Meta's AI warrior is a LLaMA

PLUS: voice modification and AGI safety

Hello and welcome to the 7,628 new folks joining from last week. That’s right, we’re growing around 1k subs a day 😳.

Milan is the city of style and something must’ve stuck with me because I spent the weekend watching Rugby and sprucing up the look of the email 💅.

Let’s get to it.

p.s. if you joined us from Substack you’ve now been migrated here - I was copy/pasting the email every day to cater for Substack readers but there’s a huge number here vs there. You’ll still get the email, just in your inbox, not in Substack. Apologies if you hate that but had to be done!

🤌 Our Picks

Highlights if you've only got 2 minutes

1/

When Sam Altman of OpenAI opines on the future of AGI we rightly listen and take heed. Especially when it comes hot in the footsteps of Eliezer Yudkowsky planting his core doomsday hypothesis right in front of us. On Friday, OpenAI’s blog saw an outline of their 3 main goals to tackle and direct AGI integration: 1) Empower humanity, 2) Ensure benefits and governance are accessible to all, and 3) Make sure we can navigate the massive risks.

For me, the most notable of their focuses is on small, gradual increments of model improvement and deployment. There is a clear emphasis on ensuring that people are ready to receive AGI by becoming slowly acquainted with the general technology. What does this mean? Not rushing into releases, and a possible push for regulation that requires independent parties to check models before training and deployment. They even go as far as suggesting possible limits on training compute power

Take from this what you will, but it’s comforting to see such big players taking safety, alignment, and integration seriously without feeling the need to go full doomer.

2/

Meta have released LLaMA, an LLM comprising 4 foundation models with 7B to 65B parameters, claiming that it outperforms many benchmark models on various tasks, including GPT-3. And all this with <10% of parameters! There’s also a big emphasis on their openness with the model, claiming to lead the way on making models accessible. But not so fast… Accusations have been floating around that Meta has been somewhat deceptive here, and that actually the weights and tokeniser are not open-sourced at all. If true, they’re claiming openness on a model that has no API, no interface, and no discernible commercial use. I’ll let you guys be the judge on this one.

3/

I’m absolutely loving the wave of voice-generation AI going on right now - I’ve even started reading out-loud to myself each evening to build up enough voice recordings to AI myself up! The latest on this front is Voicemod, who are looking to combine their voice-to-voice tech with Voctro Labs’ sing-to-sing tech to reach an amped up hybrid of speaking and singing tools. With their most recent funding round hitting $14.5m, they’re shaping up to be one of the big competitors in this space.

With the podcasting and audiobook spaces gaining so much popularity in recent years, I think this sort of tech is going to shape up to be huge for revenue generation and time-saving. But don’t fret sound engineers: they’re of the welcoming view that human factors will always be necessary to get rhythm and tone right - to make it real.

🛠️ Cool Tools

Product launches, updates and demos
  • AI-powered meeting experience with Grain HQ. (link)

  • Ask Seneca - Get life advice from a GPT3-based stoic philosopher. (link)

  • Krikey - Use AI to generate avatar animations. (link)

  • Transvribe - Ask questions about youtube videos. (link)

  • Bloop - Ask questions about large codebases as you’d speak. (link)

  • AmjadGPT - AI chatbot that acts like the CEO of Replit. (link)

  • Neeva adds multi-perspective answers to its AI search engine. (link)

  • Pickaxe - Embed AI into your business or website, no code needed. (link)

  • Landbot - No-code chatbots meet AI. (link)

  • Psychedelia - Use stable diffusion to generate psychedelic art. (link)

  • Issac Editor - Academic writing, how it should be. (link)

  • Aim - Easy-to-use and performant open-source ML experiment tracker. (link)

  • Magic Stickers - AI-generated stickers for iMessage. (link)

  • Lasso AI - Operations automation simplified. (link)

  • Spellbook - Draft legal contracts 3x faster with AI. (link)

  • Listener FM - Elevate your podcast post-production process. (link)

  • Devmaps - An open-source database of developer roadmaps prepared with the help of chatGPT. (link)

  • NPM aifiles package - A CLI that helps you organize and manage your files using AI. (link)

  • MegaPortal - An easy-to-use AI model loader designed for Apple devices. (link)

  • Lyrical labs - Write song lyrics faster using the power of AI. (link)

  • Launch an ask-my-book website in 60 seconds with Steamship. (link)

  • Kraftful - AI summaries of user feedback for product research. (link)

  • Tldr bot - Catch up on the conversation you missed with a summary of your discord chats. (link)

  • Wizi - Code search for frontend teams. (link)

  • Mememorph - Turn yourself into all your favourite memes with AI. (link)

🤓 Miscellaneous

News, podcasts, videos, blogs etc
  • AGI doomers, not AGI, will doom us all. (link)

  • Search vs. ChatGPT, ELIZA and AI futures. (link)

  • ChatGPT, GPT4 hype, and building LLM-native products. (link)

  • Can AI really be protected from text-based attacks? (link)

  • Generative AI founders aren't thinking big enough. (link)

  • How to get an AI to lie to you in three simple steps. (link)

  • Prompt engineering is probably more important than you think. (link)

  • Is it time to hit the pause button on AI? (link)

  • Anima unveils Onlybots - AI-based augmented reality pets. (link)

  • Marc Andreessen and Steven Sinofsky on AI sentience, love, ChatGPT vs. Sydney and more. (link)

  • Mapping a YouTube star. (link)

  • Generative AI is a legal minefield. (link)

  • What is data science like at NVIDIA? (link)

  • Why don’t search engines integrate ChatGPT-like bots in more helpful ways? (link)

  • On the internet, nobody knows you’re a human. (link)

  • Tech’s hottest new job - AI whisperer, no coding required. (link, paywall removed here)

  • Metaverse creator Neal Stephenson on the future of virtual reality. (link)

  • Meet the $10,000 Nvidia chip powering the race for AI. (link)

  • How one guy’s AI tracked the Chinese spy balloon across the US. (link)

  • What a 65 year old book teaches us about AI. (link)

  • Wise and Monzo founders back legal startup Robin AI’s $10m Series A. (link)

  • Don’t be deluded by the exaggerated claims made for AI. (link)

  • The generative AI genie is well and truly out of the bottle. (link)

  • Legal doomsday for generative AI ChatGPT if caught plagiarizing or infringing, warns AI ethics and AI law. (link)

  • I’m better than chatbots at the job they’re trying to take. (link)

🎓 Learn

How-to’s and resources
  • How to visualise embeddings from a vector database. (link)

  • Introduction to data-centric AI from MIT. (link)

  • ChatGPT clone in Google Sheets. (link)

  • How to build GPT-3 powered AI apps with Vercel and Svelte. (link)

👋 Too many links?! I created a database for all links mentioned in these emails. Refer 1 friend using this link and I'll send over the link database.

🔬 Research

Published research papers
  • Composer by Alibaba is getting popular for its controlled image generation. Covered it earlier but highlighting again. (link)

  • Modulating pretrained diffusion models for multimodal image synthesis from Adobe. (link)

  • Check your facts and try again - Improving large language models with external knowledge and automated feedback from Microsoft. (link)

  • MUX-PLMs - Pre-training language models with data multiplexing. (link)

  • ZoeDepth from Intel - Zero-shot transfer by combining relative and metric depth. (link)

  • SLAHMR - Simultaneous localization and human mesh recovery from videos. Caution: video heavy project page. (link)

  • Language-driven representation learning for robotics. (link)

  • Leveraging jumpy models for planning and fast learning in robotic domains. (link)

📰 Unclassifieds

Short, sponsored links 
  • Runway ML’s AI film festival is coming to San Francisco on March 22, 2023. Request to attend here.

  • Coinbase is hiring a senior staff software engineer for machine learning. Learn more >>

  • Caption Health is hiring a machine learning engineer. Learn more >>

Have a product, service, job, event, newsletter, app, book, movie, tool, or anything you'd like to share with over 37k subscribers? 
Advertise with us | Job board | Join Community

🖼 AI Images of the day

Funny memes and pics from around the web

Send this to 1 AI-curious friend and receive my AI project tracker database! Use this link.

How was today's email?

Login or Subscribe to participate in polls.

Join the conversation

or to participate.