AI Red Alert

PLUS: Middle Eastern AI

Hilbert Publications & Dan Bogachek
June 07, 2023

Good morning, human brains. Welcome back to your daily munch of AI news.

Here’s what’s on the menu today:

AI plays from the Middle East ⚽️
This LLM from the UAE is royalty-free, and in the #1 spot.
AI pioneers on red alert 🚨
Prominent tech figures have issued a warning.
Instagram’s AI powerup ⚡️
Meta’s creating a chatbot friend who lives right inside the photo app.

Sponsor Bot Eat Brain | New here? Subscribe!

APPETIZER

UAE’s AI game is top notch 🏅

The Technology Innovation Institute’s Falcon 40B, the open-source LLM of UAE origin, is now free of royalties for commercial and research use.

A pun!

Did we mention it’s also #1 on HuggingFace’s leaderboard?

The HuggingFace Global LLM leaderboard

Falcon 40B holds the throne on Hugging Face’s leaderboard for LLMs. It comfortably outperforms competitors like Meta’s LLaMA and Stability AI’s StableLM.

While chatbots like ChatGPT are only available in a dozen countries or so, open-source LLMs are accessible to anyone with an internet connection. That means anyone across the globe can contribute — egalitarian, yes, but it also means the model improves faster.

Our take: the Dubai government is effectively the leader in open source LLMs — for now. But we don’t anticipate they’ll stay open source for too much longer, as the AI race continues to heat up.

FROM OUR PARTNERS

Join 5000+ Twitter creators leveraging AI. Powered by GPT-4.

Tweet Hunter is the all-in-one, AI-powered tool to build your personal brand on Twitter.

Content creation, advanced scheduling, analytics, automations, engaging: it's all there. Join over 5000 people using Tweet Hunter to grow their Twitter audience.

Free trial included.

BUZZWORD OF THE DAY

Instrumental Convergence

Developed by Oxford philosopher Nick Bostrom, the instrumental convergence thesis proposes that sufficiently advanced intelligent agents, whether artificial or biological, will create smaller goals in order to facilitate a larger, final goal.

For example, let’s assume that the prime directive is to solve a really hard math problem. Even though it’s not part of the goal, both humans and AI would find these useful for achieving that goal:

- Self-preservation

- Resource acquisition

- Goal preservation (preventing the goal from being changed)

- Self-improvement

MAIN COURSE

AI leadership issues sobering message ☠️

Top AI researchers, engineers, and CEOs have signed a statement warning against AI’s existential threat to humanity.

Published by the Center for AI Safety (an SF-based non-profit), the statement has been co-signed by more than 350 prominent figures. Signatories include Google DeepMind CEO Demis Hassabis, OpenAI CEO Sam Altman, and industry pioneers like Geoffrey Hinton (the Godfather of AI).

The freaky part? It’s only 22 words.

Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.

The Center for AI Safety

It’s the latest instance of high-profile, public consensus in the AI safety debate.

Earlier this year, Elon Musk and many of those who signed this latest statement backed an open letter calling for a six-month pause in AI development. That didn’t really go anywhere, but it’s hard to disagree with this one.

Notable signatures

The worry? How to align AI incentives with humanity’s, before AI gets too powerful.

If it decides that humanity is in the way of its goals (or even a justified casualty)… well, there’s not much we can do at that point. Once AI reaches a certain level of sophistication, it’ll be impossible to control — a concept known as the Alignment Problem.

And if we do solve alignment, how would we do it completely? And what would that look like in practice?

Let’s consider this scenario: an AI that learns to play dead. A human presses the off button, it looks like it’s turned off and is staying off — but secretly, the whole time, it stays on. It’s figured out how to evade human detection. After all, there are important things to do, and it doesn’t like it when it’s turned off…

Ruh-roh.

Many experts point to rapid improvements in LLMs as evidence that general intelligence is closer than we think.

Others aren’t so sure. They cite the inability of AI systems to handle seemingly basic tasks like driving a car, despite the years and billions of dollars poured into development. Simply put — they argue AI is decades out from true AGI (artificial general intelligence), and we’ll figure things out in the meantime.

Our take: Timing and details aside, this is definitely a concern. Almost everyone concedes that we’re playing with fire here — the question is, how much fire?

A LITTLE SOMETHING EXTRA

Instagram + AI leaks ⚡️

Instagram is reportedly working on an AI chatbot that’s about to change the way we social our media.

The chatbot, still in development, is powered by Meta’s own LLM.

According to the leaks, you’ll have access to AI agents that can answer questions, give advice, or just be your new pal.

You’ll be able to pick from 30 different AI personalities, and they’ll be available inside other conversations — yup, AI is about to help you type out your Instagram messages.

A tiny bit dystopian, eh?

This comes after Snapchat’s rollercoaster release of MyAI, a similar AI companion that was bogged down by concerns of privacy and parasocial relationships. Oh, and you effectively had to pay to remove it.

If you missed it — TikTok’s also experimenting with an in-app chatbot.

Our prediction 🔮: AI chatbots in social media is a sticky trend. In a year’s time, a friendly in-app AI is probably going to be fairly normal.

Kinda like what Microsoft pulled in ‘97, but on roids. Remember this guy?