deflator

That's the thing though: The billions have been there for a while, but the bad corporate behavior is relatively new from tech.

Not bad! I like how it got the motion lines on the correct side. IIRC, many of the other ones you've posted have the motion lines on both sides of the pelican

I still get excited by the sheer power of current and upcoming tech

It may be the same stuff we had 20 years ago, but the specs!

What was a supercomputer before is now just a component in a datacenter, one of 72 per rack.

Makes sense to me.

We will see, since if true then it is likely the other makers of small, dense models will copy it and include high reasoning by default.

If that also makes the other dense open source models better, then you are probably correct.

The first thing I thought was: this might make Tesla's cameras also not see your funky-patterned Yaris on the road as a car. Might want to test that first!

An electric ute? I'm interested. Seems like a good move!

Just like everything else: to flood the channel

The current environment comes to mind

I second this. Anyone considering running their own hardware would also consider quantization since they would have full control over what quant to run.

The value proposition changes a lot if you can get 90% of the quality for 50% the price with a quant due to halving your hardware requirement.

Thank you for the article, though! Very informative.

Hmm. Pennywise operated on a 27 year cycle. Maybe Cthulhu does the same, but at 1MM the scale.

Eggs don't lay eggs

Very informative article. I was hazy on where NVLink sat in the local inference stack. This makes it very clear: to adapt consumer hardware that lacks multiple full x16 PCIe lanes.

Those were the days

I switched to Codex in November 2025 because I couldn't stand the smarmy, "You're absolutely right" anymore. Codex at the time offered less sycophancy and more pushback when I was actually wrong. I also feel it writes cleaner code since it follows my instructions better to not write unrequested code, not create unnecessary abstractions, and not reinvent any wheels. I use LLMs on a tight-leash in HITL for web development, monitoring traces as it goes. I rarely use it for more than just writing code that I specify and I rarely let it "have it's head" and work unmonitored. I use it on the API plan since I need it to be reliable. No outage issues that I can recall.