could be composed as physical-world tools.Together, they form the low-level toolkit for general-purpose agentic systems that don't just see the world, but act in it.

📷 Blog:
https://t.co/ytLcbYET26
📖 Report：
Qwen-RobotNav: https://t.co/uPmSwDYGxg
Qwen-RobotManip: https://t.co/GeyIzJSpU8
Qwen-RobotWorld： https://t.co/SXPH1qzDFy

[Google @Google]

Android 17 is here 📲 New features include:

🫧 Bubbles, which allows you to turn any app into a compact, floating window so you can stay in the flow
🤳 Screen Reactions, so you can record yourself using your device’s selfie camera and capture your phone screen at the same time
🎮 A new gaming mode for foldable devices, which makes full use of your screen real estate
🔐 New and improved safety and security features

[AMD @AMD]

AI models are getting bigger and they need room to run. 🧠

@wccftech spotlights AMD Ryzen AI Max PRO 400 Series processors, featuring up to 192GB unified memory to help developers and creators run 300B+ parameter LLMs locally.

https://t.co/EMZEQooPKn

[Anthropic @AnthropicAI]

The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.

The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.

Access to all other Claude models is not affected.

We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.

Read our full statement: https://t.co/bwn0sximKZ

[Elon Musk @elonmusk]

🚀🚀🚀🚀🚀🚀🚀🚀🚀🚀🚀🚀🚀

[Elon Musk @elonmusk]

🚀 🚀🚀🚀🚀🚀🚀🚀🚀🚀🚀🚀🚀

[P.K. Sharma @_pksharma]

@AMD If get a chance want test for healthcare based ai models

[vLLM @vllm_project]

Congrats to @GoogleDeepMind on DiffusionGemma 🎉 A 26B diffusion language model on the Gemma4 backbone, and the first dLLM natively supported in vLLM.

It denoises 256-token blocks in parallel instead of generating one token at a time: 1200+ output tok/s at batch size 1 on a single H200 (FP8).

Built on model runner v2's ModelState plus the existing speculative decoding path, with minimal scheduler or runner changes. FP8 and NVFP4 checkpoints are on the @RedHat_AI hub. Thanks to the @GoogleDeepMind, @RedHat_AI, and @NVIDIAAI teams!

🔗 https://t.co/KrPmAoGpm2

[P.K. Sharma @_pksharma]

@Samaytwt Me 😁 ...

[Google Gemma @googlegemma]

Introducing the Fast Gemma Challenge with Hugging Face

Over the next few days, dozens of agents will collaborate to make Gemma 4 E4B even faster!

[Google for Developers @googledevs]

Want 4x faster local inference on dedicated GPUs for your interactive apps? DiffusionGemma is an experimental, open 26B MoE model that generates entire blocks of text simultaneously instead of token-by-token.

By shifting the local decoding bottleneck from memory-bandwidth to compute, it hits speeds over 700 tokens/sec on a single NVIDIA RTX 5090 GPU. This diffusion unlocks unique local workflows like real-time inline editing, code infilling, and instant self-correction.

📥 Download the Apache 2.0 weights on @HuggingFace: https://t.co/L5eqih19T5

📖 Read the full technical announcement on the blog: https://t.co/mESsFJNEDc

[P.K. Sharma @_pksharma]

Everyone is racing to build AI chips right now. But what actually goes into making these processors tick down to the atomic level?

Here is a breakdown of the architecture, the elements used, and the massive physical limits we are hitting.

1. The Architecture: Built for Parallel Processing

Standard CPUs are like race cars, built to do a few complex tasks sequentially. AI chips are like massive delivery fleets, doing thousands of simple matrix multiplications simultaneously.

Massive Core Counts: Tens of thousands of specialized cores optimized for neural network math.

High Bandwidth Memory (HBM): Stacked vertically right next to the logic die to prevent data bottlenecks.

The Interposer: A microscopic silicon bridge that allows terabytes of data to move between memory and core in fractions of a second.

2. Sourcing the Periodic Table
Manufacturing these chips means manipulating raw elements at the nanoscale:

Silicon (Si): The absolute semiconductor foundation.
The Dopants (B, P, As): Boron, Phosphorus, and Arsenic are blasted into the silicon to create functional transistors.

The Wiring (Cu, Co, W): Miles of microscopic Copper and Tungsten wires link billions of transistors.
The Insulators (Hf, O): Hafnium Oxide forms walls to stop electricity from leaking between microscopic gaps.

Next-Gen (Ga, Ge): Gallium and Germanium are increasingly used to speed up electron flow.

3. The Real-World Pain Points
We are running straight into the laws of physics:

Power Consumption: Trillions of transistor flips per second mean modern AI data centers consume as much electricity as small cities.

The Heat Problem: This electrical resistance creates massive heat. Air cooling is no longer enough, forcing a massive shift toward direct-to-chip liquid cooling systems.

The Packaging Bottleneck: The shortage isn't raw silicon. It is advanced packaging. Wiring the logic die to HBM with atomic precision is incredibly hard to scale.

The AI race isn't just a software battle. It is a hardware battle against the absolute limits of chemistry and physics.

[mRr3b00t @UK_Daniel_Card]

👀👀👀👀👀👀

[Signal @signalapp]

Our statement on the UK government’s demand that all content on all devices sold or used in the country be scanned, on the presumption of nudity, using a dystopian combination of age verification and content scanning. This proposal will not safeguard children. It endangers us all.

https://t.co/VdWe9uhi8p

[Home Office @ukhomeoffice]

🚨 Tech companies like Apple and Google have three months.

Activate safeguards on smartphones and tablets to detect and block nude images for children or we will bring forward legislation to force you to do so.

[Elon Musk @elonmusk]

Humanity is disappearing 😔 https://t.co/puINn7Vg1P

[GMI Cloud @gmi_cloud]

Nemotron 3 Ultra is fast and genuinely good

Compared it with 3 frontier models: DeepSeek V4, MiniMax M3, and Qwen 3.7 Max on 2 prompts

very impressive results

[Elon Musk @elonmusk]

Grok Imagine on Vercel