It's Void with GUIs around runit and xbps. The problem being attempted is that Void is not very accessible to users new to Linux. This is pretty clear from the website.
> I can see why that might feel like an existential attack on Zig even if starts with a bit about how great Zig is.
The original post explicitly praised Zig, and seemed to be arguing more that Zig was a great tool for the initial version of Bun, but that Rust was better for their needs as Bun grew to a larger project with a larger contribution team.
That seems completely in line with what Zig claims its strengths are. Zig's response to "but memory bugs" has basically always been "Zig is not a language for a big tumultuous project that you're going to throw interns at."
I would be careful granting that s is less than g. There are lots of incentives other than scalping for people to create extra illegitimate accounts on Steam, and one individual can often control hundreds of bots, whereas a legitimate user almost never has more than one or two accounts.
Anthropic has alleged that this model is much more dangerous than other currently available models. Their CEO has said so publicly multiple times. It's like asking why cesium isn't banned if nuclear missiles are banned.
(whether Mythos is actually that dangerous is beside the point; considering that Anthropic claims that it is, it makes sense to regulate it)
The entity you're implicitly proposing should take the money doesn't have a good track record of spending it wisely. I'd much rather let the startup founders keep their money and build rockets with it, than spend another few trillion on building 100 meters of high-speed rail.
Is the "meddling" just running campaign ads? I don't really see how an election where voters' brains were hijacked by tiktok ads funded by foreign governments is less legitimate than one where voters' brains were hijacked by tiktok ads funded by local organizations.
If the commit was prompted by a core developer, the developer knows what the prompt was. If it was prompted by a stranger, the core developer reviewing it does not know what the prompt was. The review attention required is completely different, because with an untrusted submitter you have to meticulously hunt down intentional security vulnerabilities obfuscated in the PR.
I don't see what the core complaint is then. The guy the public voted for can refuse to spend public funding on a particular grant. There's no reason that it's somehow more pure to have the public vote for someone who appoints someone who appoints someone.
It's useful progress. Decent-fidelity local-scale inference means that you can create a product that generates throwaway images frequently without worrying about cost. Thus far every product I've seen that generates images is metered, which severely limits the value. I don't know if this is actually at the "decent fidelity" point yet.
Bugs exist in human code too. The AI derangement crowd pounces on any small bug as evidence that AI is a trebuchet, and thinks that if only we didn't use AI there would never be any bugs (like five years ago when all software was perfect, was not being enshittified, and had 0 bugs).
Why do we need any tools at all? Software worked perfectly fine when people were editing code with `ed`, so I'm going to go open timewasting issues complaining about FOSS devs using an IDE.
My understanding of caching with most models/providers is that a prefix substring of the context has to be reused for a cache hit, but not necessarily the whole entire context window. So if you prune tool calls from the history, you're going to get one cache miss on the newly-pruned history, and then you're going to be getting cache hits on every subsequent turn, with a lower number of input tokens. If you prune subsequent tool calls after that, you would still get a cache hit for the already-pruned portion of the context, just not the full context.
> It’s like if I go to Golden Gate Park and pick one flower, I shouldn’t do that, but no one cares. But if I build a machine to automatically cut every flower in the park because I want to sell them, that’s different.
It's not like that, because flowers are a physical object and moving them to one place deprives their original location of the flowers. When an LLM learns something from a webpage, the webpage is still there. Whatever 'theft' you perceive is entirely in your head; you were deprived of nothing by someone else making a copy of your thing.
It's really not uncommon for me to speak to someone who actively doesn't want any sort of adblock on their computer. I would say maybe 5% of people, anecdotally, just don't want it, even when you're in front of their computer at that very moment, and offer, and insist that it would take 30 seconds to install. It's not a majority, but I found it surprising.
Half of his recommendations for alternatives take less time to make. You (and the author) are making assumptions about what "people in general" think without any data to back it up. What you've experienced anecdotally in your social circle doesn't necessarily apply everywhere.
How is this at my expense? It's at the expense of the hedge fund they bought the oil futures from before the price went up. I don't see why I should assume that some hedge fund manager is more on my side than some insider trader.
Some providers are based in the US or EU and would face legal repercussions for lying about what they do with your data. It's a bit more than "trust me bro". Off the top of my head, you can use Fireworks, for example, which is based in California and would face the same consequences for lying about their data policy as OpenAI or Anthropic would.
I've been having the same experience. Tasks like "go through this entire module and pedantically make it match my preferred styleguide exactly" were not worth a couple dollars with frontier models. It's nice to be able to put deepseek flash on stupid, unnecessary or highly speculative tasks without thinking about the cost.
A friend of mine uses it for D&D prep and has told me that it's good for that in particular because of its ability to match the flavor/style that he's going for. He prefers ChatGPT for everything else.
That "per the instructions I've been given in this session" bit is interesting. Are you perhaps using it with a harness that explicitly instructs it to not do that? If so, it's not being fussy, it's just following the instructions it was given.
It's already illegal to threaten journalists. In America we generally make bad things illegal, not activities that could become motivation for bad things. Someone threatened me on League of Legends last week. Should we ban the game?
This isn't useful information without also knowing how common it is for newly-created accounts to place and lose bets around that size. Polymarket is a large platform with a lot of accounts being created per day. If two accounts made large bets and won and eight accounts made large bets and lost, you haven't discovered anything interesting.
He doesn't include the best solution in the 'what actually works' section: Give your LLM the same level of permissions that you would give a human you just hired in the same role. The examples given, tricking the customer support LLM into sending text messages to all users, or into transferring money, are not things that you would ever give a human customer support agent the tools to do. At some businesses that employ humans, you have to demonstrate good judgement for months before they even let you touch the keys to the case that has the PS5 games in it.
This is just counting pypi packages. Why would I go to the effort of publishing a library or cli tool that took me ten minutes to create? Especially in an environment where open source contributions from strangers are useless. If anything I'd expect useful AI to reduce the number of new pypi packages.
I don't see how this is more degenerate than betting on roulette at a casino. Prediction markets usually provide more efficient odds than casinos because the house profits from trading volume instead of from the spread, so it's essentially just a way to bet on a game of complete chance with a much better average-loss than you could get on games of pure chance in the past. If people want to bet on coinflips, it seems objectively better that they have access to a way to do that in a way where they only get fleeced for 1% of their bet rather than 5%+ of their bet.
For sporting events, for example, the alternative to prediction markets 5-10 years ago was to use a website where you bet against the house directly, and they'd usually take around a 15-20% spread, and they'd ban you and keep your account funds if they decided you're winning too much. Now you can bet on the same events on prediction market sites, with around a 1-5% spread, and the house doesn't care how much you win (so there's actually an argument that you're playing a game of skill, compared to the old format where you definitely weren't, since you'd be banned for being too skilled).