I live in Oslo, Norway. I'm driving a BMW i3 I bought new 12 years ago. It was a piece of shit the day I bought it. It's the exact same piece of shit now. I've calculate that so far, I'm at about $245 USD per month total cost of ownership including charging and toll booths across these 12 years. I suspect I can drive it for another 6-10 years and by then I expect the TCO to have dropped to about $195 a month before it starts increasing again.

I can't drive this shit heap too far, 120km on a great day in the middle of summer. And I had an incident a few years back on the countryside in -11C temperatures where I only got 50km in the mountains that day.... that was REALLY COLD but I had blankets for the 3 hour wait for the tow truck.

We'll take a vacation in a few weeks, we'll rent a crappy tesla and drive to Gothenburg or Stockholm. Still massively cheaper than wasting money on a new car that would give me better range for the week or two I need it to. But I still go grocery shopping a few times a month in Sweden because I'm not stupid and this crap car can make it there with a little extra charge each way. Which is no problem because there are chargers at pretty much every exit and they're cheap.

So, that brings me to trucks...

1) one driver 5 truck caravans. This is close and it's going to happen. In fact, I'm pretty sure within a very short time, we'll see most trucks driverless on the highway and then you'll see professional drivers being delivered to trucks for last-mile travel where the trucks will still do the driving, but supervised. And then in time, no more drivers... except in New York where some guy named Joe Joe who graduated from "Vinnie's airplane towing school" is considered more qualified than an airline pilot with a hundred thousand hours of flight time and a formal education and tows airplanes to the terminal because "we're a union shop".

2) Telsa isn't selling the trucks to make money from the trucks... oh they will... but consider that a model-X at 1/50 the size is like 1/3 the cost of the truck. The money comes from the infrastructure. Tesla will build out a charging network even if they have to dip into the college funds of Musk's 72 children to do so. And he'll use fuel cell and solar and possibly wind to charge up the battery power banks. He'll then find a way to cut deals with the truck stops that he'll install the chargers at their truck stops in exchange for parts of the food profits and for massive discounts on any fuel needed for the fuel cells. Then he'll make a deal with netflix and disney plus and all of them for the entertainment systems in the trucks and he'll charge the truck drivers for Starlink. And he'll then have a huge fleet of spare Tesla Semi's that can be quickly delivered to drivers on the road who have problems. They'll then arrange for the semi to be serviced and returned to the driver as he passes back. Of course, all at a cost.

Within 5 years, he'll have streamlined most of the US land freight infrastructure, bled everyone dry in the process and be strongly positioned to sell Tesla Semi as a service in competition with the existing fleets.

Europe and Asia (outside of Russia and India) will be a much tougher win because we prefer rail where possible. And at least here in Norway, we would still rather invest in new rails rather than new roads.

The Windows Start Menu is so frigging confusing that I use it for two things

1) press windows, type name of app, press enter
2) press start, type basic equation (I.e 5/1.35), pray it gives me numbers and not an advertisement for feminine hygiene products

I am terrified when I need to use a mouse on the start menu. It's as though it can do everything except start programs

What makes solar financially un-viable in Austin? I live a long ways from Texas so there might be something really obvious here that I'm missing. I know the solar power arguments are older than the hills at this point (I remember roof top solar panels from many decades ago) but I'm interested in what about that location makes it doomed to not work out.

Or is there just no way for Tesla to make it financially viable? Could a more honest company pull it off in a different way?

Broadcom is actually fabless, except for some mostly irrelevant research facilities.

I have a crappy old IDE (not even SATA) cdrom in my closet which I bought a usb to IDE cable for. I can't find the power bricks to any of my old laptops with working DVD or CD drives. I think it would be hillarious if they just made a CD with an autorun.inf that played Never Gonna Give You Up. How many people would actually know

I just wanted to say. I don't think C or C++ are good languages and while I think clang is a pretty decent C++ compiler, I think it's far from what it could become.

Why not use an llm to audit code?

Maybe they should use a lot of LLMs, get second and third and fourth oppinions.

Or is the article making a point that the US government is limiting themselves to one llm and if you want to hack the government, you just need to find Mythos's weaknesses?

Seriously?

I used to hand code boot sector viruses by disassembling the msdos mbr and boot sector in debug.com and then making my own.

These days, if I wanted to hack, I'd load up vllm on wsl2 running Qwen or similar, open up opencode and just tell it to hack. I would do a much better job of covering my tracks, but Windows, Linux, Mac... it really doesn't matter. They're all the same.... the trick is you start by saying "find me a cloud service somewhere which I can use for running anonymous Linux vms, or find me a way to setup payment methods I can use to anonymously create accounts on cloud services".

I think NVidia is not particularly worried about it.

NVidia is an odd duck because they are profitable. And, they have now locked in long term contracts in business, industrial, automotive, aerospace and more. NVidia also has a surprisingly small footprint. Only 42000 employees. Facebook by random comparison has 77000, IBM has 264,000. They have $80 billion in cash and infinite credit lines.

Jensen is already working on the next big thing because he knows data centers is short term. And consider this, a full populated HGX B200 Board (8 GPUs + Baseboard) which costs $350,000 a pop, they have a 75-80% margin (all public info) and they sell data centers full of then. And, oddly, even though it will be a fraction of the margin, NVidia's consumer CPUs are soon going to utterly eclipse their data center profits because they let Qualcomm foot the bill to get Windows on ARM ready, got Microsoft to tune WoA for NVidia, and my guess is, they'll get MS to seed 5000 developers with dev units at Build.. which Qualcomm didn't do... which is why they failed.

I have been testing Intel ARC heavily for a week for local AI. Intel has a lot to do. I don't think AMD even has a plan.

Jensen is creepy as hell and is using FUD to cash in. And this announcement will have him staging a dinner with someone important to make a big deal about how this will push China into the lead and the US needs to get serious.

BTW, Deepseek making their own chips is a non-starter. They already have a tiny footprint compared to their peers, they actually wouldn't profit from it. It might just ge a fund raiser.

October 13, 2023
I would have expected the layoffs closer to two years than three, but Microsoft probably had a huge accounting mess to sort from the merger. And the DDR price hikes needed to be massaged until it could be a strong enough excuse to appease the unions.
The layoffs were planned as part of the buyout. They always aee.

Notepad++ just takes the best features of Kate and rolls them into a free Notepad replacement for Windows users so they don't suffer from its eternal failings. We don't need it in Linux, we already have all of its features. Windows users do need a functioning text editor out of the box, which Notepad++ provides as a free add-on instead.

I currently burn about 30 million tokens a day. I have a 98-99% cache hit on them. An RTX3090 and RTX5070 generates most of the rest at about 200t/s. I use Deepseek online as well for now because at $0.35 a day for 5-20 million tokens/day, it's cheaper than another GPU.

I expect that when laptops with 128GB RAM and a decent NPU happen, I'll start burning closer to 100 million tokens a day.

I don't think I could spend $2000 a year on tokens if I tried... Even a laptop is a 5 year purchase, so that can't cost more than $1200 a year

RAM is a pretty good gamble. There are two korean companies, a US company and a Chinese company, The entry barrier is high enough that even now, we should be executing Infineon's former leadership publicly for threatening the supply chain. CXMT is the best thing to happen in 20 years and Norway should drop $100 billion on licensing and building a European supplier (no, the EU can't do it, this is a one country thing, Germany is the only other one who could/should try, but their government is too broken because companies like VW and BMW would steal the money, claim it went to the workers and blame China).

If you're going to risk the entire national economy on something, RAM is safe. Whether due to AI, robotics, computer vision or anything else, 128GB is about to become the next 10 year norm. Every phone maker is planning on 32GB budget phones in the next 4 years. Now that China can mass produce RAM and is scaling massively, Korea has to up the bar.

BTW, if Trump doesn't lift restrictions, CXMT will continue to steal Korean RAM tech ... while Korean and US companies have to pay licenses for Chinese tech. CXMT is at the big kids table now. And if the US doesn't allow CXMT to license US and Korean tech, CXMT will just use it for free... which will make CXMT chips cheaper than US and Korean chips and threaten the world supply chain.

Even if this is the case, Tata has always been a disaster. There is a very good reason their cars never make it out of India. Their quality control is dismal. I honestly am not comfortable buying anything made by them. I would expect heavy use of hexane, toluene, xylene... and all those other lovely chemicals that are dangerous to the workers but give iPhones that new phone shine.

I hope Apple gives buyers the option to not buy Tata. I just feel they're a company that makes money at the cost of their workers health. And unlike the Chinese when they get caught, Tata will just find better ways to hide it.

That could well be a chicken-and-egg problem. Are the buyers buying what's available because it's available, or are the manufacturers making it because the buyers actually asked for it? I'm not aware of any Equinox driver anywhere who ever complained about it not being tall enough (or it being short enough to make it unnecessarily difficult to run over small people) yet Chevy raised the hood anyways.

The other problem is that the auto manufacturers see a distorted picture of car buyers. New cars are too expensive for a large fraction of all drivers; many drivers won't ever buy a brand new car. New car dealers are selling to people with more money, and making decisions around what those more affluent people might want in their cars - or more so are telling such people what they should want.

I moved away from American AI. The risk of a cloud AI being yoinked from me as soon as I depend on it is too high.

Besides, anything Anthropic is peddling, everyone else will have shortly after.