AI Weekly Malaysia

Hugging Face and Cerebras have partnered to enable real‑time voice AI using the Gemma 4 language model, with Cerebras’s hardware providing low‑latency inference. This integration lets developers easily build voice assistants, transcription services, and conversational agents via Hugging Face’s ecosystem.

Simon Willison quotes Anthropic announcing that the US Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5, with access to be restored soon. This suggests builders in previously restricted regions may regain the ability to use these models.

Ahmad Osman argues on Latent Space that local (on-device) AI is rapidly catching up, from laptops and phones to enterprise infrastructure, enabling powerful models without cloud reliance.

The creator built a live benchmarking tool called 'How I AI Bench' using Claude Code, then ran five frontier models through 64 blind prototype generations, PRDs, and agent voice tests to review Anthropic's Sonnet 5. The results challenged common assumptions about model performance.

TechCrunch's Startup Battlefield Australia is accepting applications until July 6. Early-stage startups can pitch for visibility, funding, and media exposure. The event is based in Australia but open to regional founders.

How work changes along the exponential

Google released Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image), their cheapest and fastest image generation model. Simon Willison’s test showed decent visual quality but misspelled text, highlighting the trade-off between cost, speed, and precision.

OpenClaw, an open-source AI agent platform, has launched mobile apps for Android and iOS, enabling users to run autonomous agents directly from their phones. This brings agentic capabilities to a broader audience and opens new possibilities for mobile automation and experimental AI applications.

Claude Sonnet 5 released with near-Opus 4.8 performance at a lower API price, but uses a new tokenizer averaging 30% more tokens for English text—effectively a net price increase. Key API changes include a 128K output limit, 1M context window, removal of temperature/top-p/top-k, and adaptive thinking on by default.

Three former DeepMind researchers who built a poker-playing AI have launched EquiLibre Technologies, a Prague-based AI lab now valued at over $500 million. The company is applying imperfect-information game theory and reinforcement learning techniques, originally developed for poker, to generate profits for quantitative hedge funds.

Realta Fusion claims a fusion energy first by directly generating electricity from a plasma reaction, not just producing heat. The breakthrough moves fusion closer to practical power generation, though commercial viability remains years away.

Google has released Nano Banana 2 Lite, a faster and cheaper version of its AI image generator, aimed at making AI content creation more accessible for developers and creators.

Clicks has unveiled the final version of its BlackBerry-inspired Communicator phone in a hands-on video, featuring a physical keyboard and Android OS. It targets users nostalgic for tactile typing and a dedicated messaging experience.

IBM Research released ScarfBench, a benchmark for evaluating AI agents on enterprise Java framework migrations (e.g., Spring Boot to Quarkus). It tests multi-step, code-heavy refactoring tasks with real-world constraints like dependency management and build systems.

Etched, an AI chip startup, has reached a $5 billion valuation and secured $1 billion in sales contracts for its inference-focused chips, positioning itself as a competitor to Nvidia. This signals growing demand for specialized AI hardware beyond Nvidia’s GPUs, potentially lowering costs and increasing access for AI builders.

Anthropic released Claude Sonnet 5, a more affordable model with stronger agentic capabilities, positioning it as a cost-effective alternative to Opus, GPT-5.5, and Gemini Pro for running AI agents. The model emphasizes improved safety and lower pricing, making it accessible for startups and developers building autonomous workflows.

Acti launches a cross-app AI keyboard for iOS and Android that lets users invoke custom AI shortcuts directly from the keyboard, aiming to make AI agents accessible anywhere you type.

Simon Willison shares 'The AI Compass', a 29-question quiz that categorizes you into one of 30 AI ethics archetypes. It's a single-page React app built without a build step, using a script type='text/babel' trick.