HN user

londons_explore

39,505 karma

https://omattos.com/contact

Posts10
Comments14,434
View on HN

Utility scale power price varies widely by location and exact time of day.

LLM's aren't very latency sensitive and can therefore move to wherever power is cheapest.

Right now that's places next to aluminium smelters (which also like very cheap electricity 90+% of the time).

If there was a reliable profit to be made, a third party would do it.

I suspect the real profit is made by also being in a position to adjust laws and rules to make sure your investment survives, as well as turning low rated bonds effectively into government bonds. All those things are at the expense of others in the bigger picture.

Regressive JPEGs 5 days ago

There are already webcams which do this- but they use a mime trick for 'multipart/x-mixed-replace'.

That's basically the server telling the client 'That data I just sent you, well now replace it with this new thing'.

No JavaScript needed, and can work with plain http and jpeg

Regressive JPEGs 5 days ago

You can also deliberately have the server sending data at the right rate for the right playback time.

Easy enough to add a delay() each frame if your server is python/nodejs/PHP/whatever

Both Android Auto and Chromecast are great ideas, but so incredibly flakey.

The number of times I've heard someone say 'oh it's disconnected again' is extreme.

To be honest, some court ought to simply demand Google makes it reliable or refunds all money people paid for the phones as the product is not fit for purpose.

Insiders who have locked up stocks but still want to sell could presumably just short the stock and take out a loan secured on the stock to get the financial effects of selling, without actually selling...

I wonder if that's what's happening with ~$1T of stocks currently locked up...

I wanna see some combined treatment of drugs which have shown marginal effects in trials.

Eg. Combine the most promising 50 drugs, each at super low doses.

Low enough that side effects probably won't be bad, and if some prove entirely ineffective it doesn't matter.

But then, deliberately don't mix the ingredients precisely. Some people will get 10% more or less of each ingredient at random. Make precise records of this.

Now you have a natural experiment. Gradually increase dosage of elements that prove effective, and decrease ingredients which prove counterproductive - and you'll have a huge dataset with tens of millions of people wanting dementia treatment.

Probably not actually. Transferring one kilobyte across a network link has such a low value that the billing costs of aggregating it cost more than the revenue.

So instead you take a probabilistic approach - charge the user for a megabyte of data transfer 0.1% of the time, and bill nothing 99.9% of the time.

Now the typical cost is the same, the users bill is probably accurate to the cent, but you have divided the number of billing records by 1000.

Listening to two talkers at a time is certainly doable...

As is talking whilst listening to another conversation - eg. Giving a lecture whilst eves dropping on the people talking at the back of the lecture theatre.

However, having a two way conversation with one person whilst listening to another is really hard.

Not sure why.

This strikes me as a huge amount of custom code and technical debt. Every new software dev probably has to learn this.

Why not a sticky master election per user, and have no special data centers?