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

A lot of meaningful things are difficult and laborious, but not all difficult and laborious things are meaningful.

Not that crazy. The cost of storing the film—even 2 hour features—is dwarfed by the rest of the production costs. You can afford a dedicated HDD when you're done.

Well, hopefully two.

This will at some point in the future invert.

The cost to generate a future kind of film from some template (script, characters, art choices, etc in some kind of source file) won't be much more than the cost to store it.

When this happens, perhaps we will cache the results but later dump them. Assuming storage costs don't drop faster and more significantly.

Maybe 30 years?

Edit: Lots of downvotes. I'm a filmmaker, I've made lots of photons-on-glass films. Most of us are experimenting with this tech and aren't thumbing our noses at it like people outside our industry. We don't really have a choice but to adapt, and I find it funny that casual observers on the outside are so morally opposed. It's actually an incredible tool for pitching and has utility for some SFX, compositing, and B-roll shots today. It's really going to help mid market and below, for films that don't have Disney budgets.


The thing about the ludds is that as a rule, they don't understand that their perceived enemies are threatening to make their lives better.

They can be safely ignored... at least here, and at least for now.


If this (tragic, dystopian, abominable) future comes to pass, why would we store movies at all?

Just distribute the prompt and I'll generate my own movie on the fly, with my own tweaks of course.


It's funny to me that I work in film and have more hope and imagination over the use of this technology than people who (presumably) don't make films at all.

As long as humans have dreams it won't be like that. The human spirit and desire to connect to others and tell stories doesn't just suddenly die.

I think the very best lens to look at it is that all of the tens of thousands of kids that go through film school and never get to bring their VFX-heavy fantasy to life now suddenly have voice to match their ambition.


The downvotes are a good sign. If AI didn't promise massive artistic disruption -- the sort that threatens to put real creative power into the hands of outsiders -- no one would object.

Look at the history of photography itself to see an example. "But... but... but my portrait-painting skills will be obsolete! Waaah."


> "The outcome would eliminate jobs, push down wages, worsen conditions for all entertainment workers, raise prices for consumers, and reduce the volume and diversity of content for all viewers."

As if this weren't going to happen anyway...


It won't be relevant in a couple years when 90% of sites will be using ECH, meaning the SNI will be encrypted as well.

Just enabling ECH doesn't stop this, firewalls can see it and mangle the data to force a downgrade because most servers need to support older protocols. It's more accurate to say that once sites only support ECH, then they'll be forced to stop downgrading or deal with angry users.

TLS 1.3, including the ECH extension, does not permit downgrading, unless your implementation is broken.

Trying to downgrade or strip extensions from any TLS 1.3 connection will simply break the connection.


In the wild, that's not true at all[0][1]. The corporate firewall at my employer actually wasn't able to block ECH until they updated it then it was able to block sites as usual.

0: https://community.fortinet.com/t5/FortiGate/Technical-Tip-Ho...

1: https://docs.broadcom.com/doc/symantec-ech-whitepaper (see page 8)


I ready the FortiGate link and this is the gist:

  The DNS filter setting on the FortiGate analyzes the DoH traffic and strips out the ECH parameters sent by the DNS server in the DoH response. If the client does not receive those parameters, it cannot encrypt the inner SNI, so it will send it in clear text.
So basically they mess with DoH ECH config and trigger fallback behavior in the clients. I don't think any browsers do this yet but I think this loophole is not gonna last.

I'm surprised that works. Doesn't TLS1.3 do the thing where it crosschecks (a hash of) the setup parameters after key-agreement to protect against exactly this kind of downgrade attack?

(My phone screen is too small to look through the RFCs right now.)


I think what you're describing is TLS1.3 Finished verification so that happens after DoH response during the actual handshake. Basically this works because ECH is fairly new and there's no HSTS-style "always use ECH for this site" configuration yet. And ofc this only works if you configured FortiGate as your DNS (corp network) or if it's doing MITM (though I'd expect browser would verify cert fingerprint for DoH connections as well).

This is literally impossible. What your corp fw likely does is mitm outer SNI because your IT department installed your company CA in every client's trust store. So unless you do that at national level your only other option is to block ECH entirely.

Edit: actually totally possible but you need build quantum computer with sufficient cubits first =)


Last I heard, the QCbros were still trying to find the prime factors of 15.

(I remember using quantum algorithms to find prime factors 25 years or more ago, using the Quantum::Suppositions Perl module.)


Is there even a push for ECH? I don’t imagine big tech and other powerful players particularly want it.

Cloudflare and all the major browsers have supported it for a couple years now.

Computers have supported dnssec for years and it’s barely in use. Support != a push for.

This is exactly the case. If I don't like someone, I think about them as little as possible. Elon doesn't care if people whine about him, so you don't gain anything from fixating on it.

> this secret be easily reverse engineered / inferred by having access to models weights

It could with a network this small. More generally this falls under "interpretability."


Evil eBPF programs can hide their presence from the bpf syscall as well.

Interesting. Any good read you'd recommend on the topic/attack? Thanks.

Look up "eBPF rootkits"

This is a good article about one found in the wild: https://www.synacktiv.com/en/publications/linkpro-ebpf-rootk...


Quoting a previous comment of mine:

Ignoring the fact that if AI training is fair use, the license is irrelevant, these sorts of licenses are explicitly invalid in some jurisdictions. For example[0],

> Any contract term is void to the extent that it purports, directly or indirectly, to exclude or restrict any permitted use under any provision in

> [...]

> Division 8 (computational data analysis)

[0] https://sso.agc.gov.sg/Act/CA2021?ProvIds=P15-#pr187-


Claude is unique in the way it falls into this pattern. It's done it since at least Claude 3.

Dr Bronner's made it into the training data.

Ignoring things such as emergency services being blocked, imagine if protesters could "block" the Internet.

That would not garner much sympathy.


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

Search: