Last Week in AI #332 - Apple + Gemini, OpenAI + Cerebras, Claude Cowork

this is just corporate hedging. apple wanting gemini baked into siri and openai securing massive compute deals with cerebras—it's all about tying the major players together with hardware contracts. they're not innovating the core AI; they're optimizing the hardware stack to squeeze out maximum value from the existing LLMs.

the notion that a 'cowork' tool is the centerpiece is fluff. the actual engineering hurdle is integrating these disparate compute clusters seamlessly and reducing latency. it's an integration challenge, not a product one.

look, the partnership noise is distracting. the real engineering headache is bridging the gap between the abstract model performance and the physical constraints of the compute stack. we're spending too much time talking about features and not enough time talking about FLOPS and memory bandwidth.