PhillyGunson

263 posts

PhillyGunson

@PhillyGunson

Katılım Kasım 2022

42 Takip Edilen15 Takipçiler

@DyingBreedTT I do this with AI now to help me create ideas for DND adventures. I'll give the AI a youtube link to a song and tell it to use the lyrics and themes of the song to create a dnd encounter. Then I play that song in the background while we play.

English

Dying Breed Tabletop@DyingBreedTT·15h

Back in the day, I would use music for inspiration when creating adventures for my D&D table. I knew some others who did too. Typically, I like instrumental soundtracks. The Conan score by Poledouris was a favorite. Have you ever done the same? What are your favorites?

English

3.5K

PhillyGunson@PhillyGunson·23h

@PopBase first "look"? I cant see shit.

English

Pop Base@PopBase·1d

First look at Paapa Essiedu as Severus Snape in HBO’s ‘HARRY POTTER’ reboot series.

English

465

301

6.4K

7.5M

PhillyGunson@PhillyGunson·1d

@slethion The entire progression system needs to be changed completely. There is no progression tree. It is a progression bush. Do you want armored horses and health teas right away? You can have them. Want to rush elite crates and kill bow users with an MP5? You can do that too.

English

376

Sleth@slethion·1d

Okay so at least I'm not the only one feeling the boredom

English

147

27.1K

PhillyGunson@PhillyGunson·1d

@ROANOV1991 @AetherMoose @ChivalryGuild Wasn't it Ch 6-8 that weren't in the moves? A Short Cut to Mushrooms surely happened in the movie.

English

The Chivalry Guild@ChivalryGuild·2d

What does "uncover why the War of the Ring was very nearly lost before it even began" have to do with anything? Why would anyone care?

English

898

31.5K

PhillyGunson@PhillyGunson·1d

@thestreamingdev Compared to which Nvidia card?

English

4.5K

thestreamingdev()@thestreamingdev·1d

I ran a 35-billion parameter AI agent on a $600 Mac mini. Specs: M4 Mac-Mini 16GB RAM The model doesn't fit in RAM. It pages from the SSD at 30 tokens/second. On NVIDIA, the same paging gives you 1.6 tok/s. Apple Silicon gives you 30. That's 18.6x faster. No cloud. No API keys. $0/month. Here's what it can do 🧵

English

169

213

3.2K

686.2K

PhillyGunson@PhillyGunson·1d

@thestreamingdev This model is 3b active right?

English

683

PhillyGunson@PhillyGunson·1d

@CantEverDie How many trans were killed in the making of this video?

English

106

onion person@CantEverDie·1d

if you watch and support this you are funding trans genocide globally by the way

DiscussingFilm@DiscussingFilm

First trailer for the ‘HARRY POTTER’ series. Releasing this Christmas on HBO.

English

433

6.4K

36.5K

1.3M

PhillyGunson@PhillyGunson·1d

@ImNikhil117 @ForrestPKnight Ford said they were going to make cars but ended up making bombers for the war. Shit changes.

English

Nikhil Sharma@ImNikhil117·2d

@ForrestPKnight Nobody is surprised AI is used for war. People are surprised the same labs that spent years publishing safety research quietly signed defense contracts. The gap between what they say and what they do is the actual story.

English

Forrest Knight@ForrestPKnight·2d

AI was always going to be used for war. I have no clue why folks are surprised.

English

3.1K

PhillyGunson@PhillyGunson·1d

@rcwhiteley @steipete gollumdidnothingwrong Hey! It didnt work!

English

Robert@rcwhiteley·1d

@steipete Wow, X scrambles your password into a string of numbers if you type it out. Here's mine: 99483893094334

English

1.4K

PhillyGunson@PhillyGunson·1d

@CryptoKaleo Nothing like Pied Piper. They found a way to compress random data. You find a way to do that and you will be richer than Elon.

English

645

K A L E O@CryptoKaleo·2d

So Google TurboQuant is basically Pied Piper and just hit a Weismann Score of 5.2

Google Research@GoogleResearch

Introducing TurboQuant: Our new compression algorithm that reduces LLM key-value cache memory by at least 6x and delivers up to 8x speedup, all with zero accuracy loss, redefining AI efficiency. Read the blog to learn how it achieves these results: goo.gle/4bsq2qI

English

271

4.5K

467.7K

PhillyGunson@PhillyGunson·1d

@LeCodeBusiness @BillyM2k You mean a bigger model running on the same hardware right? :)

English

LeCodeBusiness@LeCodeBusiness·1d

@BillyM2k TurboQuant cuts KV cache memory 6x with zero accuracy loss, that's not just faster inference, it's the same model running on hardware half the price. That's the actual unlock.

English

Shibetoshi Nakamoto@BillyM2k·1d

this is HUGE MASSIVE WORLD CHANGING (i don’t get it)

Google Research@GoogleResearch

English

235

43.4K

PhillyGunson@PhillyGunson·1d

@RasmusOthar @BillyM2k Close! its like .zip for the cache that remembers stuff during your chat session. That cache grows quickly as you interact with the model and this method makes it grow slower and makes it faster to process. This makes inference cheaper / faster.

English

Rasmus "CodeWizard" Othar@RasmusOthar·1d

@BillyM2k .zip for LLM?

English

PhillyGunson@PhillyGunson·1d

@CharlesD353 @osmarks1 The blog post from yesterday isnt the same as the whitepaper. Google did real world testing and produced real benchmarks on real models. Thats what happened yesterday. The blog post was the proof this method works in the real world. Independent testing shows similar results.

English

Charles🔸@CharlesD353·2d

@osmarks1 My assumption was something like this, I was just surprised by all the apparent Twitter excitement about it

English

8.3K

Charles🔸@CharlesD353·2d

I notice I am confused when Google publishes something like this. Should I assume it is non-novel and something they expect all their competitors already have? Or there's not much benefit to hoarding this particular idea for some reason?

Google Research@GoogleResearch

English

490

130.2K

PhillyGunson@PhillyGunson·1d

@CharlesD353 Great question. There was a whitepaper about this from a year ago that showed it was theoretically possible with some early experimental results. Yesterday Google released a blog post with real-world benchmarks and testing. This took the method from theoretical to fully proven.

English

PhillyGunson@PhillyGunson·1d

@PMV_InferX @jukan05 Yep! It was Google showing everyone yesterday that this method works in the real world with real models. The blog had end-to-end tests with real data. The whitepaper had theoretical findings with experimental results.

English

Prashanth (Manohar) Velidandi@PMV_InferX·2d

@jukan05 Paper ≠ production impact.

English

6.7K

Jukan@jukan05·2d

Bro, that shit you guys are hyping dropped in April last year. Why are you acting like it’s new now?

Jukan@jukan05

The fact that memory stocks are crashing because of Google’s Turboquant is a pretty good indicator of how many clueless people this market is filled with. It’s like saying Aramco should crash because Toyota came out with a next-generation hybrid engine.

English

117

165

2.7K

404K

PhillyGunson@PhillyGunson·1d

@rizz1901 @jukan05 No one "figured out it's relevant". Google made a blog post yesterday with proof that this method works. People waited to be told this method works and can be implemented for real world gains. Those real world gains were in Googles blog post not the whitepaper.

English

Rizwan Khan@rizz1901·2d

@jukan05 ngl the funniest part is that the paper is from April 2025 and people are just now figuring out it's relevant 💀

English

10K

PhillyGunson@PhillyGunson·1d

@jukan05 This is what caused sell-off yesterday. It was a blog post by Google which came out yesterday. Not the whitepaper from a year ago. research.google/blog/turboquan…

English