Welcome to this week’s roundup of elegantly handcrafted AI news made specifically for biological audiences.
This week AI took us to church.
AI models made a human model.
And world powers shift their AI chess pieces into place.
Let’s dig in.
What did Ilya see?
After last week’s drama things at OpenAI seem to have chilled down with Sam Altman and Greg Brockman back at their old jobs. We’re still trying to work out exactly why Altman was fired in the first place and the rumor mill is working overtime.
From letters leaked on 4chan to whispers from anonymous sources we might be onto something here. Could OpenAI be sitting on a dangerous AI model that led to Altman’s firing? Is this what made Ilya Sutskever nervous?
You can’t spell faith without AI
A convicted self-driving car engineer who was pardoned by Trump just restarted his AI church. No, seriously. This article about Anton Levandowski’s AI church is the God’s honest truth.
Effective altruists also want to save our souls from AI. Effective altruism, long-termism, and politics are having a significant effect on how OpenAI operates.
AI has given us some new words as well as new meanings to old words. Sam’s deep dive into AI, language, and culture in the Library of Babel is a fascinating look at how AI is changing the way we speak and think.
“Authentic” isn’t a new word but Merriam-Webster chose it as their word of the year. Here’s why.
What’s real anymore?
Authenticity is taking a hammering as deep fakes surrounding the Israel-Palestine conflict intensify. Both sides are pumping out AI fakes and the experts say it’ll only get worse as AI gets better.
Students are sharpening their AI skills too but not in the way you’d think. AI-generated sexually explicit material is spreading in schools. A proliferation of “declothing” apps is making it easier to do, and harder to stop.
Fashion models and influencers are infamous for their Prima Donna behavior and outrageous demands. A Spanish agency solved the problem by creating an AI-generated model. She’s pretty hot and you won’t believe what she earns.
AI models in the fast lane
Even some of the data used to train AI models is fake. As data scientists scramble for more training data these researchers used synthetic data to train an AI image classifier. It works better than models trained with real images.
Improvements in AI vision models and humans struggling to keep up saw Formula 1 racing trial an AI system to enforce track limits for the first time.
If your chatbot gives you silly responses it may be because your poor attempt at writing prompts is confusing it. Meta scientists came up with something called System 2 Attention which fixes this issue in a fairly simple way, albeit with a catch.
LLMs may be getting more accurate but they’re still easy to jailbreak. A new study reveals new techniques for jailbreaking language models by pitting them against each other.
Will Stability sell out?
Stability AI’s company name is more an expression of its aspirations, than a description of its financial state. The company’s investors are getting nervous and there’s talk that Stability AI may be looking for buyers.
Cash crunch aside, Stability AI keeps delivering state-of-the-art AI products. Its new real-time text-to-image generator is just ridiculous. Now they just need to find a way to make money from it.
Introducing SDXL Turbo: A real-time text-to-image generation model.
SDXL Turbo achieves state-of-the-art performance with a new distillation technology, enabling single-step image generation with unprecedented quality, reducing the required step count from 50 to just one.
The… pic.twitter.com/0NA4aUqKkD
— Stability AI (@StabilityAI) November 28, 2023
Google material
If you want to discover new materials now you can just Google it. DeepMind’s GNoME system has discovered millions of new materials with some interesting implications.
If you’re a fan of Google’s phones then you may be due for an upgrade. Google has updated its AI Core app for the Pixel 8 Pro smartphone. If you have this model you may notice improved performance in some of the built-in AI features.
Global positioning
The UK and US cyber security agencies developed new global guidelines for AI security. The guidelines look well thought out, except for the part where it neglects to bind the signatories to follow them.
Russia and China were notably absent from the list of 18 other countries that endorsed the guidelines. Russia has been a bit of a Boris-come-lately to the AI race but Vladimir Putin has signaled Russia’s intentions to spur on AI research and development.
China’s AI march is being hampered by US export restrictions and Nvidia’s workaround isn’t working out as planned. Its new China-friendly H20 chip has hit a few snags.
As the world’s superpowers pit their AI efforts against each other, at least we’ll be a little safer at sea. Scientists have used AI to find an equation to predict rogue waves so hopefully we won’t get any more Poseidon Adventure sequels.
The business end of AI
AI is raising some tricky issues regarding business and taxation. Actors, authors, and other creatives pay tax on their earnings. Where will the state collect that revenue when AI inevitably takes those jobs? The tax breaks UK companies get for adopting AI are making this a tough discussion.
Consumers are going to feel the impact of AI too, for better or for worse. In the US the FTC has authorized new powers of investigation and compliance to hold businesses accountable for how they use AI.
Business leaders are getting excited about how AI can improve the way they run their companies. This week sees an exciting lineup of Norway’s tech leaders presenting the latest AI business solutions at the Nordic AI Summit.
In other news…
Here are some other clickworthy AI stories we enjoyed this week:
- When Microsoft tried putting an AI team together from engineers jumping ship from OpenAI it equipped them with Apple Macs rather than its own Surface devices. Not a great look for the Microsoft brand.
- The International Mathematical Olympiad (IMO) is offering $5 million to the first publicly-shared AI model to win a gold medal in an IMO event.
- American spy agencies have warned about the Emirati AI giant G42 and its work with large Chinese companies.
- AI is creating jobs, not replacing them. However, AI is having a negative effect on wages.
- Context window isn’t everything. Check out GPT-4 Turbo recall vs Claude 2.1.
This Claude 2.1 vs. GPT-4-Turbo chart absolutely blows my mind. Insane levels of fidelity from @OpenAI.
Credit: @GregKamradt pic.twitter.com/PCSOgsTjRt
— Steve Moraco (@SteveMoraco) November 22, 2023
And that’s a wrap
Do you think Q* is the next big thing we’ll see come out of OpenAI or is it being overhyped? Should we be worried or excited? Honestly, I’m feeling both at the same time.
Which side of the effective altruist conversation do you lean to? Surely it’s possible to want the best for humanity and still support speeding up AI development.
We’d love your feedback. Let us know which story was your favorite and which ones we missed.
