play on bad private servers i guess
Oh damn, I ate too many selfhosted letters, I’m afraid I’ll have to vomit.
Vanilla vmangos
TBC cmangos
WotLK AzerothCore TrinityCore
Haaaa, I think it’s finally over…
SinglePlayerProject
Urgh. Sorry. Nothing to see here, move along everybody!
If you rise anywhere above lever 5 or so, the difficulty ratchets up so much it makes the main quest nearly impossible to complete.
Didn’t Oblivion already have the difficulty slider? You could just adjust that, no?
I know level scaling is a big topic in the industry, but for me, the way it’s implemented nearly ruins what is otherwise a mostly great game.
Two of the first RPGs I played were Gothic and Gothic II which released approximately alongside Morrowind and Oblivion, and they just had no dynamic level scaling at all, so I don’t really see the appeal either. A tiny Mole Rat being roughly the same challenge as a big bad Orc just breaks immersion. If you were to meet the latter in early game it would just curb stomp you, which provided an immersive way of gating content and a real sense of achievement when you came back later with better armour and weapons to finally defeat the enemy who gave you so many problems earlier. Basically the same experience you had with Death Claws in Fallout New Vegas when compared to Fallout 3 - they aren’t just a set piece, they are a real challenge.
The games had their own problems, for example the fighting system sucked, and I’m told the English translation was so bad the games just flopped in the Anglosphere, putting them squarely in the Eurojank category of games. But creating a real sense of progression and an immersive world were certainly not amongst their weaknesses.
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.
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.
Yeah but that’s my point, right?
That
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.
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.
- 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.
“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.
All true, and maybe I have been understating my negative estimation of the whole ordeal a bit, but we are like five levels of escalation deep at this point, and all I’m saying is it’s getting harder and harder to gain a nuanced understanding of the situation. Which is important.
Anyway, given the totality of what I have seen so far LTT has either always been or just devolved into an entirely toxic work environment. Their reaction so far doesn’t inspire any confidence in the slightest. On the contrary, it reinforces all of the accusations.
This isn’t “something”. This is a “we’re sorry (not really)” video. If you watch it in the context of “I have no favorites in this game” and look it it pretty objectively, it feels like just a bait to try to stop the bleeding.
Yeah that’s what I meant with “Southpark-y ‘I’m sorry’ vibes”. For reference: https://www.youtube.com/watch?v=15HTd4Um1m4
P.S.: And that’s… “something”.
A week? To refine all the processes in that size of a company?
No, a week without videos to get started with reviewing processes. I agree with you in general though, if it stops there it’s nothing more than PR. Remains to be seen what will come of it, but the allegations by that former employee are certainly a dampener on an optimistic view of the situation.
Here you go: https://www.youtube.com/watch?v=0cTpTMl8kFY
There are some steps mentioned that they will take, like not making any videos for a week and reviewing internal processes. Getting some Southpark-y “I’m sorry” vibes there, at least it’s something though.
But the video (at least its creation if not its release) seems to predate the Twitter/X thread of a former LTT employee alleging sexual harassment and other toxic workplace behaviour: https://threadreaderapp.com/thread/1691693740254228741
So not very surprisingly most Youtube comments I’ve seen refer to that bomb dropping.
Looks cool, but somehow labeling the ethernet port as “1000 Mbps WLAN” in their marketing material doesn’t inspire confidence.
Well during that time you will have lost quite a lot of skin cells, so in a manner of speaking you literally have some skin in this game. Around 328 grams going by an average loss of 3.6kg/year I took from some random website.