Ben's Bites
Posts
A GPT-3 Alternative

A GPT-3 Alternative

💬 Open-source speech-to-speech translation, a GPT-3 alternative, A new tool for building LLM apps, Text-to-music is here

Ben Tossell
October 20, 2022

Thanks for reading! Ben’s Bites is a roundup of what’s going on in AI. Not subscribed? Do that here.Link overload? Refer 2 friends with this link and get access to the filterable+searchable database.

Hey friends!

I'm getting into a good groove with this email now (although tomorrow will be a challenge as I'm off to a wedding from 9 am 😬). In future, I'd like to make sure it gets into your inbox at a consistent time, around 9 am ET.

Let me know what you think of today's email while I sit inside to avoid the pissing rain here in the UK. (so make it nice - you always do 😘)

Let's get to it!

Prompt: List the topics in today’s email:

💬 Open-source speech-to-speech translation 📝 A GPT-3 alternative 🛠 A new tool for building LLM apps 🎶 Text-to-music is here

🫦 BEN'S BITES

Facebook Meta has released the first AI-powered speech-to-speech translation system for a primarily oral language, translating Hokkien (a variety of Chinese), one of primaly spoken languages. You can use a demo on Hugging Face here and read the full blog post here. Also, they’ve open-sourced the code !!

GPT-3 (from OpenAI) is the main text-based model out there, and it’s not open source. CarperAI (of Stability AI) is partnering with Humanloop, Scale and Hugging Face to bring an open-source alternative to GPT-3. They outline why it’s a big deal (and it is a BIG deal). CarperAI will release a chinchilla-optimal large language model explicitly trained to follow human instructions .

Lex, the largely hyped AI-writing tool has been releasing invites (I’m writing in it now) and Justin filmed a reaction video as he’s using it to write a blog post. It’s a really great demo of how it all works AND feels to use some of this technology.

We’ve all been thinking about pseudonymity-preserving meetups, haven’t we?! No? Well, thankfully Balaji has. Who else?! It’s a way to essentially meet in person while still hiding your identity and speech. A bit of a wild futuristic concept, but there’s a partial demo of it.

Generalizing in the real world with representation learning. Essentially this says that there are several assumptions and problem settings which fail to generalise to the real world. So this paper proposes ways to address those failures.

How reliable is GPT-3? (Hint: it can be more reliable than you think!)

This short thread discusses what a Transformer is. Pretty useful!

This team train diffusion models with strict differential privacy guarantees and outperformed previous methods by large margins.

👋 Too many links?! I created a database for all links mentioned in these emails. Refer 2 friends using this link and I'll send over the link database.

OpenAI is running a Hackathon for Climate Change: how our current AI models can accelerate solutions to climate change.

A form builder with AI built in. Watch the demo using AI here. This was built by a Ben's Bites reader, Dominic.

Dust is a new tool that helps you design and deploy Large Language Model apps. In their announcement, they show off a demo of how it works.

Generate a blog with OpenAI- tutorial with Python.

An open-source image database that unlocks the power of AI for ocean exploration.

The Netherlands has deployed NATO’s first killer robot ground vehicles.

Read about the experience gained by researchers, engineers, and entrepreneurs doing real machine-learning work.

People with hearing loss will be able to listen to music through the sense of touch.

Scientists from Trinity College Dublin believe our brains could use quantum computation.

New research paves way for the innovative theory of cognitive processing.

Text-to-music. MubertAI: A simple demo offering music generation by prompt and duration. Feels like a lot is going to happen quickly for this use case.