Google speeds up Gemma 4

Gemma 4 gets a speed layer

Google says Gemma 4 can now run up to 3x faster without changing the model’s output.

The update adds Multi-Token Prediction drafters to the Gemma 4 family, using speculative decoding to guess several future tokens before the main model verifies them. The drafters are available under Apache 2.0, with weights on Hugging Face and Kaggle, plus support across tools like Transformers, MLX, vLLM, SGLang, Ollama, and Google AI Edge Gallery.

The important part is practical. Standard LLM inference often gets stuck moving model parameters through memory, which makes local and on-device AI feel slow even when the hardware has unused compute.

My read: Google is trying to make open models feel less like demos and more like working software. Speed matters most when a user is staring at a screen, waiting for the next token to appear.

The ripple is that smaller teams may get more serious about running capable models on laptops, phones, and consumer GPUs. Google also gets to push Gemma deeper into developer workflows before the model race moves again.

Fast enough may become the real test.

Apple pays for the AI it promised

Apple is paying $250 million over iPhone AI features that buyers say were advertised before they were real.

The settlement covers some US buyers of the iPhone 15 and iPhone 16 between June 2024 and March 2025. Apple admits no wrongdoing, but eligible customers could receive between $25 and $95.

The case centers on Apple Intelligence, especially the promised Siri upgrade that was supposed to make the assistant more useful and personal. On a dark stage in 2024, Tim Cook sold Apple’s AI future with the usual calm confidence, but the product did not match the pitch fast enough.

My read: this is less about one delayed feature and more about Apple trying to sound caught up in AI before it was ready. That is unusual for a company that usually waits, ships late, and claims polish as the point.

The ripple is simple. Apple’s AI marketing will get more careful.

The real question is whether customers still believe Apple can make Siri matter.

Samsung rides the AI memory squeeze

Samsung’s stock jumped more than 15% as investors chased the companies feeding AI’s appetite for memory chips.

The rally pushed Samsung’s market value past $1 trillion and helped lift South Korea’s Kospi above 7,000 for the first time. Last week, Samsung reported first-quarter operating profit of 57.2 trillion won, more than eight times higher than a year earlier.

The key detail is supply. AI systems need fast DRAM, NAND storage, and high-bandwidth memory, and new chip capacity can take two to three years to come online.

My read: investors are treating Samsung less like a slow hardware giant and more like a direct beneficiary of AI infrastructure spending. That makes sense, but the stock move also assumes tight memory supply keeps pricing strong.

There is still pressure. SK Hynix holds an estimated 55% of the HBM market, while Samsung has roughly 25%.

The room is bright for Samsung right now, but this rally depends on memory staying scarce.

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.