Given that providers of open source models can offer Kimi K2.5 at input $0.60 and output $2.50 per million tokens, I think the cost of inference must be around that. We would still need to compare the tokens per second.
There's a program called Hilo [1] in Québec where it's using the Hydro-Québec Rate Flex D [2] to automatically stop the heating during peak demand.
> With Rate Flex D, you can save quite a bit of money, since most of the time in winter, you’ll be charged less than the base rate, except during occasional peak demand events, when you’ll be charged more than the base rate.
"Ripple control" is vintage technology - hourly usage meters are not necessary.
Everyone imagines that consumers would change their behaviour if they were given price information. In my experience, I've yet to see any good data showing that on average consumers save electricity due to smart meters.
In New Zealand, I think the power companies design their consumer products to be unhelpful (what's the equivalent term here for dark patterns in marketing?). I believe few consumers watch their instant usage or review their hourly usage. Personally I changed away from a plan that used spot prices (after seeing the debacle from snow in Texas, and realising the rewards were low and that judging/managing the risks was hard).
I find Google search useful. I find the AI generated snippets they inject into the top of the page a distraction. Roughly 9/10 times the information is generic, unhelpful, or plain wrong.
I stopped finding their search competitive around 2015 but kept checking with gradually less frequency, so I can't really comment on whether anything new has added or hurt their results.
Some of their semantic web based info results seemed good and would have benefited by more complex algorithms to figure out when to show it.. But I kind of got the sense that they went into the Amazon territory where no one was authorized to sort anything out any more to get out of some kind of local optimums.
I originally moved to duckduckgo to use their !g/!b/!.. for quick search comparisons when I was very much a power user of search, but these days ddg's first results or results after adding a qualifying word are usually adequate for my usage and my power search skills are in decline.
There's a John Oliver episode about the chocolate industry, you can check it out on YouTube.
Of course there's no logo saying the chocolate is made with child slavery, I hope you didn't mean to be snarky here.
There are initiatives to try to alleviate the problem, the only example I have in mind right now is Tony's Chocolonely (which explains the problem and their solution on their website) but I'm sure others could come up with other initiatives.
The trends from top companies of chocolate is that they try to do as little as possible to alleviate the problem of child slavery in chocolate production.
It's expensive, easy to bypass, and has little enforcement teeth.
Expensive because you need to pay an inspector (ideally multiple to avoid corruption) to visit these farms.
Easy to bypass because the kids aren't working the fields all the time and there's plenty of warning an inspector is coming.
And the enforcement mechanism is you don't get the label... But like there's plenty of buyers for slave chocolate so that mostly means a minor hit on income.
NGO inspectors are better than nothing, but we should be clear eyed in how much they can solve. The only way to significantly reduce slave labor is harsh penalties on companies that trade in slave goods. That was pretty much the only thing that collapsed the ivory market.
You could buy a large plot of land in Ghana or Nigeria. Start a large cacao plantation. Hire locals to do the labour intensive work for you. And then fly in management from Western European countries to oversee the operation and to make sure that no children set foot on the plantation.
But this hasn't applied for self-driving cars, there has been huge amounts of money poured into it and very smart people working on the problem and yet the results have not really been up to the level.
Hardware products have a much harder time becoming more popular. They are often restricted to a few countries and require significant investment from customers, whereas software can be distributed overnight to a hundred different countries at once.
Hence, the economics of both are vastly different. It is much harder to reach the viral inflection point with hardware than software, even with great products.
