Your curated collection of saved posts and media
๐GPT or iGPT https://t.co/EF1baAIINN https://t.co/Qa24tJhecn
Mech interp has been very successful in tiny models, but does it scale? โฆKinda! Our new @GoogleDeepMind paper studies how Chinchilla70B can do multiple-choice Qs, focusing on picking the correct letter. Small model techniques mostly work but it's messy!๐งตhttps://t.co/SLFEOqltYR https://t.co/LT29ry8o9t
How is ChatGPTโs behavior changing over time? If you are developing with LLMs or in this case GPT-3.5 or GPT-4, it's definitely worth taking a look at this report. There is suspicion in the AI community that models like GPT-4 are changing/degrading in performance and behavior.โฆ https://t.co/Z7J5pKpRnm https://t.co/a5csuHKaUf
Localizing Object-level Shape Variations with Text-to-Image Diffusion Models @Gradio demo is out on @huggingface demo: https://t.co/D2hMIVgess https://t.co/Q6iM9mrcLp
.@huggingface LLM leaderboard is saying fine-tuned 30b llama v1 beats 70b llama v2 chat. I have a creeping feeling that something in our LLM benchmarks is not working that great. https://t.co/OUwfXwHKkn
LLAMA-v2 training successfully on Google Colab's free version! "pip install autotrain-advanced" ๐ฅ Yes, you can also use your local machine! https://t.co/VOvocAQ46c
So the suspicions about the dumbing-down of GPT-4 may actually be right! Here is some initial hard evidence that GPT-4 is actually getting less capable (and GPT-3.5 is getting more so), since launch. Also, why it is hard to build on AI, when model abilities are quietly changed. https://t.co/xyGyPdhhIE https://t.co/uszx7m0qV9
Augmenting CLIP with Improved Visio-Linguistic Reasoning paper page: https://t.co/PHbgZCUuRi Image-text contrastive models such as CLIP are useful for a variety of downstream applications including zero-shot classification, image-text retrieval and transfer learning. However,โฆ https://t.co/Eu1TgBgmyb https://t.co/JP6iuPXSMV
๐ฆx๐ฆ = ๐ช Experiment with Llama 2 now via LlamaIndex! We made a special release (v0.7.10.post1) to help you get started super easily ๐ https://t.co/ddLURZwmPG https://t.co/CkfBHHE1Cb
Many are asking how Llama 2 compares to other popular models. It is clearly better compared to other models of similar size and the best OS model based on the benchmarks! ๐ฅ Compare against dozens of other models: https://t.co/szZabnc2bY See screenshot below โคต๏ธ https://t.co/ifblNapfwW
When language models โreason out loud,โ itโs hard to know if their stated reasoning is faithful to the process the model actually used to make its prediction. In two new papers, we measure and improve the faithfulness of language modelsโ stated reasoning. https://t.co/eumrl2gxk1
Just realising that training Llama 2 70B emits as much CO2 as a full transatlantic flight from New York <-> Paris (back and forth) ๐ GPUs are getting pretty efficient! also, this can be offset; planes don't. https://t.co/cVxtS0WQcJ
Measuring Faithfulness in Chain-of-Thought Reasoning paper: https://t.co/cSSNX0zkOK Large language models (LLMs) perform better when they produce step-by-step, โChain-ofThoughtโ (CoT) reasoning before answering a question, but it is unclear if the stated reasoning is a faithfulโฆ https://t.co/qGLFZb3Y5G https://t.co/gL0JSXRRN3
Meta releases Llama 2: Open Foundation and Fine-Tuned Chat Models paper: https://t.co/bhG3W56DCW blog: https://t.co/iPHa0PL0DU develop and release Llama 2, a collection of pretrained and fine-tuned large language models (LLMs) ranging in scale from 7 billion to 70 billionโฆ https://t.co/bMAVVNjQbU https://t.co/5Zqk3QAq6m
Announcing Bing Chat Enterprise, the AI-powered chat for work with commercial data protection! Now you can take full advantage of Generative AI creativity at work with the confidence your confidential work information wonโt leak outside your company. https://t.co/8q47eijGzl https://t.co/OrkZbdLeIi
@_akhaliq If you're looking for long form technical analysis of the paper -- ie what you need to actually know, here's my piece: https://t.co/NwDokIYq3W https://t.co/MS3QfMunBi
BIG NEWS ๐ฅณ๐ Building Chatbots apps just got wayyy easier: announcing the new ๐ฒ๐๐๐๐ธ๐๐๐๐๐๐๐๐ class ๐ The *fastest* way to build to build a Chatbot UI in Python -- including streaming, undo/retry, API, all out of the box! Let's take a look at a few examples... https://t.co/vsoWnmN7uP
Incredible to see big tech's massive contributions to open-source #AI on @huggingface ๐๐ค 1. @Meta: 689 models including the likes of MusicGen๐ต, Galactica๐, Wav2Vec๐๏ธ, RoBERTa๐! 2. @Google: 591 models powering AI with BERT๐ฆ, Flan๐ฎ, T5๐ข, mobilnet๐ฑ... 3. @Microsoft: 252โฆ https://t.co/68Xp0DtauI https://t.co/mIVJ7eAC30
First attempt at using #PIKALABS @pika_labs. #aivideo #texttovideo #AIgirl #AIcommunity https://t.co/4wpVHrsaQe https://t.co/Lu5SbJ6ZsE
The other day I discovered a little environment variable buried in the @huggingface Hub Python docs: ๐ท๐ต_๐ท๐๐ฑ_๐ด๐ฝ๐ฐ๐ฑ๐ป๐ด_๐ท๐ต_๐๐๐ฐ๐ฝ๐๐ต๐ด๐ It has changed my life! Docs say 2x faster, but in my testing it's 3-5x faster ๐๐ (and it's just as fast for uploads!) https://t.co/u1XVULnidb
Introducing LLongMA, a series of OpenLLaMA models, trained at 8k context length using linear positional interpolation scaling. The model was trained in collaboration with @theemozilla of @NousResearch and Kaiokendev. https://t.co/uMCy5Da14Z
Text-to-Video task page just landed at @huggingface Tasks ๐ค In this page you will learn how you can generate videos from text ๐ฟ๐ฅ Get started here ๐ https://t.co/rCvsEVSavw ๐ https://t.co/NQboUEsMTS
ไฝ ๆถๅฐไปๅคฉ็Daily Papersไบๅ๏ผ๐ ่ฎข้ ๐ https://t.co/W7m6clPrtf https://t.co/RZtmaguAw7
Today completes the first cohort of our new course on Prompt Engineering for LLMs. Participants worked on projects like customer support email generation, prompt injection detectors, LLM-based evaluators, and LLM-powered conversational assistants. The discussions in this cohortโฆ https://t.co/6iOdsGzUkp https://t.co/ASaS2XH3tq
The code interpreter feature on ChatGPT is the most mind blowing thing I've seen yet. All I did was upload a CSV of SF crime data and ask it to visualize trends(!!) https://t.co/pkFdPqgAzb
VideoGLUE: Video General Understanding Evaluation of Foundation Models paper page: https://t.co/Y97nZAXGm9 We evaluate existing foundation models video understanding capabilities using a carefully designed experiment protocol consisting of three hallmark tasks (actionโฆ https://t.co/WJxQbZNRc2 https://t.co/sgnEheZKj2
mPLUG-DocOwl: Modularized Multimodal Large Language Model for Document Understanding paper page: https://t.co/t52BJ76bYv Document understanding refers to automatically extract, analyze and comprehend information from various types of digital documents, such as a web page.โฆ https://t.co/J6lkXD3dSc https://t.co/Xk2P2cUaTX
The biggest problem with our RL diffusion paper was that nobody could run our Jax+TPU code. No more! I've reimplemented DDPO in PyTorch, plus replicated our results using LoRA for low-memory training! Links below ๐ https://t.co/J7J51mpPot
A lot of users just wanted to just ask on perplexity via their search bar on Chrome thanks to muscle memory. We listened and have it ready now! Install the Perplexity Default Search Chrome Plugin and all your searches are now just Perplexity queries! https://t.co/tTTf2Kh3gv https://t.co/wpa7Qi1ml2
The hairy ball theorem states that in odd dimension d, vector fields on the tangent plane of a (d-1)-sphere necessarily contain a singular point (where it vanishes). https://t.co/5UNiMnjI42 https://t.co/I6RsLEvzry
Made a quick demo reel for the future HD version of the AI Web TV stream ๐ https://t.co/sF4GVqsOJh
The https://t.co/4n2b8GFMaM email just went out https://t.co/2GUVYgiXbs
