😺 Superintelligence

PLUS: ChatGPT fights clones

Welcome, humans.

Getting plugin access is the nerd’s version of Halloween. Install and repeat. 🎃 

Here’s what’s going on in the world of AI today:

  • Meta open-sources model that supports 1,100 languages.

  • OpenAI proposes an IAEA-like regulatory body for AI.

  • A fake AI image of an explosion at the Pentagon went viral.

  • 66% of The Neuron readers think the EU AI Act goes too far.

OpenAI Proposes International Regulatory Body ⚖️ 

There’s synthetic biology. Then there’s nuclear energy. Then there’s AI.

And according to the cofounders of OpenAI, AI thwarts every other technology in terms of existential risk. In a blog post, they recommend a regulatory body like the International Atomic Energy Agency but for AI.

An agency that could …

  • Require audits.

  • Inspect systems.

  • Place restrictions on deployment.

  • Track computer and energy usage.

… all to make sure that superintelligence doesn’t destroy the world. Sounds like a good idea to us!

The blog makes it clear that smaller projects should not be subject to the same “burdensome” restrictions. This is a divergence from the EU AI Act, which mandates all models to undergo licensing.

Why it matters: It all sounds reasonable, but some worry that OpenAI is only pushing regulation to win a monopoly on the market through “regulatory capture.” In other words, pull the ladder on startup competitors right after winning over the market.

The first half of All-In’s latest pod offers insightful viewpoints on this issue!


Do not click this ad

(At least not all of you at the same time!)

Remember Learn.xyz, the generative AI learning platform where you can ask literally anything? Last week when we featured them, you all broke their site.

So they said: Heck, let’s do it again!

Try it out yourself and skip right past their waitlist with our secret link.

"PS: They are also hiring in their SF office"

New Meta Model Supports > 1,000 Languages 🌎 

Today, most speech recognition tools only support roughly 100 of the world’s 7,000 known languages.

Meta unveiled a speech-to-text (& vice versa) model that supports 1,100 spoken languages and identifies another 3,000. It’s secret? The Bible.

Meta researchers trained the model on vast swathes of text from the New Testament (which contain the most publicly available speech data in the world).

While the data comes from religious texts, the model performs well across non-religious domains too. It’s also more accurate: Compared to OpenAI’s Whisper, Meta’s model has ½ the number of errors while covering 11x more languages!

Meta also released the code to the public in pre-trained models on GitHub, which could mean upcoming advancements in areas that deal with languages like:

  • Content subtitling.

  • International communication.

  • Translation services (ahem, Google Translate)

The future roadmap involves increasing the model’s coverage of even more languages and supporting different dialects. Does this include Dothraki?!

Why it matters: Meta is on fire, churning out innovative open-source models weekly. Watch out, OpenAI!

Around the Horn 🦄

  • Intel announced a generative AI model for science called Aurora that’s targeting 1 trillion parameters.

  • ChatGPT had to fight through thousands of clones to reach the top of the App Store charts.

  • Researchers created LIMA, a new language model that performs better at complex tasks than GPT-4 and Bard with very, very little training data

  • An AI-generated image of an explosion at the Pentagon went so viral that it triggered a brief dip in the stock market.

  • Ask HN: Those with success using GPT-4 for programming – what are you doing?

Treats To Try 🍿

  • Featured: Use Scribe AI to write your user guides, process overviews, and SOPs. Record your screen while walking through a process and automatically turn it into a step-by-step guide.

  • Featured: Ogimi is an AI-powered personal meditation coach that makes meditation easy with customized sessions just for you. Try for 50% off 3-months today!

  • Quizbot generates quizzes from your content in seconds using AI.

  • Eraser’s DiagramGPT is a whiteboard for engineering teams that generates diagrams from prompts.

  • Horizon AI identifies, configures, and manages the best LLMs in <10 lines of code for unique use cases.

  • Get a summary of any article online with just 2 clicks using the TLDR AI iOS app.

DM us links on Twitter: @nonmayorpete & @noahedelman02.

Tuesday Ticker ⏰

Here are the results from last week’s poll:

Thanks for all of the thoughtful responses! They help us learn and get better.

Now, for this week’s poll:

Is AI the greatest threat to humanity we've ever seen?

Login or Subscribe to participate in polls.

A Cat's Commentary 😻

That's all we have for today. See you cool cats on Twitter if you're there: @nonmayorpete & @noahedelman02.

What'd you think of today's email?

Login or Subscribe to participate in polls.