- Ben's Bites
- Posts
- AI Feedback = Human Feedback
AI Feedback = Human Feedback
PLUS: X's AI policy and Mustafa Suleyman on AI regulation
Hello folks, today, I’m interviewing the founder of Glean, an AI-powered workplace search, that has raised over $100M at a $1Bn valuation, backed by Sequoia, BoxGroup General Catalyst, Lightspeed, Kleiner Perkins and more. It was founded in 2019. Post your question suggestions here.
Here’s what we have today;
Our picks
1/
Researchers from Google compared reinforcement learning from human feedback (RLHF) to reinforcement learning from AI feedback (RLAIF). Human evaluators preferred the RLHF and RLAIF outputs over the base model ones around 70% of the time. But the interesting part is that both techniques led to equal improvements. That means RLAIF can achieve human-level performance and address the scalability issues of RLHF.
2/
X’s privacy policy confirms it will use public data to train AI models. Some privacy policy changes on X (rip Twitter) say that now they will use the user data to train machine learning models. Elon has said he’ll use public tweets for X AI, so no surprise there. However, he can back out of it by saying we meant improving our algorithm, not training LLMs (like Zoom did).
3/
Mustafa Suleyman on getting Washington and Silicon Valley to tame AI. Mustafa mentions that a few of the voluntary commitments top AI firms made with the White House last month should be legally mandatory. He also makes his case against open-sourcing AI models.
from our sponsor
Doppl – Your digital self, owned by you.
The most accurate AI representation of you, built on the highest security standards. You control your data. You own your Doppl.
Doppl uses cutting-edge machine learning tech to create the most accurate and interactive representation of a unique you via photos, videos, texts and audio. Your AI twin, or Doppl, is a natural evolution of your digital identity. It also serves as a timeless memory of you.
Sign up for our waitlist and be the first to receive the latest updates on our upcoming launch.
From the community
Prompting for AI Ops BootCamp - Learn how to delegate your work to AI.
How to create vector-style illustrations with MidJourney?
PlayHT (YC W23) is Hiring Senior ML Engineers to work on LLMs and generative AI.
GPT-4 capability forecasting challenge - Test your ability to predict how well GPT-4 will perform at various types of questions.
Cool Tools trending product launches from the last 24 hours
BabyFoxAGI - Mod of Baby AGI with self-improving task lists and new skills.
Rephrase AI - Create professional-looking videos with a digital avatar in minutes.
Novel - NPM package for an open-source Notion-style WYSIWYG editor with AI-powered autocompletion.
Automorphic - Incremental finetuning for language models with just 10 samples.
Elto - Human-in-the-loop AI call center, 50% cheaper than typical call center/BPO.
myGPTBrain - QnA over your personal data & bookmarks.
Tactic - Turn Multiple Documents into Comparative Tables with GPT.
Singify - Make AI music covers with your favorite artists anytime.
Lily - Your AI therapist. Confidential help, anywhere, anytime.
Open Interpreter - An open-source, locally running implementation of OpenAI's Code Interpreter.
Prompt2model - Generate deployable models from natural language instructions.
SDXL emoji on Replicate - An SDXL fine-tune based on Apple emojis.
Stockmusic - AI generated royalty-free music.
Ben’s Bites News top posts from the last 24 hours
Belebele benchmarks from Facebook to enable direct comparison of model performance across all languages.
Generative AI in Video and the Future of Storytelling (with Runway CEO Cristobal Valenzuela).
Refact Code LLM - 1.6B state-of-the-art LLM for Code that Reaches 32% HumanEval.
CityDreamer - Compositional Generative Model of Unbounded 3D Cities.
UK government sets out AI Safety Summit ambitions.
Large Content And Behavior Models To Understand, Simulate, And Optimize Content And Behavior.
Plastic Surgery and Artificial Intelligence - How ChatGPT Improved Operation Note Accuracy, Time, and Education.
NVIDIA CEO Meets with Indian Prime Minister Narendra Modi.
Mushroom pickers urged to avoid Amazon books for foraging mushrooms, the books appear to be written by AI.
TinyLlama project aims to pretrain a 1.1B Llama model on 3 trillion tokens.
Ben’s Bites Insights
We have 2 databases that are updated daily which you can access by sharing Ben’s Bites using the link below;
All 10k+ links we’ve covered, easily filterable (1 referral)
6k+ AI company funding rounds from Jan 2022, including investors, amounts, stage etc (3 referrals)
Reply