OpenAI, Google & Anthropic ban the use of the generated output content from their AI models to train other AI models, under their terms-of-service. However, they’ve been using other online content for their own model training. They can’t have it both ways.
@hardmaru
-
SlimPajama-627B: Large Deduplicated Open-Source LLM Dataset
By
–
SlimPajama-627B: the largest extensively deduplicated, multi1corpora, open-source dataset for training large language models. Sometimes less is more! https://
reddit.com/r/MachineLearn
ing/comments/1467jvm/np_introducing_slimpajama627b_the_largest/
… -
Running State-of-the-Art Models Locally Six Months Prior
By
–
This is far from state-of-the-art, but I was already able to do this on my local machine with open-source models six months ago: https://t.co/aEauJTXeZs
— hardmaru (@hardmaru) 10 juin 2023This is far from state-of-the-art, but I was already able to do this on my local machine with open-source models six months ago:
-
Otter: Multi-modal Model with Enhanced Instruction Following
By
–
Otter is a multi-modal model based on OpenFlamingo (open-sourced version of DeepMind's Flamingo developed by @laion_ai
), trained on MIMIC-IT dataset and showcasing improved instruction-following and in-context learning ability. -
Killer App for Multi-Modal Foundation Models Finally Arrives
By
–
Finally, a killer app for multi-modal foundation models! 🀄️🤖 https://t.co/t9w9PSjyDl
— hardmaru (@hardmaru) 10 juin 2023Finally, a killer app for multi-modal foundation models!
-
Google Does Not Release Imagen Editor for Responsible AI Reasons
By
–
Note that due to concerns in relation to responsible AI, we are not releasing Imagen Editor to the public.
-
QR Codes Generated with Stable Diffusion and ControlNet
By
–
QR codes created using Stable Diffusion and ControlNet. This is Art.
-
Real Networking Beats AI Resume Tricks for Competitive Edge
By
–
Saw this AI post trending on LinkedIn. The best way to have a ‘competitive edge’ isn’t to ‘use AI to write your resume‘ or to ‘get AI to automatically mass apply to jobs’ for you. Instead, network in the real world. Meet people in real life and ask them to have coffee with you.
-
GPT models generate simple stories for efficient grammar learning
By
–
A related recent paper uses GPT-3.5 and GPT-4 to generate a synthetic dataset of short stories that only contain words that typical 3 to 4-year-olds would understand. They show that such a dataset leads to very efficient learning of grammar and reasoning!
-
StableRep: Synthetic Images Improve Visual Representation Learning
By
–
StableRep: Synthetic Images from Text-to-Image Models Make Strong Visual Representation Learners Another paper demonstrating the power of training ML models using *only* synthetically generated content leading to good improvements in learning efficiency! https://
arxiv.org/abs/2306.00984