Hacker Newsnew | past | comments | ask | show | jobs | submit | lja's commentslogin

You would need thousands to tens of thousands of images, not just 50 to produce an adversarial network that could use the API as a check.

If someone wanted to buy it, I'm sure reality defender has protection especially because you can predict adversarial guesses.

It would be trivial for them to build "this user is sending progressively more realistic, rapid responses" if they haven't built that already.


I really think a "OpenAI Me" is what's needed.


That's because this product isn't for you then. My team has been evaluating vector databases for years and everything on the VectorChord page resonated with me. We run one of the world's largest vector databases and we'll likely benchmark vectorchord to see if it lives up to its promises here.


Hi, we’re here to help if you need assistance (via GitHub issue, Discord, or email). Could you let us know the scale of your vectors—are they 1B or 10B?


10B, feel free to email me luke@ the domain in my profile.


Do you have a blog per chance? Or any recommended reading on pre processing / data chunking strategies to improve results?


I recently came across a project that looks promising: [WordLlama](https://github.com/dleemiller/WordLlama?tab=readme-ov-file#s...). It appears to be well-suited for semantic chunking, though I haven’t had a chance to try it out yet.


Maybe check this https://zilliz.com/pricing

Yan easily store 1B data into Zilliz Serverless and the cost is incredible cheap


You cannot compare Serverless with dedicated instance, this is unfair. A large number of writes and queries will eat up your wallet.


That's true. Now we can offer the new dedicated Cost-optimized CU,pricing is

$0.318/hour ($228.96/month)

Which means you can store 20M 768 dimension data in 228$ per month


Zilliz gave us a near 6-figure a month quote for our database.


"It's funny but it doesn't apply to me." - most of the people in tech


That's true of a lot of standup comic jokes that rely on generalizations. Because they're generalizations.


Tech is pretty diverse these days, so that isn’t hard to achieve. If they lean on big corporate tech, though, I think that can provide a lot of relatable laughs. But really, that’s already true without a roasting comedian (like with Trump, the memes write themselves).


To be fair, not _just_ in tech. :)


Apply in what way? That John Mulaney could write a joke about it? He has literally dedicated his life to that skill. Of course he can.


Any OF models would be met with HN users over-explaining their own economics to them and how it's a terrible business that'll never work. These models will also learn they don't even have a moat to differentiate themselves from other offerings and should keep their development jobs. :)


I don't think people discovering these profiles on HN is the concern here.

The true problem with OF models is due to an iterated mutual tit-for-tat strategy between OF models and popular groups on social media platforms:

1. OF models (or people acting on their behalf) want to promote themselves using popular groups/pages/channels on social-networks — they spam posts to these places, seeming to be authentic engagement, in ways that get people curious to look at their profile; and where their profile on the social network then directs those people to their OF profile.

2. The popular groups/pages/channels on social networks are inundated by spam from these OF models, and so attempt to use automated measures to detect and block posts from posters who link to OF on their profiles.

3. OF models/their agents try to work around this by indirecting their OF profile behind "make an About You page with links" services like carrd.co.

4. The popular SN groups respond by also blocking profiles containing links to these "About You page" services (because, keep in mind, the SN profile already works as an "About You page", so there's no need to link to one of these external "About You page" services — you could just put the same links you'd put on such a page into your SN profile instead. The only people who do link to "About You page" services from their SN profiles, are OF models.)

5. And OF models/their agents try to work around this, by finding ever-more-obscure "About You page" services — and/or profile pages on other, more obscure social-media services, to use as an "About You page" — to get ahead of this moderation.

This at.hn service would sadly be exactly the kind of service referenced in step 5.


I used to enjoy visiting link.tree links to find ways to support content creators but now it’s a way to disseminate their OF spam


I am wondering why we don't see more OF models using custom domains instead. $10/yr shouldn't be too big of a price for them. Perhaps those About You page services don't support that?


I'm guessing it's because they think that even with a custom domain, some deep headless-Chrome-scraper-based OF-link scanning (or just human flagging!) will catch on and get them blocked within a few weeks anyway. Which would make the $10/yr for a custom domain, more like $10 per time-you-get-caught.

It's the same reason that scammers/spammers rely on these services: the only hosting you can "acquire scalably" is free subdomain hosting.

(Now, why they don't use something like a Google Doc, that can't be blocked based on its URL and can't easily be text-extracted by a scraping bot? I have no idea.)


low tech savvy-ness


> tit-for-tat strategy between OF models

groan.


agree completely except the last sentance.. smells like a cheap backdoor to auto-profiling plus shenanigans immediately following that


I can see both groups working with models and curves

Just different types

https://www.youtube.com/shorts/AE4IxYq4nig


This is a punny comment


I think you underestimate both HN and OF users.

If this post is any indication, part of HN would build tooling, while the rest would vote it up.


I think it's important to contemplate the human component of mergers. If lots of gainfully employed people get let go so that fewer people can consolidate expenses and raise shareholder value I think it should be considered as part of the decision making framework.


It is a consideration; redundancies are eliminated in acquisitions. It's capitalism; not laborism.


The taxonomy of AI is the following:

AI

-> machine learning

...-> supervised

........-> neural networks

...-> unsupervised

...-> reinforcement

-> massive if/then statements

-> ...

That is to say NN falls under AI but everything falls into AI.


Where did you pull this “taxonomy” from?


This looks so much like the bin packing problem [0] if you generalized it you could do a lot with it.

[0] https://en.wikipedia.org/wiki/Bin_packing_problem


Yep, I'm targeting an optimization engine at a more specific problem that might be useful to lots of consumers.


Would you make that open source? I'm interested in this problem. I'm sure there are some OS bin packing algorithms but I not familiar with ones that have separation and linking rules.


I'm using Optaplanner underneath, which is already open source.

https://www.optaplanner.org/


There is a lot of DMCA hate in these comments.

I run a DMCA take down service where we remove revenge porn for people that are victimized by it. DMCA is the only tool we have to take these images down so as much as there are holes and gaps in that law, it's not all bad.


DMCA is a corrupt, offensive attack on consumer rights. Just because it can be used for laudable purposes doesn't change that.


Not saying it makes DMCA perfect, just wanted everyone here to know there are some really awesome, clever uses of DMCA that people use for good and if we torch the DMCA we should make sure we don't leave those folks out in the cold.


Thanks for sharing the use case. However using it to defend the entire DMCA sounds like a false dichotomy. Surely there's a way to make rules only for those "awesome" uses, so that they don't have to be "clever" uses. C stands for Copyright and it's mainly used that way in practice.


Totally agree with everything you said, I just want to remind people that half of the story of DMCA is that we have really poor controls over our privacy and ironically DMCA is one of our only tools.

It does have many, many broken aspects.


I don't really see how this is helping, you're just dragging your service down by getting it involved with this for no good reason.


Given the homogenous demographic of HN, it’s safe to say we aren’t losing any business from here. There is nothing to drag down.


I applaud you for using the DMCA for good. You're doing good work, and please keep doing that. That doesn't mean that the DMCA can't also be used for evil, though.

It would be nice if it could only be used for good, but maybe that's impossible. It may also be possible to replace it with something more specifically tailored to take down revenge porn. I certainly think revenge porn is far more serious than a mere copyright violation.


Ah, a service predicated upon the Streisand effect. What could possibly go wrong?


They are saying, "We are going to make our existing managers work harder, make the existing engineers deliver more because the $22 billion in profit we made in 2022 wasn't enough. We should consider how we can sacrifice more to make even more profit."


No I think Zuckerbergs words should be read at face value.

Managers with small amount of reports will be converted to ICs and strong managers will be given more reports. The org will be more lean and will have less overhead per manager of converting work between layers.

I’m summarizing above but in the article are specific goals to address this and I didn’t pick up on any signs that employees should overwork themselves.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: