Such different time now than early 2025 when people thought Deepaeek was going to kill the market for Nvidia.

Actually the fact the inference of a SOTA model is completely Nvidia-free is the biggest attack to Nvidia every carried so far. Even American frontier AI labs may start to buy Chinese hardware if they need to continue the AI race, they can't keep paying so much money for the GPUs, especially once Huawei training versions of their GPUs will ship.

By "completely Nvidia-free" do you mean Nvidia wasn't used for training nor inference? Because if it's only inference, we know that Opus already can run on TPUs. Not to mention Gemini.

Yep but they don't run on Chinese hardware that is going to be available to everybody and will cost a lot less than NVIDIA stuff. So now you have a full non-US pipeline for AI, and soon they'll have the training GPUs as well.

That's like saying Raytheon would outsource building drones from Saheed makers (don't know who exactly).

Not gonna happen

They might still kill the market for NVIDIA, if future releases prioritize Huawei chips

Next up: Google I/O on May 19?

I have to imagine they'll go to Gemini 3.5 if only for marketing reasons.

"our newest and most expensive model yet"

I guess I'm most surprised that there are any NYC cops who don't deface their plates: https://www.nytimes.com/2022/12/17/nyregion/license-plate-vi...

I'd seen the technique, but never visualized like that. Very cool.

> Thanks to email security scanners this feature is largely broken.

One person's feature is another's anti-feature. I'm glad it's dead.

Same. I've built dozens of small tools and scripts and never felt the need to try something else.

> so many internal contradictions that when they're all listed out like this, you can just pick one that justifies what you want to justify.

More like the Bible of Software Engineering then

"vibe code" now just means "coded with AI" which should not be anymore of an insult than "IDE coded".

I'm much more critical of closed-source, subscription, wrappers over open source software of simple prompts.

> Splitting TPUs into dedicated training vs inference chips feels like an admission that the bottleneck has shifted from FLOPs to memory bandwidth + latency.

With the expected scale of inference, it makes cost sense to make dedicated hardware for each task if the workloads are even slightly different. Probably similar to the video decoding chips in TVs not being very cheap/efficient compared to chips capable of encoding video.