I see. a brief google search didn't bring up anything in relation to the leading zero concept, but that helps. at a brief glance, their use of the leading zero seems like ... clever marketing?
my analysis of it is that it's a way of making people wonder "oh why is he writing it like that?" like I did, lead them to the foundation, and have them engage with it and be aware of it in the future; i.e. marketing. it's quite clearly not a practical thing. the probability that by the time 10000AD rolls around we're still using the same year system, we're still alive as a species, we're still technologically capable as a species, and we don't have the capacity to understand older years minus the leading zero seems near enough zero to be zero. call it what you like, marketing, inspiration, whatever, but it's a sneaky way of leading people's thoughts onto a particular pathway, which I call marketing
to be clear, having read through their website, I think what they're doing is great, and this isn't a criticism
I was born in the 20th century. I was filling out a medical form and put my birth year in with 2 digits. The web app took that to mean just the 2 digits. No, I wasn't born in AD 70.
That's the kind of programming that makes you reluctant to put anything into it.
Not using leading zeros seems fine if you're using AD near it to indicate that it's not 1942.
The "fair use" part takes a lot of place in this article.
It talks a lot about what happens if you use more tokens than what you're allowed, but curiously doesn't pip a word about what happens if you use less - for example maybe with a partial rebate on your next billing cycle ?
I think "fair" should mean "fair for all parties involved", currently it's rather a "we don't want to incur any risk" policy, since I don't see how it's fair for my end of the contract.
I'd rather pay for my actual usage at any other provider than pay for min(actual usage, 25$) at Kagi.
It’s the exact opposite. They are incurring a huge amount of risk with this.
6 hours ago most users didn’t have access to this feature at all. Now we have $4-8 of raw token credits a month to use on a well-built feature.
I’m paying $9 a month with the annual subscription, and it was worth it just for Search. Now they’re giving me $17 worth of value for the same price.
Their margins must be razor thin, and they’re only able to offer this much value because they’re counting on most people not using all credits. If everyone did, or if they gave rebates, they’d go out of business.
The other point of view is that they are now forcing users to pay for both search and AI, even if they do not want to use or fund the development of the later.
You used to be paying 9$ for search, now you are paying 9$ - x for search and an unknown amount for AI.
I am also an existing happy subscriber to Kagi. I currently pay for unlimited X. Now I pay for limited X, where I can't even see if I'm approaching limits or not. Anyway, my main point is that I'm getting LESS for the same money.
I'm going to take a charitable interpretation of your post and assume that you have simply misunderstood Kagi subscription tiers, and you have mistakenly assumed that I was on the first tier. I am not on the first tier. I am on the tier which they continue to advertise with the words "Unlimited Kagi Assistant". That is the product I am paying for. Now, instead of getting "Unlimited Kagi Assistant", I am getting "Limited Kagi Assistant".
I'm not paying for search, I'm paying for Unlimited Assistant. No, they did not "have some limits". Or at least, the product they sold me was very clearly labelled "Unlimited". The word "Unlimited" is literally in the name, and it was described as such in more wordly descriptions.
I'm not going to argue about your interpretation of the word unlimited and wish that its use was either banned or strictly enforced in this (and similar) contexts. That said, until accountability is legislated and enforced, it is not reasonable to assume that unlimited means unlimited. Just as marketing abuses the word, customers abuse the concept. That's especially true in a domain where processes can be automated and distributed, by some means or others.
> it is not reasonable to assume that unlimited means unlimited
sure, but I am not abusing the unlimited plan. it is reasonable to assume that "unlimited" means that a normal user like myself will not run into limits.
The article claims that 95% of users won't be affected when enforcing the policy. Assuming the claim is true, not affecting 95% of users is pretty much the definition of not affecting normal users.
That is totally unsurprising. I would be curious to see what they are doing to achieve that. Likely automating something and not using it like a typical human would.
Maybe your usage might not be affected, I propose to check for a month and if it really goes over the limit, you have the option of canceling the subscription.
It definitely is worded in a way that it can either be interpreted as (unlimited searches and unlimited premium ai) or (unlimited searches and premium ai).
It also may not help that they did not enforce the fair use policy until now. At least that is what I read out of their blog post.
But the fair use policy has been included a long time. I checked 2023-12-25 and found it there, might be available earlier, but no interest in looking harder.
They advertise with the phrase "Unlimited Kagi Assistant", so I'm not sure how you can interpret that as "unlimited searches and premium ai". Kagi Assistant refers to the same thing as "ai" here. It's a different product than Search.
"As an existing happy subscriber" -- goodness! Even after years of seeing what happens after such illogical evangelism, hn never ceases to surprise you and brings the fandom out in full force. It's like with Apple. You say once "your phone switches on and off on its own - maybe something is amiss with the hw/sw" and there are dozens of replies already blaming you "you must be holding it wrong!". It's a whole new level of apologism.
I don't follow what either you or the OP are upset about. Where is the apologism and why do you think it's needed?
Kagi rolled out a free feature to its existing customers without increasing the price of their plans. The limits of that plan seem quite generous, as well. The only way I can make sense of the OP's post is that the OP wants the Kagi subscription price to decrease. Perhaps that's fair, but it doesn't make any sense here because you're strictly getting more for your money. If you're paying $10/month for a subscription, yesterday you couldn't use Assistant and today you can for the same $10/month. Placing a cap on how much you can use seems quite reasonable given the service costs money to operate. If you choose not use it, you're no worse than you were yesterday... you can happily go about using the search service you were already paying for.
Is the problem that the free usage isn't unlimited? Is it that not using the free service doesn't reduce the search price? Or is it that those using Assistant more than you appear to be getting more value for their money? I'm not trying to be dense, but I really don't see what's even remotely controversial about this announcement.
Lately, every subscription I have is increasing the fee without giving me anything. Kagi is giving me something extra without charging me any more money. I'm sure the nefarious intention is to make their service more attractive to non-subscribers and grow their userbase, but interests can align.
It’s probably worth noting that they do offer a refund of your subscription price if you don’t perform searches in a given month, which is pretty much the opposite of how every other company with a subscription works.
It wasn’t a day one feature, so there’s some chance that a thing like this could roll out.
The article probably focuses on the overage because that’s what most users are going to be concerned about.
Few other companies seem to try to do things in the interests of their users and balance that against making enough money to keep existing.
Huh? The title of the blog post is "Kagi Assistant is now available to all users!". Their users are people paying for what up until now was just their search service. They're now rolling in Assistant as a value-add. Your subscription price didn't increase. You're strictly getting more for what you were already paying. If you don't use it all, you're no worse off than you were yesterday.
If you want metered billing, there's no shortage of AI services that offer that option. Kagi even offers one by way of the FastGPT. You can also pay to use their search API if you don't think the subscription is worthwhile. You can cobble something together with Open WebUI pretty easily.
I have Kagi Family plan for my household. I've been paying for the Ultimate upgrade for my account in order to access Assistant, but given how infrequently others in my family would use it, it never made sense to upgrade them. Still, it would have been convenient if they could occasionally access Assistant. And now they can. And my bill didn't increase. And they're being incredibly transparent about what the limits are and why they're there. I'm a really happy customer today.
As an early adopter I first got forced off my grandfather plan to the regular one(at least I got a T-shirt). Now I have a limited number of searches that I have to keep track of and this has made me only use Kagi if necessary. This has dropped my number of searches significantly but at the end of the year I’m still being charged for renewing my plan even though I haven’t used a quarter of my allotted searches.
I don’t care about LLMs so this brings nothing of value to me. Give me an email account or some backup storage and open source office suite and I would be willing to pay and pay more.
I’m seriously considering not re-newing my subscription for the first time in ages.
Then don't renew. Nobody is forcing you to pay for the service, and from what you wrote, it sounds like the service is not what you need/want (anymore).
It's a good idea, but from the docs it looks like the high level abstractions are wrong.
If my data pipeline is "take this table, filter it, output it", I really don't want to use a "csv file input" or a "excel file output".
I want to say "anything here in the pipeline that I will define that behaves like a table, apply it this transformation", so that I can swap my storage later without touching the pipeline.
Same things for output. Personally I want to say "this goes to a file" at the pipeline level, and the details of the serialization should be changeable instantly.
That being said, can't complain about a free tool, kudos on making it available !
Hey, not sure I get your point here. I believe the abstraction provides what you're describing. You can swap a file input with a table input without touching the rest of the components (provided you don't have major structural changes). Let me know what you meant :)
Which does the 100x speedup too and is a "safe" way to adjust memory access to numpy strides.
Whether the output is correct or not is left as an exercise to the author, since the provided benchmark only use np.zeros() it's kind of hard to verify
I just measured this with the np.ascontiguousarray call (inside the loop of course, if you do it outside the loop when i2 is assigned to, then ofc you don't measure the overhead of np.ascontiguousarray) and the runtime is about the same as without this call, which I would expect given everything TFA explains. So you still have the 100x slowdown.
TFA links to a GitHub repo with code resizing non-zero images, which you could use both to easily benchmark your suggested change, and to check whether the output is still correct.
I wonder why are the example videos this specific clip compilation format.
It feels to me that to navigate that, you essentially have to index 500 10-seconds videos, and that looks a lot easier than retrieving information that is in an actual 1 hour long video, because the later one will have a lot more of easy to mix-up moments. So maybe it hides an inability to answer questions about actual long videos (in the paper, the other example videos cap at 3 minutes length for what I can see).
On the other hand, maybe it's just for results presentation purposes, because it is much more readily "verifiable" for everyone than saying "trust us, in this very long video, there's the correct answer unarguably".
So if someone happens to more about that, I'd be very interested
The planet has never been, is not and probably won't be in any near future in peril. Humanity is. Humans disappearing is the concern of ecology. The planet will adapt just fine with or without us, and species disappearing will be replaced in time by new ones.
Not having children is not effective at perpetuating the human race, and thus not usually considered as an ecological solution.
That being said, you don't have to have children to be an ecologist either, but you certainly should realise that your efforts are directed towards future humans, not towards the planet which does just fine anyway.
Here's a conclusion that you can also take from the same article :
"In the UK, ages 1-59 year olds are dying at almost eight time the rate if they don't have any baby tooth left and don't absurdly love dinosaurs".
The fact that people have to actually write the whole data analysis down is mind boggling, because it's so trivial to understand the reason.
Still, this is the conclusion :
"Thus, the results we see in the actual UK all-cause deaths for fully vaccinated and unvaccinated is not unexpected, and can be fully explained by the Simpson's paradox artifact since the observed ratio of vaccinated:unvaccinated all cause death rates, 1.82x in week 30, is less than the expected background ratio of 2.41x based on their disparate age distributions."
That's not "a lot of hand waving to try and explain it away". They do a very methodical and sensible mathematical analysis of the data at hand, including the initial data point, to explain a very simple thing : people tend to die older overall, and older dying people tend to vaccinate more.
If you translate (instead of ass) cul as "bottom", it is more like the "bottom of a sack" (i.e. no-exit), if your interpretation was correct, it would be a no-cul-de-sac.
Is focusing on github stars specially meaningful ? Is there any functional difference for the repo between having 6k stars (as of today) and 54k ? I don't think so honestly.
Github stars are just not something useful to monitor closely. I had a look at my starred repos on github, 95% of those I have no idea what it is anymore anyway.
So yeah, github didn't do any extra effort to restore what's imo essentially a vanity metric. Makes sense to me ?
