Shannon Sands(@max_paperclips) 's Twitter Profileg
Shannon Sands

@max_paperclips

Software developer & aspiring cognitive architect https://t.co/JAoBrqMLXN Proudly TESCREAL & shitpost/acc. 🇦🇺 pride

ID:1629406731163021313

calendar_today25-02-2023 09:04:05

22,3K Tweets

3,7K Followers

2,8K Following

Shannon Sands(@max_paperclips) 's Twitter Profile Photo

There's an alternative timeline where Lua, not Python, is the language of ML, and modders everywhere are highly in demand to produce AI products. Embedding AI in other applications is far simpler, environment weirdness doesn't exist and the main gripe is 1-indexing

account_circle
7oponaut(@7oponaut) 's Twitter Profile Photo

kache 'A new unregulated operating system is enabling the activity of dangerous criminal organizations across the globe'

@yacineMTB 'A new unregulated operating system is enabling the activity of dangerous criminal organizations across the globe'
account_circle
Shannon Sands(@max_paperclips) 's Twitter Profile Photo

This is unironically why Nous Research and similar open source research crews matter. Academics produce pure theory, but can't really scale it (they're not that interested either, more papers to write). Big commercial labs have >9000 GPUs and are happy to stick with what works

account_circle
Vaibhav (VB) Srivastav(@reach_vb) 's Twitter Profile Photo

Let's fucking go! Mistral just released 7B v0.3 🔥

> Base + Instruct model checkpoints released
> Extended vocabulary to 32768
> Supports new v3 Tokenizer
> Supports function calling
> Uncensored (no moderation tactics used during fine-tuning)

Thanks for the sweet surprise,

account_circle
Awni Hannun(@awnihannun) 's Twitter Profile Photo

Check-out mactop - a nice new tool for performance monitoring on macOS, great for use with MLX!

- Easy install: brew install mactop
- Simple (< 1k loc)
- Customizable display
- Actively maintained

Thanks for the tool Monozukuri !

account_circle
Siddharth Sharma(@siddrrsh) 's Twitter Profile Photo

Introducing ambientGPT: an open-source and multimodal MacOS foundation model GUI

Run GPT-4o and open-source models with full ambient knowledge of your screen.

Foundation models have long been confined to the browser. With ambientGPT, your screen context is directly inferred as

account_circle
Far El(@far__el) 's Twitter Profile Photo

in response to this bill, I am proud to announce off-switch-llama-3-8b. a super compliant llama-3-8b with an off-switch to shut it down in case of FOOM. Follow the instructions in Readme here: huggingface.co/pharaouk/off-s…

in response to this bill, I am proud to announce off-switch-llama-3-8b. a super compliant llama-3-8b with an off-switch to shut it down in case of FOOM. Follow the instructions in Readme here: huggingface.co/pharaouk/off-s…
account_circle
Michael Timothy Bennett(@MiTiBennett) 's Twitter Profile Photo

I fear you've been mislead about risks based on recycled 16th century mysticism (this preprint explains my position: techrxiv.org/users/684323/a…)

I fear you've been mislead about risks based on recycled 16th century mysticism (this preprint explains my position: techrxiv.org/users/684323/a…)
account_circle
Daniel Jeffries(@Dan_Jeffries1) 's Twitter Profile Photo

If open source is bad for security why does the US army use Linux?

Why does it power nuclear subs in the US and the top supercomputers?

Why does it power all of our stock market infrastructure, like the New York Stock exchange?

Why does it power every cloud?

Why did the NSA

account_circle
Shannon Sands(@max_paperclips) 's Twitter Profile Photo

New training-on-evals with a twist - 'deliberately nerfing your model on benchmarks so it slides through under the line'

account_circle
Teortaxes▶️(@teortaxesTex) 's Twitter Profile Photo

Recall over 1M
Multimodal
Codes well
And yet it *still* won't put a scratch on ARC.

Aidan McLau is among the few who note limits of ICL over long context.
François Chollet is in the tiny club of people who get to dunk on LLMs with clear conscience of having done their research.

account_circle