We can probably blame copilot for the email about new verification reqirements not going out to everyone. Maybe even for the reports of people who jumped through all the hoops and still got blocked as if they hadn't. But rolling out new verification reqirements, then not monitoring how many developers fulfill your new reqirements and following up is entirely on Microsoft employees. That's management failure and disregard for developers on their platform

If we assume an average power draw of 20W per laptop, that's 300W for each 15 laptop unit, or about €57/month in Hetzner's Finish DC (including aircon)

Not sure about network. A 1Gbit uplink with 10TB traffic (and €1/TB after that) is provided. Upgrading that to 10Gbit is probably similar to the €51/month cost for the same uplink for dedicated servers, so another €15 for each 15 laptop unit. Plus around €2/month/IP, but you can probably bring your own if you find a cheaper subnet to buy

So yeah, you are right that the math does not work out. But it is pretty close to break even. I think you can break even on this if you find a more space efficient way to cram them into the rack and don't pay yourself any salary

https://www.hetzner.com/colocation/

This is also true for an entirely different reason: all three of these fields would benefit hugely from having more balanced gender ratios

Getting the right people might be an issue. But the tech sector is actually a great example of how to do that. The tech sector has also grown beyond what it could sustain with just people who are in it for the love of tech, and how it managed to do that would be a great case study. A good work-live balance, consistently high salaries and great outreach were certainly among the contributing factors. All of which boil down to throwing money at the problem in a smart way

If you're only using flagship model providers then openrouter's value add is a lot more limited

But for the party with the most responsibility for blowing it all up I'd like to nominate Rupert Murdoch. Most visibly with Fox News, but really his entire media empire

I could pay API prices for the same models, but aside from paying much more for the same result that doesn't seem helpful

I could pay a 4-5 figure sum for hardware to run a far inferior open model

I could pay a six figure sum for hardware to run an open model that's only a couple months behind in capability (or a 4-5 figure sum to run the same model at a snail's pace)

I could pay API costs to semi-trustworthy inference provider to run one of those open models

None of those seem like great alternatives. If I want cutting-edge coding performance then a subscription is the most reasonable option

Note that this applies mostly to coding. For many other tasks local models or paid inference on open models is very reasonable. But for coding that last bit of performance matters

https://energy-charts.info/charts/price_spot_market/chart.ht...

As noted in the methodology below, they are measuring the gas-implied level as the marginal running costs of a combined-cycle gas turbine plant: the price of the natural gas necessary to generate a given amount of electricity, plus the cost of the necessary carbon credits to burn that natural gas. Then they compare that to the actual electricity price

We love moving plants and animals around if they're useful or pretty. Conservation efforts that try to stop this for certain species are a relatively recent thing

Kaparthy's notes on improving nanochat [1] are one of my favorite blog-like things to read. Really neat to see which features have how much influence, and how the scaling laws evolve as you improve the architecture

There's also modded-nanogpt which turns the same kind of experimentation into a training speedrun (and maybe loses some rigor on the way) [2]

1 https://github.com/karpathy/nanochat/blob/master/dev/LOG.md

2 https://github.com/kellerjordan/modded-nanogpt