With AI assistance - another Eric Barone could conceivably create a similar game in a fraction of the time

Yes, the density vs power tradeoff could make a lot of sense in a few applications.

I wonder if it would start to make sense for SRAM caches to shrink (or not grow as fast) in favor of adding on chip DRAM cache at some tier.

Or AMD’s putting sram on a separate chip. I’m surprised they didn’t go there yet except for extra L3 instead of all of it. At some point the costs will shift the decisions.

They claim win + win*10:

> provides up-to 50% area reduction and reduced power consumption by a factor ten

I know a freelance translator and they have more business than ever, AI being "mostly" right is not good enough in many businesses. You need oversight and most importantly - accountability in the form of a human that can put their stamp of approval.

You can, and there are many techniques to do so - it's not super precise but you can essentially "delete" concepts out of a model.

What's your distinction between real time vs multiple steps? All computation is done in steps.

Is it all one big computation? Its turtles all the way down.

Claude get me a job that pays $200k/yr. Make no mistakes.

Claude get me a job that pays $200k/yr. Make a breakthrough.

Sending email is already cheap, not free. Just so happens the cost is subsidized/hidden for most regular people and its so cheap that they give it away for free. For those of us that run transactional email, the cost is ~ $0.0001 per email for AWS SES, already at the fraction of a fraction of a cent you suggest.

Yes, but I want the rate to be variable depending on volume. I want it to be $0.0001 for a simple personal email. But if you send 10,000 emails today, I want it to cost you $100 (or more).

Many schemes to do this. Things like:

The first 10 emails is $0.0001 each.

The next 10 will be $0.001 each.

And so on. You can always fiddle with the thresholds and costs.

I'm 10000 people sending 1 message each, to get the cheap cost for every message.

You're forgetting the 5 cent charge per recipient who hasn't put you in their approved list.

Anyone have any idea what the architecture/vendors they are using for inference/compute?

Getting the compute to run inference for multi-trillion parameter models at any sort of scale and performance is daunting. There are a handful of vendors that have systems that can do this (~ Nvidia NVl-72 class) that pretty much only the frontier labs and hyperscalers effectively have access to.