sergiopreira

So Google is basically admitting PyTorch/XLA on TPUs didn't work — TorchTPU looks like them rebuilding what should have worked day one. Its hard to run production ML on a toolchain engineers can't trust, no matter how fast the silicon is.

DeepSeek is commoditizing frontier capability... Opus 4.6-level benchmarks at a fraction of the cost changes also who can access these tools.

Stuff that was prohibitive six months ago is now up for grabs. We keep on working on the infra level now, swithcing models whenever we run out of credits, or want a different result. The question is how do we build context, architecture and ensure the agent is effective and efficient..... wouldn't it be good if we simply used less energy to make these AI calls?

Anthropic keeps conflating two distinct strategies — be the best model for developers to build on, or be the company that ships Claude Code. Those two have opposite policy conclusions. Restricting third-party harnesses maximizes Claude Code revenue; allowing them maximizes model-layer lock-in through developer habit. The whiplash is the symptom of not picking. Pick for crying out loud!

Most 'runs on Mac' ports are a wrapper around a cloud call or a quantized shell of the original model. Going after the CUDA-specific kernels with pure-PyTorch alternatives is the kind of work that ages well, because the next CUDA-locked research release is three weeks away. One question: how much of the gather-scatter sparse conv is reusable for other TRELLIS-like architectures, or is it bespoke to this one?

An interesting question is whether the tokenizer is better at something measurable or just denser. A denser tokenizer with worse alignment to semantic boundaries costs you twice, higher bill and worse reasoning. A denser tokenizer that actually carves at the joints of the model's latent space pays for itself in quality. Nobody outside Anthropic can answer which it is without their eval suite, so the rugpull read is fair but premature. Perhaps the real tell will be whether 4.7 beats 4.6 on the same dollar budget on the benchmarks you care about, not on the per-token ones Anthropic publishes.

Agreed, but right now they're only pushing papers, and what for? So companies can then move to the most favourable state?

This should be done at a federal level, and the US needs to be the global driving force behind it, ensuring all countries adopt and adhere the same principles. We need the equivalent of the International Atomic Energy Agency to manage this going forward.

Could this be even more dangerous than a nuclear bomb?

Aha, yes indeed! Given AI right now....

A key theme in the book is the relationship between power and technology. Pynchon explores the idea that those who control technology also wield immense power, and that this power can be both destructive and corrupting.

Super befitting to the current discussion around AI.

Well, they don't seem to want revenue... the amount of comments I've seen of ads staying in pending, and not being able to advertise...

Drop in the ocean though. Pocket change for Elon.

This will spread to corporates as well now. The proliferation of AI is going to make everyone more wary of data being extracted and used incorrectly.

I'm currently building a development platform that has a long term vision of helping to solve this....

I would love to learn from more experienced engineers how they manage the changing requirements, the legacy issues, etc. when managing complex applications. I'm not talking about the code quality itself.

Are they? Or are they simply implementing the innovation of others?

Is anyone actually using ChatGPT to help them with some mundane coding tasks? If so, what?

It would be interesting to understand what % of searches actually need ChatGPT. I figure more than 80% don't require it.

Haha! Brilliant. Next is to configure based on "highness"....

Yes and no. Platform engineers are generally very skilled developers+ devops that are able to built internal solutions for their colleagues (developers) to be able to deploy to production more effectively. So, basically, a platform engineer is removing obstacles and improving efficiency of your CI/CD pipelines. This can be done with workflows, UIs, etc. A cloud engineer will be able to manage cloud deployments, but may not be able to build these internal tools effectively.

Very cool video. Nicely done! I tend to eat very little salt, but only because I don't like it very much. Will probably look into this in more detail.

Very cool list. I'm starting a 1 book a week this year, and kicking it off with 1984 by Orwell. Will definitely take a look at this list when I choose the next one!

Yes. Without a doubt. Newspapers were supposed to be curators of valuable content, but most went down the road of click-bait for advertising. I've also tried paid solutions, but most fail with the exception of the economist. Its the only one I pay for today....

We've been doing this for over 6 months at work, and whilst it seems to be ok, its still not 100% clear its working... I think you need to have many other things working well before implementing such a routine.

Not sure, but sounds like a great gimmick for dating apps.