HillmanC

[HillmanC @CrestHillMan]

Worth a bit of more thought. Apps with community and interaction elements perhaps

[Nikita Bier @nikitabier]

Consumer AI apps are all chasing the most obvious ideas. Breakout apps don’t work that way. The technology should be invisible in the background.

The smart builders aren’t making AI girlfriends. They’re using AI bots to seed a dating app until they reach a critical mass of users

[HillmanC @CrestHillMan]

Quality feedback and swift iteration matter!

[Sam Altman @sama]

everything 'creative' is a remix of things that happened in the past, plus epsilon and times the quality of the feedback loop and the number of iterations.

people think they should maximize epsilon but the trick is to maximize the other two.

[Sanyam Bhutani @bhutanisanyam1]

Best tutorial on setting up LLMs locally! 🙏

@Rob_Mulla made an end to end video teaching how to install, run with GUI and connect a Large Language Model to your own data on your own machine.

All open source, running offline:

https://t.co/aTgckhqMlQ

[William Wang @WilliamWangNLP]

Wow, gzip-embedding is wild. IMHO, this paper is more creative than 95% of ACL's main conference papers. How come it was only accepted as findings? #ACL2023NLP #NLProc

[Tim Dettmers @Tim_Dettmers]

QLoRA: 4-bit finetuning of LLMs is here! With it comes Guanaco, a chatbot on a single GPU, achieving 99% ChatGPT performance on the Vicuna benchmark:

Paper: https://t.co/J3Xy195kDD
Code+Demo: https://t.co/SP2FsdXAn5
Samples: https://t.co/q2Nd9cxSrt
Colab: https://t.co/Q49m0IlJHD

[Yao Fu @Francis_YAO_]

The core differences between GPT4 and 3.5 is the ability to perform complex tasks. In this post, we present a complete roadmap towards LLMs complex reasoning abilities, covering the full development stages: pretraining, SFT, RL, CoT prompting, and eval. https://t.co/1rzx02LNlx

[HillmanC @CrestHillMan]

@isabelunraveled Nice next buy for me

[Isabel🌻 @isabelunraveled]

every time i open this book i need to stop my reading session almost immediately to go create something. it is unlike any other text i’ve encountered in terms of creative fuel and its articulation of the process that is transmuting ideas into art.

[George E. Dahl @GeorgeEDahl]

We've just released the first version of our Deep Learning Tuning Playbook! This is our attempt to distill our process for actually getting good results with deep learning. We emphasize hyperparameter tuning since it has been a large pain point. https://t.co/PjeJVWeOzS

[Mengye Ren @mengyer]

Wondering about how to train deep neural networks without backprop? Check out our ICLR 2023 paper: https://t.co/P9KVcZvSBu

Forward gradient computes gradient information from forward pass. But it is slow and noisy — it computes the directional gradient along a random weight perturbation. Gradient variance explodes with deeper and wider networks.

Our key insight is to utilize local activity perturbation and we introduce a new architecture with many local losses throughout the network so that each loss is associated with a small number of weights.

Both supervised and self-supervised (contrastive) learning works. Much better than a lot of backprop-free methods on large scale problems.

Joint work with @geoffreyhinton @skornblith @lrjconan

Our ICLR poster will be presented on Tuesday May 2, in Kigali, Rwanda!

[Jim Fan @DrJimFan]

The future of MLOps is *LMOps*. Wherever there’s standardization, there’s opportunity.

Enter Lamini: FaaS = finetuning as a service! Llama + custom data is becoming the new norm. Defending your moat in AI starts with owning the model.

Built by my longtime friends @realSharonZhou & @GregoryDiamos. Congrats on finally coming out of stealth mode! Eager to see what’s next 🦾

[HillmanC @CrestHillMan]

New models will be multi-prompt, agentic, knowledgeable and more far-reaching in their real-world applications https://t.co/fRXgoEU5QY

[HillmanC @CrestHillMan]

https://t.co/TiecAJqzFB

[Ali Ghodsi @alighodsi]

We are open sourcing Dolly, a ChatGPT-like model that can do instruction following, created for $30, trained 3 hours on 1 server. The secret in magical human-like interactivity probably lies in a small dataset.
https://t.co/6opU6NEArG

[Cristóbal Valenzuela @c_valenzuelab]

The most important user interface of the next decade

[AK @_akhaliq]

Visual ChatGPT: Talking, Drawing and Editing with Visual Foundation Models

build a system called Visual ChatGPT, incorporating different Visual Foundation Models, to enable the user to interact with ChatGPT by 1) sending and receiving not only languages but also images 2) providing complex visual questions or visual editing instructions that require the collaboration of multiple AI models with multi-steps. 3) providing feedback and asking for corrected results

abs: https://t.co/0nYSa7CupY

[HillmanC @CrestHillMan]

Wise words indeed https://t.co/AcUkSkGe7R

[Roman Semenov 🌪️ @semenov_roman_]

I've tried to give GPT access to the internet and the blockchain. What could possibly go wrong?

[Miles Deutscher @milesdeutscher]

Many of next cycle's top performers haven't been released yet.

These projects have potential to 50-100x next bull run.

The next $SOL or $MATIC could be sitting right in front of you.

🧵: Here are the top 13 unreleased projects I'm most looking forward to. 👇