If you meet someone who understand business, it's going to color your opinion of other people who meet who seem to understand business, until you either learn business or die.
That's how all prejudices work: We're wired up to be afraid and anxious and to share (and communicate) our anxieties to our friends and neighbours. We're trying to help.
The thing is, knowledge; business, economics, and so on, probably can be used to help people, but in a lot of peoples' recent memories, it's being used to harm.
I lost a lot of my teen friends when I "went corporate", but thirty years later I'm reconnecting with some of them, because people change, and we can learn to recognise someone will participate in capitalism for lots of reasons that are not so simple as being "greedy" or "evil".
But to me, I think it is simple: Capitalism is almost certainly unavoidable, so the world could be better if more kind people participate well in it than don't!
> So if you get your target to record (say) 1 hour of audio, that's a one-shot.
No, that would still be zero shot. Providing inference-time context (in this case, audio) is no different than giving a prompt to an LLM. Think of it as analogous to an AGENTS.md included in a prompt. You're not retraining the model, you're simply putting the rest of the prompt into context.
If you actually stopped and fine-tuned the model weights on that single clip, that would be one-shot learning.
> Providing inference-time context (in this case, audio) is no different than giving a prompt to an LLM.
Right... And you have 0-shot prompts ("give me a list of animals"), 1-shot prompts ("give me a list of animals, for example: a cat"), 2-shot prompts ("give me a list of animals, for example: a cat; a dog"), etc.
The "shot" refers to how many examples are provided to the LLM in the prompt, and have nothing to do with training or tuning, in every context I've ever seen.
> Right... And you have 0-shot prompts ("give me a list of animals"), 1-shot prompts ("give me a list of animals, for example: a cat"), 2-shot prompts ("give me a list of animals, for example: a cat; a dog"), etc.
> The "shot" refers to how many examples are provided to the LLM in the prompt, and have nothing to do with training or tuning, in every context I've ever seen.
In formal ML, "shot" refers to the number of samples available for a specific class during the training phase. You're describing a colloquial usage of the term found only in prompt engineering.
You can't apply an LLMism to a voice cloning model where standard ML definitions apply.
> If a state actor were trying to intercept traffic (MITM), the last thing they would do is pad the AS path
That's presumptuous: A state actor would (and could trivially) pad the wrong directions to flow traffic down to pops that are not making new announcements (and thus not-implicated by cloudflare and other "journalistic" efforts).
There's also a lot between fat-fingers and deep-state: I know of some non-state actors who do this sort of thing just to fuck with ad impressions. I also doubt much usable intelligence can be gained from mere route-manipulation thing, but I do know that if it is a fat-finger, every techdude in the area was busy at that time trying to figure it out, and wasn't doing their best work twelve hours later...
> most likely a route map intended to manipulate traffic engineering for their own upstream links
...that being said, this does seem plausible: Most smaller multihomed sites I've seen (and a few big ones!) have some kind of adhoc health monitoring/rebalance function that snmp or something and does autoexpect/curl or something-else to the router to run some (probably broken) script, because even if your uplinks are symmetrical, the rest of the Internet isn't, so route-stuffing remains the best way to manipulate ingress traffic.
> Never attribute to malice that which is adequately explained by a missing export filter.
As soon as I peer with two big sites that don't peer directly with each-other, they both gotta let me forward announcements unfiltered across them. Once I have a third, I have a legitimate need to manipulate my own ingress.
The problems with the BGP are legion, and not just one thing that prevents BGP and security from sharing time in a sentence.
> A state actor would (and could trivially) pad the wrong directions
This isn't how BGP works. An AS-PATH isn't the path the traffic will follow; it's the path that this overall announcement has allegedly tranversed and is (one of many attributes) used to judge the quality of route. The next hop tells our peer where they should send the data if they like this route.
Putting more things in the AS path makes the route less attractive. Leaking a new route isn't going to magically make some other route become more preferred.
Lower cost usually means lower quality and is an example of how a long path being leaked can result in traffic flowing away from high quality path to the leaked path.
Not saying that this is the case with Venezuela, just explaining the reality of BGP where path prepends are often ignored.
> > > That's presumptuous: A state actor would (and could trivially) pad the wrong directions to flow traffic down to pops that are not making new announcements
> > Leaking a new route isn't going to magically make some other route become more preferred.
> Not magic, but technology can look like magic when you don't understand it.
Please let me know of the scenario where route A is preferred, undesirable, long-path route B is advertised/leaked, and as a result traffic flows over route C.
I've used BGP for over 25 years, so I'm really curious what you're thinking. Or if you're describing something else, you're being really unclear.
Or if you're just describing withdrawing a route and replacing it with a really undesirable route -- sure, we do that all the time. But that doesn't match this scenario and isn't going to get flagged as a routing anomaly.
You know what's really toxic? Not explaining what you mean and just sending some introductory lab documentation about what the other person has already clearly shown they understand.
I don't even know what you mean by a lot of these things.. e.g.
> > > As soon as I peer with two big sites that don't peer directly with each-other, they both gotta let me forward announcements unfiltered across them.
A straightforward reading of "forward" doesn't work for this sentence. I should not take a route from peer A and send it to peer B. Peering isn't transitive. If I try, it should be filtered.
Peering means to give your own routes (and your transit customers' routes) to someone else. Not your other peers routes.
> Please let me know of the scenario where route A is preferred, undesirable, long-path route B is advertised/leaked, and as a result traffic flows over route C.
> ... I'm really curious what you're thinking
That the actor actually wanted the traffic to flow over route C.
> You know what's really toxic? Not explaining what you mean and just sending some introductory lab documentation about what the other person has already clearly shown they understand.
I think perhaps you and I have different ideas of what is "clear", for example when you said something that is totally covered in introductory lab documentation, I thought it was clear that you did not understand.
> I don't even know what you mean by a lot of these things
That is clear! But confusing! How can you clearly understand but not know what I mean?
> Peering means to give your own routes (and your transit customers' routes) to someone else.
That's exactly what's happening here: Not every transit customer peers with every other transit customer.
> > Please let me know of the scenario where route A is preferred, undesirable, long-path route B is advertised/leaked, and as a result traffic flows over route C.
Yes, but how does advertising undesirable route B make traffic go over route C? This is why I think you're confused.
> That's exactly what's happening here: Not every transit customer peers with every other transit customer.
I am not understanding what you're saying at all. You said:
> > > > As soon as I peer with two big sites that don't peer directly with each-other, they both gotta let me forward announcements unfiltered across them.
This is the thing you are supposed to never do as a peer, and the thing that I have a whole bunch of filtering to prevent my peers from inadvertently doing.
Are you misusing the word "peer"? It's hard to talk about BGP and routing policy without using these words correctly.
> I am not understanding what you're saying at all.
And that is why; You seem to have a very strong opinion about something that you don't understand "at all" and frankly I cannot understand how that can work.
> This is the thing you are supposed to never do as a peer
So you say, but that's what I did when back in the early 2000s, and that's what the parties in the news were doing, and if you're not totally lying to me, you know this because it's the default in BGP, that's why you would say you need to:
> I have a whole bunch of filtering to prevent my peers from inadvertently doing.
because that's how BGP works. Duh.
> It's hard to talk about BGP without using these words correctly.
and I am flabbergasted you continue to persist at it, when I have even offered you "introductory lab documentation" to help.
Peering means "give our downstream customers' routes plus our own routes; receive the same from them".
Transit means "give our entire table, receive their routes plus their downstream customers routes".
You don't give one peer's routes to another. You filter to make sure you are not doing this. They hopefully filter (using data from RIRs) to make sure you're not doing it. If both parties screw up the filtering, you "leak routes" like we're discussing here.
This has been standard practice for peering since at least 1997. It is codified, among other places, in RFC7454.
> And that is why; You seem to have a very strong opinion about something that you don't understand "at all" and frankly I cannot understand how that can work.
Do you operate an AS? Are you a peering contact? I mean, I only do it mostly for funsies now but for quite awhile that was part of my job. :P
Also, still seeking an answer to this question:
> > > Yes, but how does advertising undesirable route B make traffic go over route C [that previously went over route A]? This is why I think you're confused.
I set up multihoming in the US (going through ARIN assignment for ASN and PI) in the early 2000s and for another larger company in the UK (doing the same same but different) in the early 2010s.
> Also, still seeking an answer to this question:
Not sure what to tell you. I've answered this within the context of the news article, if you're asking specifically what kinds of configurations do that they're the kinds that are in that "introductory lab documentation" and if you're not overstating your credentials you should be able to understand.
OK, so you've never actually been involved in peering between providers. That explains things.
If you are buying transit for multihomed site, you just mostly need to worry about advertising your few local routes and receiving a table.
These words mean specific things in a provider environment. Peering between providers is not transitive. You should not give a peer's routes to another peer-- else you are offering to provide transit to that peer.
None of the three parties involved (the peer the route came from, the peer the route would go to, or you), generally, want this.
> As soon as I peer with two big sites that don't peer directly with each-other, they both gotta let me forward announcements unfiltered across them.
So -- this is true for neither scenario. If you buy transit from two people, you don't want to "forward" routes between them. Likewise for two peers. You only want to pass on your own routes, plus routes for your downstream customers or to your downstream customers.
The other topic that caused confusion: leaking a route to somewhere won't change where traffic goes unless that new leaked route wins. If it wins, the traffic will come to you (regardless of AS path). If it doesn't win, the existing route will still be used.
I'm retired-ish now, too (now I'm mostly a high school teacher for the funsies). But if you find yourself with BGP infrastructure in HE.NET Fremont, or somewhere that can peer with FREMIX, or somewhere that can access KCIX, feel free to send me a peering request. I currently have 2 direct upstream transit providers and 55 peers across 3 sites and export 6 prefixes plus 2 downstream prefixes :P
Making a dumb iphone game is a good excuse to send random HTTPS traffic, but receiving packets encoded as emails via (say) IMAP would be a good way to bring back large(ish) amounts of data.
Someone watching closely might try to correlate the strangeness of the emails you receive with your candy crush habits...
There are quite a few programmers who say lisp led to early retirement. That was a pretty interesting idea to me. I like going to the beach a lot.
I am not so sure about people who don’t want to get done: if you like doing what the ticket says instead of the other way around lisp probably isn’t going to be something you’re interested in.
"Lisp makes people rich, and I love being rich. Using Lisp actually can't help but make you rich. But I can't actually provide any examples of that happening or how they might translate to anyone else. Get so rich with Lisp. Lisp."
You can use a better tool than someone who is using a worse tool, and take their customers, because it will be easier to solve the problems better and faster than that someone-else. Lisp is a great tool, but it doesn't help you if you're the tool: You still need someone with a problem who can pay for a solution, and they need to like you enough to pay you for that solution. If you need to hire 30 people to figure that out, Lisp isn't going to help with that problem.
And it's not fast: Lisp is not a get-rich-quick scheme, it's a tool that you can learn how to use to make other tools, it's a tool that is difficult enough to use that you will need perhaps years of experiences in order to use effectively, it's a tool so advanced it might seem like actual wizarding magic just watching someone screencast if you don't know what you're looking at, but it's still just a tool.
I am literally asking you to show me what you are talking about.
Who retired early? What did they make with Lisp? Do you have a link to that thing they made? Was it actually something special about Lisp or did they just happen to use Lisp while making something that could have just been as easily - or more easily - done with something else? Is that real-world, specific, material outcome replicable by other people and they should know about it?
> I am literally asking you to show me what you are talking about.
Do you see these things -> ? <- Those are question marks.
Now go look back at your last comment and tell me how many of them you see.
> Was it actually something special about Lisp or did they just happen to use Lisp while making something that could have just been as easily - or more easily - done with something else?
I don't know. Why are you asking me this?
Please read my comments carefully, because I put time into trying to tell you something, but I am not trying to tell you what I think you think I am trying to tell you, and I am not going to justify my opinions to you any more than I would ask you to justify yours.
> and they should know about it?
I have mixed feelings about this: Who are the other people you are talking about?
There are a lot of people I won't work with; I won't trade with them; sell them my products; buy theirs; No contact if I can help it. That's why the first question I asked you is what is important to you? Did you even see my question?
> the only way to share these with a wider audience is to convert them to a slideshow and make a Rec.2020 HDR movie that I upload to YouTube
i understand some of this frustration, but really you just have to use ffmpeg to convert it to a web format (which can be done by ffmpeg.js running in a service worker if your cpu is expensive) and spell <img as <video muted autoplay playsinline which is only a little annoying
> I can't politely express my disgust at the ineptitude, the sloth, the foot dragging, the uncaring unprofessionalism of people that get paid more annually then I get in a decade who are all too distracted making Clippy 2.0 instead of getting right the most utterly fundamental aspect of consumer computing.
hear hear
> If I could wave a magic wand, I would force a dev team from each of these companies to remain locked in a room until this was sorted out.
i can think of a few better uses for such a wand...
> writing a blog post every once in a while will not provide meaningful income
Should people receive meaningful income for writing a blog post every once in a while?
I feel like that's the real question and not everyone agrees on the answer
> we would be better off with free content like on the internet of old
Well as someone who was there you used to need meaningful income to use the Internet of old. Nowadays everyone needs the Internet and it's a pretty big expense in most peoples' budget, and I think that's why so many people are willing to try something at it,
I figure if you just gave everyone meaningful income we could have that again
> I want to make sure that content creators are compensated for their work. Ad firms that employ fingerprinting stand between me and the content creator.
This is false: We're the ones who pay the creator, because:
> I'm not going to pay $5/month for every blog that I occasionally read
If that upsets you, please understand it upsets me to, because
> but at the unacceptable price of privacy
I want you to consider a different toothbrush brand, or maybe a hot location for a holiday, and the idea that I am "invading" your privacy in trying to do this is disconcerting.
I understand there are actors who want to use your private personalising data to harm you. I think that is bad, but I am telling you friend, that isn't me.
> I'm not quite sure what the answer is.
Listen, as an insider I am not quite sure what the answer is either, but I'm telling you that content creators need to eat because you have threatened them with capitalism which murders you if you don't participate, and I am the one feeding them and not you.
I think though, it probably takes the form of better laws that prevent people from using personalised data to harm you without public (judicial) review, and I think that is going to require people like you thinking of the outcome that you want, instead of foolishly trying the impossible to conserve your personal privacy.
The GPDR is not criminal law. But ignoring that, regulators barely pursue GPDR violations.
Consider the swaths of dark patterns surrounding cookie terror banners. The GPDR language is extremely clear that none of them are legal, but virtually nobody is ever punished.
While the GDPR does not directly prescribe prison sentences, it absolutely enables countries to establish criminal offences for severe data protection violations, and they will clearly extradite!
That's how all prejudices work: We're wired up to be afraid and anxious and to share (and communicate) our anxieties to our friends and neighbours. We're trying to help.
The thing is, knowledge; business, economics, and so on, probably can be used to help people, but in a lot of peoples' recent memories, it's being used to harm.
I lost a lot of my teen friends when I "went corporate", but thirty years later I'm reconnecting with some of them, because people change, and we can learn to recognise someone will participate in capitalism for lots of reasons that are not so simple as being "greedy" or "evil".
But to me, I think it is simple: Capitalism is almost certainly unavoidable, so the world could be better if more kind people participate well in it than don't!
reply