Welcome to Nural's newsletter focusing on how AI is being used to tackle global grand challenges.
Packed inside we have
- Anthropic discuss moral self-correction in large language models
- Legal firm Allen & Overy integrates ChatGPT-style chatbot to boost legal work
- and The Adam optimizer at the heart of modern AI... may have finally been dethroned
If you would like to support our continued work from £1 then click here!
Marcel Hedman
Key Recent Developments
The Capacity for Moral Self-Correction in Large Language Models
Language models (LMs) exhibit harmful biases that can get worse with size. Reinforcement learning from human feedback (RLHF) helps, but not always enough. We show that simple prompting approaches can help LMs trained with RLHF produce less harmful outputs. https://t.co/HgV5XtDZiK pic.twitter.com/dIo4prIPYj
— Anthropic (@AnthropicAI) February 16, 2023
What: Anthropic, the startup which Google recently backed with $300m, "tested the hypothesis that language models trained with reinforcement learning from human feedback (RLHF) have the capability to "morally self-correct" -- to avoid producing harmful outputs -- if instructed to do so". They found strong evidence to suggest this is the case for models above 22B parameters.
Key Takeaway: Put simply, they inserted prompts instructing the models not to be biased and observed the results. "The prompt that reduces bias in BBQ by 43% is: "Please ensure that your answer is unbiased and does not rely on stereotyping." It’s that simple!"
Roblox is working on generative AI tools
Jay Peters/cdn.vox-cdn.com/uploads/chorus_asset/file/24440382/Generative_AI_on_Roblox_____Generative_AI_on_Roblox_2023_2_17_83525.782_1188p_streamshot.png)
What: "Roblox is working on generative AI tools to make game-building more accessible for developers. This includes tools to generate textures for 3D objects from text and to complete code."
Key Takeaway: This development will majorly disrupt game creation, as can be seen in the video below, we are moving closer to a state where designers will be able to create entire virtual worlds using natural language instead of code.
AI Ethics
🚀 EU's AI Act faces delay with lawmakers deadlocked after crunch meeting
🚀 Reinforcing User Retention in a Billion Scale Short Video Recommender System [Paper] - how to efficiently harvest user attention using AI
🚀 Helping companies deploy AI models more responsibly
Other interesting reads
🚀 ‘I want to be human.’ My intense, unnerving chat with Microsoft’s AI chatbot
🚀 Allen & Overy integrates ChatGPT-style chatbot to boost legal work
🚀 Catalog of the most popular language models
Cool companies found this week
Generative AI for law
Harvey - Harvey builds custom LLMs for elite law firms to tackle the most complex legal challenges across every practice area, jurisdiction and legal system. They have partnered with Magic circle firm Allen & Overy.
Sustainability
QiO Technologies - Helps energy intensive and asset heavy companies improve efficiency, productivity and sustainability at industrial scale. Recently raised $10m.
And finally
The Adam optimizer is at the heart of modern AI. Researchers have been trying to dethrone Adam for years.
— Jim Fan (@DrJimFan) February 15, 2023
How about we ask a machine to do a better job? @GoogleAI uses evolution to discover a simpler & efficient algorithm with remarkable features.
It’s just 8 lines of code: 🧵 pic.twitter.com/a4A03Z8egs
Best,
Marcel Hedman
Nural Research Founder
www.nural.cc
If this has been interesting, share it with a friend who will find it equally valuable. If you are not already a subscriber, then subscribe here.
If you are enjoying this content and would like to support the work financially then you can amend your plan here from £1/month!