Welcome to Incremental Social! Learn more about this project here!
Check out lemmyverse to find more communities to join from here!

Muehe

@Muehe@lemmy.ml

This profile is from a federated server and may be incomplete. Browse more on the original instance.

Muehe OP ,

So, I think it’s pretty stupid to argue whether “convicted felon” should be in his opening lede line for Wikipedia.

True though that may be, I don't think it's surprising that this would happen, and since making the post I have been falling down a rabbit hole of finding out how Wikipedia is handling situations like this, partly through taking more than a glancing look at the talk pages for the first time ever, and it's fascinating.

Currently my deepest point of descent is this sub-thread on the Admin board about the "consensus" boxes on top of talk pages being an undocumented and unapproved feature.

Muehe ,

(not actually everything, but I get your hyperbole)

How is it hyperbole? All artificial neural networks have "hallucinations", no matter their size. What's your magic way of knowing when that happens?

Muehe OP ,

Did that too a while back, but anecdotally it feels lees buggy through Steam, especially regarding updates.

Also clicking the Stop button in Steam doesn't leave behind zombie processes off Battle.net.exe and Agent.exe, which I had to manually kill when using Lutris. Assume that's due to Protons(?) pressure-vessel thingy.

Muehe OP ,

AFAIK you can set Lutris up to use GE or Proton builds.

Muehe OP ,

Oh so that's what you meant. Thought you meant don't use Lutris at first because of how you worded it. That makes much more sense.

Muehe OP ,

The WINE_SIMULATE_WRITECOPY=1 %command% is the Steam launch option you set, with %command% meaning roughly "what Steam would do without any launch options set".

The whole process was a bit finicky and I did it a few month ago, but from what I remember it went something like this:

  • Download battle.net installer
  • Add it as non-Steam game to run it
  • Locate the newly created prefix in Steam directory
  • Add the Battle.net.exe in it as a non-Steam game, then remove the installer (not the other way around or the prefix will be deleted)
Muehe OP ,

Neither. Battle.net is Blizzards game launcher and store. They also own that domain, but the name usually refers to the binary.

Muehe OP ,

I'm involved in the development of an addon for the Classic WoW versions (Questie), and the thing I do there is such a convoluted process that not doing it feels like letting my fellow devs and the users down. But you can do development on the PTRs and beta servers, so I haven't given money to Blizzard in a long time. Now you could argue that this is even worse in regards to supporting Blizzard than just paying for a game, but I rationalise it to myself with the fact that the newer clients will inevitably be used for private servers just like the old ones were (some already are actually).

Muehe ,

there must be a bit more to it than that. AIM, Skype, and several others were viable options with existing userbases.

Once upon a time in a messenger landscape far far away there lived a king called XMPP. It had a lot of powerful children, WhatsApp, Facebook Messenger, Google+, and even Skype amongst them. And they all worked together in a big federation towards the commonwealth of all, freely sharing their metadata. But then some of the children grew greedy, jealously guarding their own gardens behind higher and higher walls, breaking down the federation. And thus the era of the warring messengers began. But prophecy foretells of a prince to unite all the disparate standards in one big Matrix again, completing yet another revolution of the XKCD 972 wheel of time.

For real though it was phone numbers. WhatsApp always worked based off of phone numbers, which is an identity confirmation method that was immediately familiar to most people at the time, even more so than email.

Muehe ,

Firefox has no extensions so no Adblock.

That's because so far every browser on iOS had to use WebKit as it's HTML rendering engine, meaning that even if you installed another browser manually you were basically still using Safari under the hood. IIRC the new DMA rules include allowing other browser engines like Gecko, so Mozilla is probably already working on making addons available. I mean they are available on Android, so why wouldn't they make them available on iOS now that they finally can?

Muehe ,

it’s definitely more pro-privacy than Brave or FireFox. I’ve never had to jump through a captcha to use Google in those browsers.

You have this backwards. Google showing you captchas is basically them saying they can't match your browser to any know (shadow) profile they have already stored. So they aren't sure you are a human and if so which one specifically. Getting harassed with a captcha is essentially like a badge of honour for your browsers privacy settings.

Muehe ,

Oh you are right, I misread that. Thanks for pointing it out.

Muehe ,

Great, thanks, now I'm envisioning a post-solarpunk dystopia where humanity dies due to some idiot circumventing the 4th law of robotics because of the horny.

Scientists aghast at bizarre AI rat with huge genitals in peer-reviewed article | It's unclear how such egregiously bad images made it through peer-review. (arstechnica.com)

Scientists aghast at bizarre AI rat with huge genitals in peer-reviewed article | It's unclear how such egregiously bad images made it through peer-review.::It's unclear how such egregiously bad images made it through peer-review.

Muehe ,

At this point somebody should really create a /c/Aipocalypse community or something to collect stuff like this.

Muehe ,

pandoc.org is probably what you are looking for, but you might have to create a custom reader/writer or find one on the internet.

The White House wants to 'cryptographically verify' videos of Joe Biden so viewers don't mistake them for AI deepfakes (www.businessinsider.com)

The White House wants to 'cryptographically verify' videos of Joe Biden so viewers don't mistake them for AI deepfakes::Biden's AI advisor Ben Buchanan said a method of clearly verifying White House releases is "in the works."

Muehe ,

Cryptography ⊋ Blockchain

A blockchain is cryptography, but not all cryptography is a blockchain.

Muehe ,

Now? Want?!

Muehe ,

Well I guess that depends very much on what you mean by being on life support. Like financially speaking? Oh yeah, they are more or less entirely dependent on Google. Regarding user numbers? Sure, Statcounter says 3.3% currently. Technologically speaking? Not really, quite the opposite actually. Besides Apples WebKit and Googles fork of it called Blink there is but one game in browser engine town, and its name is Gecko.

Muehe ,

I sometimes forget that this picture exists, and then I happen upon it in places like here and it just smacks me in the face how perfectly it encapsulates the total and utter loss of decorum in politics. I mean it was never perfect obviously, but in past times there was a somewhat reasonable expectation of politicians being civil and them losing their office if they were publicly caught out not to be. It was rare, but it happened. Yet here you have the supposedly "most powerful man in the world" just dropping every pretence and hustling for some company in a flagrant abuse of his office. It's so brazenly corrupt. And the worst thing is this was just another Tuesday for Trump, mild shit-storm, on to the next fucked up thing he did. Society never even had time to realise what a historic moment this was. It was just dropped on the pile.

Muehe ,

The highest GDPR fine was 1.2 billion.

This isn't the GDPR but the DMA. That said, fines there are even steeper, 10% of global revenue for the first offence, 20% for repeated offences.

Muehe ,

https://knowyourmeme.com/memes/loss

TL;DR: Somebody made a really weird episode of a web comic and the configuration of the figures in the panel has become a meme named "Loss" after the name of that episode. Pic related:

pic related

The joke here is that evolution gave you pattern recognition to avoid predators, but now you are using it for useless things like recognising this comic is spatially organised in the same way the Loss comic is.

Muehe ,

Just here to leave the daily reminder that API reimplementation may constitute fair use under certain circumstances.

Muehe ,

“Inclusive models” would need to be larger.

[citation needed]

To my understanding the problem is that the models reproduce biases in the training material, not model size. Alignment is currently a manual process after the initial unsupervised learning phase, often done by click-workers (Reinforcement Learning from Human Feedback, RLHF), and aimed at coaxing the model towards more "politically correct" outputs; But ultimately at that time the damage is already done since the bias is encoded in the model weights and will resurface in the outputs just randomly or if you "jailbreak" enough.

In the context of the OP, if your training material has a high volume of sexualised depictions of Asian women the model will reproduce that in its outputs. Which is also the argument the article makes. So what you need for more inclusive models is essentially a de-biased training set designed with that specific purpose in mind.

I'm glad to be corrected here, especially if you have any sources to look at.

Muehe ,

First, there is no thing as a “de-biased” training set, only sets with whatever target series of biases you define for them to reflect.

Yes, I obviously meant "de-biased" by definition of whoever makes the set. Didn't think it worth mentioning, as it seems self evident. But again, in concrete terms regarding the OP this just means not having your dataset skewed towards sexualised depictions of certain groups.

  1. either you replace data until your desired objective, which will reduce the model’s quality for any of the alternatives

[...]
For reference, LoRAs are a sledgehammer approach to apply the first way.

The paper introducing LoRA seems to disagree (emphasis mine):

We propose Low-Rank Adaptation, or LoRA, which freezes the pre-trained model weights and injects trainable rank decomposition matrices into each layer of the Transformer architecture, greatly reducing the number of trainable parameters for downstream tasks.

There is no data replaced, the model is not changed at all. In fact if I'm not misunderstanding it adds an additional neural network on top of the pre-trained one, i.e. it's adding data instead of replacing any. Fighting bias with bias if you will.

And I think this is relevant to a discussion of all models, as reproduction of training set biases is something common to all neural networks.

Muehe ,

Yeah but that's my point, right?

That

  1. you do not "replace data until your desired objective".
  2. the original model stays intact (the W in the picture you embedded).

Meaning that when you change or remove the LoRA (A and B), the same types of biases will just resurface from the original model (W). Hence "less biased" W being the preferable solution, where possible.

Don't get me wrong, LoRAs seem quite interesting, they just don't seem like a good general approach to fighting model bias.

Muehe ,

a neural network with a series of layers (W in this case would be a single layer)

I understood this differently. W is a whole model, not a single layer of a model. W is a layer of the Transformer architecture, not of a model. So it is a single feed forward or attention model, which is a layer in the Transformer. As the paper says, a LoRA:

injects trainable rank decomposition matrices into each layer of the Transformer architecture

It basically learns shifting the output of each Transformer layer. But the original Transformer stays intact, which is the whole point, as it lets you quickly train a LoRA when you need this extra bias, and you can switch to another for a different task easily, without re-training your Transformer. So if the source of the bias you want to get rid off is already in these original models in the Transformer, you are just fighting fire with fire.

Which is a good approach for specific situations, but not for general ones. In the context of OP you would need one LoRA for fighting it sexualising Asian women, then you would need another one for the next bias you find, and before you know it you have hundreds and your output quality has degraded irrecoverably.

  • All
  • Subscribed
  • Moderated
  • Favorites
  • •
  • incremental_games
  • meta
  • All magazines
    •