There should be, that’s just how fiber works. If they lay a 10 Gb line in the street, they’ll probably sell a 1 Gb connection to a 100 households. (Margins depend per provider and location)
If they give you an uncapped connection to the entire wire, you’ll DoS the rest of the neighborhood
That’s why people are complaining “I bought 1Gb internet, but I’m only getting 100Mb!” - They oversold bandwidth in a busy area. 1Gb would probably be the max speed if everyone else was idle. If they gave everyone uncapped connections the problem would get even worse
I assume they’re talking about this api
Any tools that interface well with it?
Lots of tools, but it depends on where you want to use it. For example, inside Obsidian you can use it as a text generator
Inside VSCode you can use something like AI Genie
If you just want to use it raw, you can use postman
Ok, sure. So in a tech race, if energy is a bottleneck - and we’d be pouring $7tn into tech here - don’t you think some of the improvements would be to Power usage effectiveness (PUE) - or a better Compute per Power Ratio?
What benefits to “AI supremacy” are there?
I wasn’t saying there was any, I was saying there are benefits to the race towards it.
In the sense of - If you could pick any subject that world governments would be in a war about - “the first to the moon”, “the first nuclear” or “first hydrogen bomb”, or “the best tank” - or “the fastest stealth air-bomber”
I think if you picked a “tech war” (AI in this case) - Practically a race of who could build the lowest nm fabs, fastest hardware, and best algorithms - at least you end up with innovations that are useful
For all our sakes, pray he doesn’t get it
It doesn’t really go into why not.
If governments are going to be pouring money into something, I’d prefer it to be in the tech industry.
Imagine a cold-war / Oppenheimer situation where all the governments are scared that America / Russia / UAE will reach AI supremacy before {{we}} do? Instead of dumping all the moneyz into Lockheed Martin or Raytheon for better pew pew machines - we dump it into better semiconductor machinery, hardware advancements, and other stuff we need for this AI craze.
In the end we might not have a useful AI, but at least we’ve made progression in other things that are useful
Well @ @TheGrandNagus and @SSUPII - I think a lot of Firefox users are power users. And a lot of the non-power Firefox users, like my friends and family, they’re only using Firefox because I recommended them to use it, and I installed all the appropriate extensions to optimize their browser experience.
So if Firefox alienates the power users - who are left? I’m gonna move on to Waterfox or Librewolf, but they are even more next-level obscure browsers. My non-tech friends know about Chrome, Edge, and Firefox, so I can convince them to use one of those… But I kinda doubt I can get them to use Librewolf. If I tell them Firefox sucks now too, they’ll probably default to chrome
If AI integration is to happen […], then this to me seems to be the best way to do it.
Well, to me the best way to do it would be for Mozilla to focus on being the best bare-bone, extendable browser.
Then - if people want an AI in their browser - people should be able to install an AI extension that does these things. It’s a bit annoying they’re putting random stuff like Pocket, and now an AI in the core of the browser, instead of just making it an option to install extendable
So the full story would be that Elon stayed up until 5:30 a.m playing Elden Ring in a Vancouver hotel - was very stressed, saw on Twitter that people knew he was raging in Vancouver based on the Jet Tracker - stressing him out even more -
Though “Fuck it, maybe I can’t beat Malenia, but at least I can beat this asshat on Twitter tracking me!”
…If only FromSoftware had added some pay-to-win elements… Like “For A Small $1 billion Micro-Transaction you get the uber Malenia slayer sword!” -
We would be living in a totally different timeline
I suppose it’s not allowed them. That kind of sucks, it is pretty convenient to just use a replicate.com machine and use a large image model kinda instantly. Or spin up your own machine for a while if you need lots of images without a potential cold-start or slow usage on shared machines
I wonder why they chose this license, because the common SD license basically lets you do whatever you want
Well I have Copilot Pro, but I was mainly talking about GitHub Copilot. I don’t think having the Copilot Pro really affects Copilot performance.
I meanly use AI for programming, and (both for myself to program and inside building an AI-powered product) - So I don’t really know what you intend to use AI for, but outside of the context of programming, I don’t really know about their performance.
And I think Copilot Pro just gives you Copilot inside office right? And more image generations per day? I can’t really say I’ve used that. For image generation I’m either using the OpenAI API again (DALL-E 3), or I’m using replicate (Mostly SDXL)
This model is being released under a non-commercial license that permits non-commercial use only.
Hmm, I wonder whether this means that the model can’t be run under replicate.com or mage.space.
Is it commercial use if you have to pay for credits/monthly for the machines that the models are running on?
Like is “Selling the models as a service” commercial use, or can’t the output of the models be used commercially?
I use Copilot, but dislike it for coding. The “place a comment and Copilot will fill it in” barely works, and is mostly annoying. It works for common stuff like “// write a function to invert a string” that you’d see in demos, that are just common functions you’d otherwise copypaste from StackOverflow. But otherwise it doesn’t really understand when you want to modify something. I’ve already turned that feature off
The chat is semi-decent, but the “it understands the entire file you have open” concept also only just works half of time, and so the other half it responds with something irrelevant because it didn’t get your question based on the code / method it didn’t receive.
I opted to just use the OpenAI API, and I created a slack bot that I can chat with (In a slack thread it works the same as in a “ChatGPT context window”, new messages in the main window are new chat contexts) - So far that still works best for me.
You can create specific slash-commands if you like that preface questions, like “/askcsharp” in slack would preface it with something like “You are an assistant that provides C# based answers. Use var for variables, xunit and fluentassertions for tests”
If you want to be really fancy you can even just vectorize your codebase, store it in Pinecone or PGVector, and have an “entire codebase aware AI”
It takes a bit of time to custom build something, but these AIs are basically tools. And a custom build tool for your specific purpose is probably going to outperform a generic version
What we have is machine learning, just an algorithm that takes input and gives you output. It can’t act on its own.
Isn’t that basically what “real learning” is as well? Basically you’re born as a baby, and you take input, and eventually you can replicate it, and eventually you can “talk” for example?
But in the training data something was off, suddenly your AI is racist and gives every black person a lesser amount.
Same here, how is that different from “real learning”? You’re born into a racist family, in a racist village where everyone is racist. What is the end-result; you’re probably somewhat racist due to racist input - until you might unlearn that, if you’re exposed to other data that proves your racist ideas were wrong
If a human brain is basically a big learning computer, why wouldn’t AI eventually reach singularity and emulate a brain and beyond? All the examples you mentioned of what it can’t do, is just stuff it can’t do yet
In your original comment, it seemed like you were suggesting hashing only before transmission
Ok, that wasn’t what I was suggesting, no. That would effectively make your password hash the password itself - and it would kinda be stored in PlainText on the server, if you skip the client auth and send that value to the server directly through the api or something
how does such a service (like Proton Mail) perform this in a web browser without having access to the data necessary to decrypt all of the data it’s sending? […] do you send down an encrypted private key that can only be decrypted with the user’s password?
Yes, pretty much. I can’t really find a good, detailed explanation from Proton how it exactly works, but LastPass uses the same zero-knowledge encryption approach - which they explained with some diagram here - with a good overview of the client/server separation of it’s hashing.
I’m not really sure how it opens up replay attacks, since it doesn’t really change anything to the default auth. There are already sites that do this.
The only difference is that instead of sending an http request of { username = "MyUsername", Password = "MyPassword" }
changes to { username = "MyUsername", Password = HashOf("MyPassword") }
- and the HashOf(“MyPassword”) effectively becomes your password. - So I don’t know how that opens up a possibility for replay attack. There’s not really any difference between replaying a ClearText auth request vs an pre-hashed auth request. - Because everything else server side stays the same
(Not entirely auth related), but another approach of client side decryption is to handle decryption completely client site - meaning all your data is stored encrypted on the server, and the server sends you an encrypted container with your data that you decrypt client side. That’s how Proton(Mail) works in a nutshell
No, the client side hashing doesn’t substitutes anything server side, it just adds an extra step in the client
there is no possible way to handle sensitive data without storing it in memory at some point
Since we’re nitpicking here - technically you can. They could run hashing client side first, and instead of sending the password in plain-text, you’d send a hashed version
[From the github comment]
The issue I see with the RFC is not wanting to allow users to add tags to ease the burden on moderators. This comes from a lack of users with privileges, so moderators are overworked and need to rely on bots.
If the tags are just kinda “plain old hashtags” - and not something cool like I mentioned in the previous post 😉 -
Possibly you could have a look at how Gazelle handles tags, where it’s just a voting system. For example, this is “Kanye West” https://i.imgur.com/adTe4t8.png - then tags are no longer a boolean yes/no system, but a user-voted system. And then it’s no longer a moderation concern to have to correct tags, and you don’t need “User privileges” to manage the tags either.
It’s just a pretty chaotic system though - you might still want moderators to remove bad tags and/or ban users from creating tags if they’re always adding nonsense.
Could be some point based system like Stackoverflow - users with n points can vote on existing tags, users with n+ points can add their own new tags
Yea, noticed that last week. Is already fixed again in latest revanced.
Delete microG, revanced manager, and YouTube revanced
Download and install the new gmscore, which replaces microG: https://github.com/ReVanced/GmsCore/releases/tag/v0.3.1.4.240913
Download and install latest version of Revanced Manager: https://github.com/ReVanced/revanced-manager/releases/tag/v1.20.1
Download and install YouTube 19.09.37 from APKmirror: https://www.apkmirror.com/apk/google-inc/youtube/youtube-19-09-37-release/youtube-19-09-37-android-apk-download/