R to @SullyOmarr: Everyone who liked this tweet will have their gpt4 access cut by morning...

RT by @daryl_imagineai: Holy shit lmao

More like 53.8% amirite? 🥁🎤

R to @daryl_imagineai: I 👏 your achievements. I may only further this discussion by recommending this book “Programming Massively Parallel Processors: A Hands-on Approach” which explains the nuances in architectures of CPU vs GPU (aka PPU) and the effects on ML computations.

R to @cpavel866: Thanks for mansplaining how ML works to a data scientist who built and owns multiple AI companies. The point is, if through R&D we can get *consumer* cpu (e.g. M2) performance 'good enough', why pay a premium for unused capacity from a monopolist with supply chain bottlenecks?

R to @cpavel866: Thanks for sharing some numbers. I'm just curious, what about memory bandwithd? I noticed llama.cpp on Mac M1 with Metal enabled performing 2x-2.5x faster than CPU-only llama.cpp

R to @daryl_imagineai: Consumer grade RTX3090 is about 5x faster in matrix operations (the foundation of deep learning compute) yet 10x cheaper than flagship AMD Epyc CPU. Yes you can run enterprise grade ML models on 3090s/4090s.

R to @cpavel866: Because they're cheaper to purchase, easier to acquire and consume less energy

Cyberpunk is now 😂

RT by @daryl_imagineai: Jon Stewart's show will no longer be shooting its upcoming third season after disagreements with Apple about his planned episodes on China & Ai. Apple approached Jon & said that his views needed to align with the company to move forward, so he quit. https://www.theverge.com/2023/10/19/23924549/jon-stewart-apple-ai-china-cancel

R to @gerrylynch: Universally, except if you'd consider asking Surinamese people.

In today's episode of 'Not strange, just colonialism consequences':

Remember this gem?

Man, the 'how traumatic it is' just hit a bit too hard this morning 😢

As per usual, having a manifesto is a pretty good indication that you and the movement you represent are cringeworthy. And so it is too with the e/acc techbros who were last season's crypto gm wagmi nft bros.

Because you'd be too busy maintaining dashboards

Maybe my own visual model for desserts is too narrow, but shaved parmezan cheese on baklava? 😳

Good. https://www.reuters.com/legal/litigation/google-says-data-scraping-lawsuit-would-take-sledgehammer-generative-ai-2023-10-17/

The posterior #distribution when I apply my slightly opinionated #bayesian prior

RT by @daryl_imagineai: Not at the moment: but you can use it in a colab here: https://github.com/Vaibhavs10/insanely-fast-whisper