The wait is over for Google's power users. After weeks of teasing, the company has officially rolled out Gemini 3 Deep Think to Google AI Ultra subscribers. This isn't just a speed bump; it's a fundamental shift in how the model processes information, moving from rapid-fire token prediction to a deliberate, iterative "System 2" reasoning process designed to tackle complex math, science, and logic problems.

Unlike standard models that rush to a linear conclusion, Deep Think leverages advanced parallel reasoning to explore multiple hypotheses simultaneously. Think of it as a team of experts debating a problem in a conference room rather than a single intern guessing the answer.

This "iterative rounds of reasoning" approach allows the model to self-correct and refine its logic before presenting a final answer, reducing the hallucination rate on multi-step tasks that usually trip up LLMs.

The performance metrics are turning heads. Google claims Deep Think achieves a 41% score on "Humanity's Last Exam", a notoriously difficult benchmark designed to be "un-googleable," without using any external tools.

Even more impressive is its performance on the ARC-AGI-2 visual reasoning test, where it scored an unprecedented 45.1% with code execution. These numbers suggest Google has successfully closed the "reasoning gap" that OpenAI opened with its o1-preview models earlier this year.

The "chat" era is evolving into the "thought" era. By gating this capability behind the $249.99 per month Ultra tier, Google is betting that professionals and researchers will pay a premium for an AI that takes its time to be right, rather than just being fast.

Amazon Web Services (AWS) is tired of waiting in line for GPUs. In a major strategic pivot, the cloud giant has rushed out its new Trainium3 chip, a piece of custom silicon designed explicitly to break Nvidia's monopoly on AI training. This isn't just an iterative update; it's a declaration of independence for the AWS ecosystem.

The specs are turning heads. Trainium3 promises to be significantly more power-efficient than Nvidia's current H200 lineup while offering comparable performance for specific workloads. TechCrunch notes that Amazon is teasing an "Nvidia-friendly roadmap," suggesting a hybrid approach where customers can mix and match chips, but the underlying message is clear: you don't need Nvidia for everything anymore.

Amazon is playing the long game. By controlling the entire stack, from the data center metal to the Trainium3 silicon and the software layer, they can offer lower prices to startups desperate to cut compute costs. This is a direct shot at Google's TPUs and Microsoft's Azure Maia chips. CryptoRank highlights that this competition could finally start to deflate the "GPU bubble" that has kept AI training costs astronomically high.

Can Amazon convince developers to switch? Nvidia's CUDA software moat is deep and wide. Trainium3's success won't depend on its FLOPS, but on whether AWS can make migration painless. If they succeed, the era of Nvidia's absolute dominance might be showing its first cracks.

🗣️ Anthropic releases bot to interview users on work habits

🧸 Algorithmic playmates reshape how the next generation grows up

🖳 Grok 4.20 destroys others LLMs in Crypto trading

🤝 Google integrates Gemini into Replit for faster app creation

🐠 Softbank CEO say humans intellectually inferior to AI by 2035

🗞️ Meta partners with multiple news publishers to use content

Here’s what to do:

🖍️ Step 1: Sign up for the Runway Gen-4.5 trial. Think of a simple, atmospheric scene (e.g., "A cyberpunk detective walking in rain," or "A coffee cup shattering in slow motion").

📝 Step 2: Write a prompt that focuses on movement. Instead of just "a cat," try "A cinematic close-up of a cat leaping across a gap, fur rippling in the wind, 4k, shallow depth of field."

🎥 Step 3: Use the "Camera Control" feature. Set a specific camera move, like a "Truck Left" or "Zoom In," to give your shot intention.

🎞️ Step 4: Generate the clip. If the physics look off, tweak the "Motion Score" slider and try again.

Share your best 4-second masterpiece with us!

From reality-bending AGI-esque LLMs to chips that challenge the status quo, the landscape is shifting fast. Keep experimenting, keep questioning, and remember: the future won’t write itself (yet). See you next time! 🚀

Stay informed, stay curious, and stay ahead with Jumble!

Zoe from Jumble