😺 Mistral AI

PLUS: we tried something new today....

Welcome, humans.

New section alert: upload a picture of you reading The Neuron on your phone or computer and we’ll round up some of your submissions next Friday. No AI-generated photos allowed—you can’t fool us! Here’s Noah’s setup from yesterday:

Evanston, Illinois

Submit your pics here!

Here’s what you need to know about AI today:

The AI ecosystem in France is booming right now.
How to use Mistral 7B on your iPhone or laptop.
Lawmakers want to ban unconsented AI deepfakes.
Zuck is pivoting Meta's focus from the Metaverse to AI.

*P.S. we’re off on Monday for MLK day—have a great weekend!

Meet the 22-person company giving OpenAI a run for its money.

This might just be the most underrated and lowkey player in AI:

Based in Paris (!).
Raised $415M at a $2B valuation (!).
Only 22 employees (!).

That’s right—we’re talking about Mistral AI!

not exactly the most intimidating team we’ve ever seen

It’s a startup building AI foundational models like ChatGPT-4, except Mistral’s models are open-source, meaning total public access—anyone can download and use them free of charge.

Mistral stands out in a bunch of big ways:

First, it delivered the market's best open-source AND free chatbot in less than 7 months. One benchmark even showed Mistral Medium outperforming Claude and ChatGPT-3.5 (the free version).

TBH, it’s a bit tricky to get Mistral up and running.

The easiest way to use it is on Poe. You can also download it on your MacBook Pro or try this new $1.99 iPhone app called Offline Chat that lets you use Mitral locally (i.e. with no internet).

Second, it’s Parisian (= from Paris). It’s pretty uncommon to see AI unicorns outside of SF right now, much less the world’s croissant capital, but Mistral is part of what’s become an incredibly impressive AI ecosystem in France:

Third, Mistral DGAF (doesn’t give a f*ck). Forget polished PR—this company dropped its latest chatbot as a torrent link on X with no context, caption, or marketing.

Yet, its nonchalant attitude has stirred up some controversy, particularly around the implications of making potent AI models like Mistral's freely available.

The concern is that open-source could potentially arm bad actors with powerful AI tools for nefarious purposes, like spewing hate speech.

On the other hand, most AI experts we know advocate for open-source as a cornerstone for collaboration, innovation, and transparency in AI. Case in point: Perplexity’s latest models for (our fav) AI search engine were built on top of Mistral and Llama.

P.P.S. We've added more visuals and tweets in today's edition—tell us how you like this new format!

Should we have more or less images in stories?

FROM OUR PARTNERS

Unlock Mistral's latest model for a 30x cheaper.

You get it—Mistral’s models are absolutely smashing the leaderboards right now.

And OctoAI is the fastest and easiest place to run Mistral’s models, including its latest marvel, Mixtral 8x7B.

It’s 4x cheaper per token than running GPT-3.5 Turbo and 40x cheaper than running GPT-4-Turbo, plus you get:

Ultra-fast response times.
Infinitely scalable infrastructure.
High accuracy & the flexibility of open-source.

Check out how Mixtral performs and sign up for OctoAI to receive free tokens to put it to the test.

Around the Horn.

How AI replaced the Metaverse as Zuckerberg’s top priority for Meta.
Lawmakers are trying to ban the creation of unconsented AI deepfakes.
More than 1/3 of all VC money in 2023 went into in AI startups.
An “Average day in France” according to DALL·E 3.

Intelligent Insights.

Everyone and their mom has an opinion on AI—here are some of the more legit ones we came across this week:

The Architect of Microsoft’s Vaunted Cloud Bundle Turns His Gaze to AI (link).
Signs and Portents—some hints about what the next year of AI looks like (link).
5 Predictions About How AI Will (and Won’t) Affect the 2024 U.S. Election (link).
Andrew Ng: ‘Do we think the world is better off with more or less intelligence?’ (link).
Shaping the future of advanced robotics by Google DeepMind (link).