TradingPlaces

The more accurate way of looking at it is using the "Office and accounting equipment” deflator, which puts it at $1291 in 2022 dollars, about the price of a decent laptop, which makes sense to me.

https://apps.bea.gov/iTable/?isuri=1&reqid=19&step=4&categor...

The Selectric remains the world’s best feeling keyboard.

I miss the needless formality of Apple Garamond.

Terrible site. But I downloaded the data referenced. My conclusion that central banks are choosing the dollar less as a reserve currency. This is a long term trend of managing risk by spreading it out more. At first that was the euro, but now it’s a much wider range of currencies, including Chinese.

What it means is less demand for USD and subsequently the distortion of its price will lessen. It could also make the budget deficit much more problematic if it goes very far.

We’ll find out next week

The article does not say if it’s a Mac Studio or Mac Pro. I think that matters. My understanding is that there is no TDP on Apple Silicon, and that it has a lot of runway in a less constrained thermal environment like the new Mac Pro vs Mac Studio.

While I’m still surprised they didn’t put a second Ultra in the Mac Pro, I’m betting there’s a wider delta than people imagine between the two form factors.

Violations of the Stringer Bell Rule will always get my attention.

https://youtu.be/Ly82nabRRYc?t=75

Chris Lattner and Tim Davis. Google in on A-round https://www.modular.com/about

Don’t sleep on the Modular inference engine

https://www.modular.com/engine

Came for everyone ripping on a16z. Was not disappointed. Thanks y’all.

OP here. That’s exactly right. In the article, I describe CUDA and the rest of the software suite as their competitive moat to hardware competition from AMD or AI accelerators.

This is one of those fluky months in the household survey. All the negative number in employment can be accounted for by 1) government wage/salary worker and 2) unincorporated self employed. There is pretty obviously some huge error in there, especially 2) down 4.1% MoM. Self employed are not counted in the employer report

In the household report private non farm employment was up 395k. In the employer report private nonfarm payrolls were up 339k

Kagi’s Orion browser also has this. I like it.

The model may be the same, but the chatbot is not. They have made the responses shorter to save inference expenses, which are huge in a model the size of GPT-4.

If you prefer data to anecdotes:

https://www.sanfranciscopolice.org/stay-safe/crime-data/crim...

As people note, you cannot substitute locally for the Azure GPU cloud that GPT-4 runs on. But I believe that will change, and maybe quickly. After years of explosive exponential growth in model size, all of a sudden, small is beautiful.

The precipitating factor is that running large models for research is very expensive, but pales in comparison to putting these things into production. Expenses rise exponentially with model size. Everyone is looking for ways to make the models smaller and run at the edge. I will note that PaLM 2 is smaller than PaLM, the first time I can remember something like that happening. The smallest version of PaLM 2 can run at the edge. Small is beautiful.

Inference is far more expensive on GPT-4. My take has been the same and I think it’s a cost-saving move. The responses are shorter and less complete than they were just a few weeks ago.

Right now, I don’t think it’s possible to scale a service really big based on GPT-4 because of cost.

One of the strangest and most awesome things in LA. RIP Milt.

I did when I got an Apple ][+. Barely picked up my soldering iron again.

Came for everyone roasting a16z. Was not disappointed.

Regulation that puts up a barrier to entry that prevents competition: good

All other regulation: bad

From Yale Law professor and frequent shitposter Scott Shapiro’s new book, Fancy Bear Goes Phishing https://www.penguin.co.uk/authors/122489/scott-shapiro

A thousand times “no,” for every “yes”.

My read is this is primarily radio filters, and all the stuff they were already buying from Qualcomm. There are a lot of them in 5G iPhones. Biggest loser here in my read is Skyworks, who provided a lot of the front-end-modules.

Great product; happy subscriber. I never bumped up against the 1k searches, but thanks anyway, Vlad!

Foxconn somehow made $137m profit in the past 12m.

Lambda has a very interesting benchmark page https://lambdalabs.com/gpu-benchmarks

If you look through the throughput/$ metric, the V100 16GB looks like a great deal, followed by H100 80GB PCIe 5. For most benchmarks, the A100 looks worse in comparison

Badhla Solar Park in India is the biggest now, I think. 2.24 GW

https://goo.gl/maps/LUXNbviB7GuZ8Vup6

Hi Vlad! Kagi subscriber here. Been thinking about this since Neeva dropped that, and I think bootstrapping is why you’re still running and they aren’t. Good luck!

I think this was more about keeping the Imagineers in Glendale, California, where they all wanted to stay. This move was a key source of internal dissatisfaction with Chapek.

Background: https://seekingalpha.com/instablog/49663329-trading-places-r...