I just did some quick testing on my own benchmark that tests LLMs as customer support chatbots, and found out that deepseek-v4-flash (scored 90.2%) was better than qwen3.5-27b (89%) and qwen3.5-35b-a3b (89.1%) and roughly equal to gemini-3-flash-preview (90.5%), but deepseek-v4-flash had the lowest cost of all of them by far. Half the cost of gemini-3-flash and an order of magnitude less cost than the qwen models.
Have you noticed the deepseek-v4-pro performing worse than deepseek-v4-flash? It performed even worse than qwen3.5-27b. I found it surprising and I'm wondering if there is a bug on my software because I had to implement sending the `reasoning_content` otherwise the API failed with BadRequestError.
It's written by Kyle Mitchell https://kemitchell.com/ who is a lawyer who has published many unusual licenses (like the Parity license) that are not approved by the FSF or the OSI. He uses simple language.
While freedom of association is fine on a personal level and maybe for small businesses, big corporation that are monopolies or oligopolies shouldn't have this freedom. They should be regulated and forced to serve everyone otherwise they have the power to exclude some people completely from such services with no alternative. Due to their power, their decisions affect people as if they are government decisions, yet we don't have a say on it like we do with the government so it's even worse than government censorship yet some people justify it because they are "private companies" as if that means something.
Although they gave birth to more children before the industrial revolution, most died, so the population was almost constant. If we define "birth rate" as the number of children that survive long enough to have their own children, then I believe birth rate was close to 2 back then. I believe long term the birth rate (according to my definition) always converges to 2 because anything else is not sustainable. Nothing can exponentially increase forever. Maybe the population can exponentially decrease to 0 but it's unlikely because there always be some group that will be willing to have more children for cultural or religious reasons.
Good idea. Random selection is interesting but I don't know if it can work today. A solutions for the issue you mentioned "Nobody knows anything about any of the candidates" is a system that allows people to vote only for people they know personally, and use some algorithm (maybe something like the PageRank algorithm that Google used) that rates each citizen according to the votes they get but also the votes are valued according to the rating of each citizen. That way the rating flows to the people who are really trusted by the people and not the best funded career politicians. Just an idea. maybe there are problems with that too if it can be gamed but it's worth trying.
install rclone and run `rclone serve webdav .` to start a WebDAV server on the directory that contains your tiddlywiki file. easy and never had issues with saving that way.
I agree that "tribalism" exists. I'd add that sometimes political disagreements are actually differences in morality. And there is no way you can persuade someone to change their moral beliefs. Everyone accepts their moral beliefs as "axioms". But I still believe it's worth discussing politics in order to learn what kind of person someone is and their morality.
The fact that there are companies that have the power to shift public policy is the problem. Because they use that power for their interests that don't align with our interests.
TL;DR: two methods (cosmic microwave background, and cepheids) are used to measure the rate of expansion of the universe that give different results. The "Hubble tension" is this disgreement about this value. Recently a new method using the tip of the red-giant branch (TRGB) gave a value that is in the middle of the other two.
The EUPL is copyleft for any derivative work (as defined by EU law) so the scope of copyleft is not clearly defined.
But EU law allows linking without producing a derivative work [1] "linking two programs does not produce a single derivative of both (each program stay covered by its primary licence)".
So this is weak copyleft. It's something like LGPL with network protection.
They have modified the EUPL license though, which means they have created another license which is incompatible with EUPL. I believe they should use a different name for the modified license. This is confusing.
> So really it seems this should be considered as licensed as those two.
If you create a different work based on the original (EUPL licensed) work and some AGPL code, you still have to abide by the EUPL but you license the resulting work as AGPL, so others have to abide by the AGPL only.
I wrote my own in-process task queue in Go tailored to my needs.
I needed auto-scaling and ability to connect different pools into a pipeline.
https://github.com/cmitsakis/workerpool-go
The 2nd though experiment about a stablecoin that that goes up 20% per year is interesting.
So in order to guarantee a stable price, the algorithm has to be able to confiscate some of your coins whenever it needs. This makes sense, but it makes stablecoins less appealing as a store of value.
Don't bother with leetcode. I recommend you get the minimal knowledge required and start building the MVP right away. That way you will have the motivation to learn.
1. choose a popular language, and familiarize yourself with the basics so you can at least read simple code
2. choose a web framework (assuming you want to build a web app), and watch/read a tutorial that uses that framework to build a web app
3. start building the MVP and search stackoverflow for everything you want to do
Have you noticed the deepseek-v4-pro performing worse than deepseek-v4-flash? It performed even worse than qwen3.5-27b. I found it surprising and I'm wondering if there is a bug on my software because I had to implement sending the `reasoning_content` otherwise the API failed with BadRequestError.