usef-
usef- on Hacker News
Yes. I do see people describe Kimi models as as open weights all the time, and they have somewhat similar licensing restrictions to this
> they seem to be the ones going around saying they're going to replace peoples jobs, and usurp the value of human labor.
I don't see why an alternative provider that is quiet but also scrambling to make AI smarter is any safer for jobs. Wouldn't we rather orgs speak honestly about the risks of the industry?
I find it odd that we talked for decades about companies being evil for covering up problems (chemical companies, etc), yet it turns out people hate the ones that talk about risks openly even more.
That's one issue from one distro 18 years ago. Have there been recent things? This thread is about a today's MacOS, and an issue known publicly for more than a week.
Yes, even though I'm ok with its use in many circumstances, if I can immediately tell it's AI that usually means the person was very lazy when making it. It's boring when everything looks the same.
Oh, I agree it's not ideal. But the kind of people who were willing to move to a new IDE are also the people who will now be first to ditch IDEs completely. The people who stuck to their traditional IDEs are likely the ones paying subscriptions.
I actually think they could do something more radical than Fleet today after seeing the (alpha, but impressive in use) ArtCraft apps, and translations like Bun. They could embrace still being the best traditional IDE, translating their trademark intelligence, with a lot of the heaviness removed. The thing that struck me about the ArtCraft apps was how fast and light their interfaces felt compared to the originals.
Nice idea. I've been thinking lately that the much-maligned genre of "edutainment" is going to come back tremendously now that games are cheap and quick to make for any concept you might want to learn.
Unlike normal game projects, they don't need to be as raw fun as a professional game, nor cover a popular-enough topic to be sellable to others, they just need to be fun enough to make you stick with the learning process for an afternoon.
Many of the improvements and refreshed design were "backported" into the traditional IDEs.
And fleet was made to attract the VSCode generation, who they were starting to lose new user momentum to. But VSCode is no longer the biggest threat.
It may or may not have, but it does match my experience that haiku is quite good at coding implementations on its own now
It seems pdf chose jpegxl, so they would have shipped with a decoder regardless. But, yeah.
What are you viewing with?
Note that progressive rendering is now supported by avif, and the comparison in the OPs post makes is appear quite a lot better.
It does sound like the remaining areas jpegxl is better at are almost all non-web use cases.
I'm not sure what the mistake was there? I thought their whole thing was judging orgs by cost:benefit calculation, which this seems to fit.
> They pick battles they can win and don't engage with ideas that really are challenging to their reasoning methods or priors
What's an example of one?
I've never seen someone describe the movement as infallible. They are pretty self critical if you actually read their writing.
If the translation is correct, Finland converted about 14,000 hectares of forest in the 2010s per year (without replacement), and cut about 173,000 hectares per year in general (with replacement).
I suspect how much people think this was an honest mistake with their 330 hectares (apparently following the forestry rules but missing EIA) will depend on people's priors.
1. https://ilmastopaneeli.fi/hallinta/wp-content/uploads/2025/1...
That's fine, but why not make it clearer that they're extremely early alpha?
Their websites look more polished than the apps themselves, have download options for non-technical users, and in no place that I can see (as of writing) do they mention that they are early development: https://getartcraft.com/apps/wordcraft
In "the before times" it was at least the norm to place at the top of a github repo and download section "this is early alpha" or similar to not let down potential users. This project currently seems like a great way to burn user willingness to try new apps.
Note that they're currently a buggy mess in my experience. That's not unusual for an alpha, but I'll personally wait to call them a gift until I see if the team is capable of polish+reliability. I do think their marketing is ahead of their product at the moment.
It also lost much of its quality advantage over plain jpeg, given the improvements in jpeg encoders: the jpegli encoder (from google's jpegxl team) has improved plain jpeg encoding significantly.
Most photo/design apps don't seem to be using SOTA jpeg encoders, though, so I think webp looks better in comparisons like OP's than it actually is. You maintain all the ultra-fast decoding of jpg when you use a modern encoder, too.
What would that achieve for their goals, compared to listening in to what they're saying?
It's impressive for its total size if you need local inference.
Though it's significantly slower in Token/s and also thinks a lot more without matching the same intelligence (xhigh qwen27b scores lower than haiku's medium setting, and haiku-med is $0.05 per task compared to Qwen27B's $1.01 on AA's comparison)
It seems like more a backup if you need to work offline, imo, unless time doesn't matter and/or your electricity is free. Or you want independence from the labs (fair enough).
I think they mean the new Haiku, which is mildly above Luna now . If you have a plan written by a smarter model (so the hard parts are solved) they can be great at implementation.
Maybe, but it's a reality that most of the general public do prefer cookbooks with pictures. I suspect openai would end up doing this anyway just by targeting the general consumer. It's configurable.
I'm more worried about picture accuracy: usually the benefit of pictures in recipes is to see what you're aiming at (eg, how finely chopped something is), but I don't know if the image model is up to that level of detail.
On subscription pricing a $20 Anthropic subscription gives >$500 equivalent tokens, which is not so different, and you get smarter models. API pricing has decent margins.
And Opus 5.5 is really good.
You're judging purely by token cost I assume, not cost per completed task?
The benchmark in the article showed it as lower per completed task than luna, but I guess we'll find out how representative that is. Anthropic has generally been fairly honest in their benchmarking though.
Yes, I would have believed it more if they were targeting one app. But doing all seven at once immediately made me think they likely had no level of care at all (or a giant team?). Today's AI can be impressive but it's not up to adobe-scale apps without a
lot of attention and feedback, if you have standards at all. There's a huge surface area to polish in just one of these apps.
I checked GitHub and it has only two committers and four weeks effort: One person +550k lines, another +3k lines.
They don't use fire extinguishers for lithium battery fires though (as the article mentioned: submerging it in special sand). It would possibly be a negative if a member of the public grabbed a visible extinguisher during a battery fire?
And I'm surprised that the only reports seem to be from eight years ago, given the amount of old electronics that go through the stores.
Maybe I'm misremembering, but do most luxury stores have fire extinguishers visible to customers (as opposed to staff)? I don't remember seeing them
There's usually a reason for making artistic choices, though. What do you think it is in this case?