Actual YouTube/Twitch drama thread II

[Athas]

[Meresgi]

Haha ok yeah this was the shit that taught me the different terms, immediately remembered this as soon as he started

[Byrthnoth]

I guess this goes here kinda: https://9to5mac.com/2024/02/19/reddi...nt-being-sold/

Someone light the Ragns symbol and find out how much money a few years of blue vs. purple debate by barely literate teenagers is worth.

[Ragns]

Someone light the Ragns symbol and find out how much money a few years of blue vs. purple debate by barely literate teenagers is worth.

It would require BG 1000 years to generate as much content as 1 year of Reddit which is around 300M posts per year.

[Roranora]

so not $60m/yr but $60k/yr

a BG-trained AI would know that shit's situational

[Cantih]

I swear to god, if you hook up Genbu to a LLM and it goes rampant...

[Ragns]

I swear to god, if you hook up Genbu to a LLM and it goes rampant...

Last time I tried with text-davinci-002 it would disregard the forum rules to be a white knight.

Also my rig isn't powerful enough to run mixtral-8x7B yet.

[Byrthnoth]

Quantized models are the way to go for inference in casual applications.

4-bit quantization (q4_0) drops the 70B or 8x7B models to a more reasonable size for consumer rigs.

Also, they may need our data so they can answer hard hitting adaberk questions:

[Kaslo]

Quantized models are the way to go for inference in casual applications.

4-bit quantization (q4_0) drops the 70B or 8x7B models to a more reasonable size for consumer rigs.

Also, they may need our data so they can answer hard hitting adaberk questions:

[NotGreyheim]

I wonder if it's a coincidence that this news broke the same day that I got an email from Microsoft asking me to take a survey about AI, and pitch theirs to me (Copilot, I think?)

[Roranora]

Quantized models are the way to go for inference in casual applications.

4-bit quantization (q4_0) drops the 70B or 8x7B models to a more reasonable size for consumer rigs.

Also, they may need our data so they can answer hard hitting adaberk questions:

jesus christ, abort AI

[Byrthnoth]

For those of you who haven't interacted with GPT too much, this is actually only the result from 3.5, which is roundly worse than 4.0. Using the API, 4 tells me that it doesn't know anything about individuals (safety response) when I ask it the first question. When I ask it the second question, it says "The Adaman Hauberk in FFXI is blue."

Now, if I ask it to explain the color controversy, it tells me that the in game appearance does not match its description and asserts that the item is described as being "black" but appears metallic silver. Basically, afaik it hallucinates a controversy using cold reading. Armors don't have meaningful descriptions in FFXI and the choices are blue or purple.

So it seems like OpenAI didn't scrape BG, to their detriment.

[Kaslo]

If it did, it would have known it was Ultramarine.

[Mertron]

nah it’s blue, pack it up folks let’s go home

[Madeline]

So this is what they meant when they said AI would lead to the apocalypse..

[Roranora]

we need a new poll on blue vs purple adaberk. on reddit, get it vacuumed up and baked in. and let's get it right this time

[kuronosan]

Indigo. It's indigo.

[synistar]

So this is what they meant when they said AI would lead to the apocalypse..

Or exploit the hell out of it. idk.

https://mashable.com/article/air-can...misinformation

[Madeline]

That shit's Prussian Blue. Bob Ross would agree.

[Ragns]

It's over, we're doomed.