Uladzislau Bayouski

Two days, ninety-plus sessions, and a quiet shift in what AI is actually for. Most people watched the smart glasses demo and the Gemini benchmarks. The headlines wrote themselves. That's the surface. Underneath it, Google quietly retired the idea of AI as an assistant and replaced it with AI as an operator. The Short Version The keyword across both days — repeated in nearly every announcement — was agent . Not chatbot. Not assistant. Software that takes an instruction, decomposes it, and operates on systems on your behalf, often without you watching. Sundar opened with the numbers that made that inevitable: 19 billion tokens per minute through Google's APIs, 2.5 billion monthly users on AI Overviews, AI Mode crossing a billion in twelve months. At that scale, the assistant model collapses. Nobody types a billion prompts. The work has to happen on its own. Day 1 — the consumer stack: Gemini 3.5 Flash — 4× faster than competing frontier models, beats Gemini 3.1 Pro on ev...

Google Just Solved One of AI's Biggest Hidden Bottlenecks Every time an LLM generates a response, it quietly runs one of the most memory-hungry operations in computing — the KV cache . As conversations get longer and models get bigger, it becomes a wall. Expensive. Slow. Stubborn. Google Research just published a paper that hits that wall with a sledgehammer. The Short Version Traditional quantization methods save space on data, then spend it right back storing bookkeeping constants. You compressed the model, then padded it back out again. 🤦 Google's answer — TurboQuant — is a trio of algorithms that eliminates the overhead entirely, not by compressing it, but by redesigning the geometry so it was never needed in the first place. The results: KV cache down to 3 bits — no retraining required Memory footprint reduced by 6x Up to 8x speedup on H100 GPUs Long-context benchmark accuracy? Essentially unchanged And it outperformed methods hand-tuned to specific datase...