Hursh Desai

[Hursh Desai @hurshd0]

Weeb mode activate

[Cristiano Giardina @CrisGiardina]

ChatGPT Advanced Voice Mode speaking Japanese (excitedly)

[Han @HanchungLee]

vector search bm25 search

[Hursh Desai @hurshd0]

With recent Scarlett voice model issue shouldn’t we build like open source voice models that are not based on any celebs and has its unique personality to it?

[Hursh Desai @hurshd0]

playing with suno to do some ai rap battles

[Hursh Desai @hurshd0]

@karpathy @suno_ai_ I like the rap version though

[Hursh Desai @hurshd0]

Can't decide where to bet at lol, AOH1996 or LK-99

[Cas (Stephen Casper) @StephenLCasper]

New paper: Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback

We survey over 250 papers to review challenges with RLHF with a focus on large language models. Highlights in thread 🧵

[Michael Moor @Michael_D_Moor]

Excited to announce Med-Flamingo, a new multimodal few-shot learner specialized for the medical domain!

Last week, we uploaded the pre-print, now it’s finally live!
Paper: https://t.co/v2B7HptMpT
Code: https://t.co/zGtDyZ9NHP
Model: https://t.co/aS1HYyb3ls
A short 🧵.

1/n

[Harrison Kinsley @Sentdex]

If you're not already using room temp superconductors, then you're already behind. 👇🧵[1/1024]

[Yann LeCun @ylecun]

No more GIL!
Python code can now breathe multiple gulps of fresh air simultaneously.

[Hursh Desai @hurshd0]

my new flex 💪

[Philipp Schmid @_philschmid]

The new @awscloud p5.48xlarge instance is now GA!
The instance comes with 8x H100 GPUs! 🚀

But at $98.32/hr or $72K/month, your wallet may cry. 😫

[Philipp Schmid @_philschmid]

The new @awscloud p5.48xlarge instance is now GA!
The instance comes with 8x H100 GPUs! 🚀

But at $98.32/hr or $72K/month, your wallet may cry. 😫

[Tim Dettmers @Tim_Dettmers]

The result of long days of CUDA optimizations: the new bitsandbytes release includes 4-bit inference, which is up to 4.2x faster than 16-bit inference (bsz=1). Full HF integration for all models. No code change needed.

Bnb is growing rapidly, just shy of 1M installs/month🧵

[Hursh Desai @hurshd0]

Where my followers at ?

[Hursh Desai @hurshd0]

Make this the👇 new meme

[Hursh Desai @hurshd0]

Eeny, meeny, miny, moe,
Personality and attitude of LLMs on show,
Who will bring the right vibe,
To our team, they'll thrive.

[Hursh Desai @hurshd0]

Eeny, meeny, miny, moe,
Searching for context length high and low,
For the best candidates in MoE

Eeny, meeny, miny, moe,
Which candidate should we choose, you know?
Their context and reasoning will weigh,

[Yam Peleg @Yampeleg]

Which model should you use?

The AI wars TL;DR:
- Long context tasks -> Use Claude 2.0.
- Internet required tasks -> Use Bard.
- Harder reasoning tasks -> Use GPT-4.
- Anything with code -> Use Code Interpreter.
- Long (~essay) + Internet -> Use Bing.

And all are crazy good at this point.
It is much much closer.

If you didn't try them lately: You should.
You would probably be surprised how much bard and claude improved.. night and day..

*Pi is also very capable.
I just find it hard to control when I try it for work related tasks.

[Hursh Desai @hurshd0]

We need LLMs vs Good Ole ML algos UFC cage fight 💪

[Stephen Diehl @smdiehl]

A 14-line Python script using gzip outperforming a 345m parameter transformer model is probably the most hilarious result I've seen all year.
https://t.co/E7o9gFNA6u

[Hursh Desai @hurshd0]

man if being on twitter pays this well, should i just give up my job

[Hursh Desai @hurshd0]

That's called implosion bro