Why OpenAI's safety leaders are leaving

Read time: under 4 minutes

Welcome back, Superhuman

The defining motto of Silicon Valley is, “Move fast and break things.” But when it comes to super-intelligent AI systems, some industry leaders are fighting to do just the opposite: Move slowly so everyone stays safe. That debate reached OpenAI’s doorstep this week.

Today’s Insights

• An exodus at OpenAI raises questions about AI safety

• Prompt: Social Media Posting

• Salesforce wants to solve AI’s data problem

• 5 new AI tools to boost your productivity

• Everything else you should know today

• Chart: How workers are using AI

• AI-Generated Images: Vintage Chairs

NEXT IN AI

Why the heads of OpenAI’s safety team are stepping down

Executives leave companies all the time. But when it’s the person tasked with preventing one of the world’s most powerful AI systems from becoming self-aware and going rogue…it’s only fair that onlookers might raise their eyebrows.

After around a decade at OpenAI, chief scientist and co-founder Ilya Sutskever announced last week he’s stepping down. Sutskever’s work as a graduate student at the University of Toronto helped shape the field of generative AI. But no less important, he co-launched OpenAI’s superalignment team last year.

That group’s goal is to make sure that even as OpenAI runs full speed toward artificial general intelligence — the point when AI becomes better than humans at most tasks — that mission won’t accidentally “lead to the disempowerment of humanity or even human extinction.”

Here’s what went down:

• Soon after Sutskever announced his resignation, his co-leader, Jan Leike, revealed he was quitting, too.

• Leike said he and the company had reached a “breaking point” because he thought it wasn’t spending enough on preparedness.

• “These problems are quite hard to get right,” Leike wrote on X. “And I am concerned we aren’t on a trajectory to get there.”

• Sources told TechCrunch the superalignment team was severely under-resourced — despite OpenAI’s promise to dedicate a fifth of its compute to safety.

It’s a tricky balance: OpenAI wants to build a super-intelligent system because it believes it’s best-equipped to handle that technology with care. Other companies feel the same way about their own AI initiatives. That competition can help drive innovation, but it might also lead AI firms to quickly push out new products — even when a more cautious approach would be safer.

A counterpoint: OpenAI gets props for speaking openly about potential risks and investing millions to make sure those dangers don’t pan out, something other AI companies have been reluctant to do. Also, it’s getting easier to build a model from scratch in your basement. OpenAI’s proponents argue it’s much better to have a well-intentioned company developing a potentially dangerous technology — rather than someone who has more malicious goals in mind.

PRESENTED BY AE STUDIO

Hire a world class AI team

Trusted by leading startups and Fortune 500 companies

Building an AI product is hard. Engineers who understand AI are expensive and difficult to find. And there's no way of telling who's legit and who's not.

That's why companies around the world trust AE Studio. We help you craft and implement the optimal AI solution for your business with our team of world class AI experts from Harvard, Stanford and Princeton.

Our development, design, and data science teams work closely with founders and executives to create custom software and AI solutions that get the job done.

Book a free consultation session today. Get in touch here

PROMPT OF THE DAY

Social Media Scheduling

Prompt: Create a social media schedule: You are an expert social media manager. I want you to create a schedule for social media posts over one month, starting from [insert date that schedule will start]. The frequency of posting will be [daily/every two days/every weekday/weekly]. My business is called [insert name] and we sell [insert products or services]. For each post, include the day it will be published, a heading, body text and include relevant hashtags. The tone of voice we use is [professional/casual/funny/friendly]. For each post, also include a suggestion for an image that we can use that could be found on a stock image service.

You can adapt the prompt to your specific needs.

_{Source: Forbes / Bernard Marr}

SPONSORED BY SALESFORCE

Salesforce wants to solve AI’s data problem

Nearly all business executives say they’re feeling the heat to incorporate LLMs into their workflow. AI automates certain tasks that might have once taken hours to complete. And it can reveal insights that were hidden under piles of complex data.

So why have so many companies held off on adopting LLMs? Because more than half of C-suite leaders across nine countries say they don’t yet trust AI to give accurate and helpful information, according to a new Salesforce-YouGov study.

The problem comes down to data: Say you’re working on a presentation to show why there’s demand for a new campaign — but you cite a study that’s decades old. If you’re working with bad information, the final product will probably suffer, too. Salesforce says it’s the same with AI. At this point, a lot of companies use LLMs that are trained on statistics or information gathered from third parties. That data might not be relevant to a firm’s needs.

So, what’s the solution? Salesforce thinks the answer lies in AI-powered virtual assistants like its Einstein Copilot that are grounded in a company’s proprietary information to help solve problems. If you want to figure out how a new product might perform, for example, you can ask Einstein Copilot to analyze customers’ behavior during previous launches. It also includes built-in safety measures that keep your data safe and secure. Eight in 10 executives say that’s the biggest factor that would help them place more trust in AI. As Salesforce’s SVP John Kucera puts it, “When AI is grounded in a company’s own complete and accurate data, it creates greater trust in the technology, which drives more adoption and ultimately, ROI.”

PRODUCTIVITY

5 AI Tools to Supercharge Your Productivity

✅ fforward ai: Analyze customer interviews and discover valuable opportunities that tie your customers together.

✅ Algomo: A custom chatbot that helps you generate leads and automate customer engagement.

✅ Phew AI Tab: Organize, manage, and retrieve your tabs via AI-based grouping.

✅ TestSprite: AI-powered, fully automated testing solutions so you can launch 10x faster.

✅ Prompt Whisperer: Tailored prompts for professionals helping you to generate code, craft emails, design social media content, and manage data with ease.

^{* indicates a promoted tool, if any}

AI & TECH NEWS

Everything you need to know today

• Win-Win Scenario: Reddit finalized a deal that lets OpenAI train its models with the platform’s content. In exchange, Reddit can now use ChatGPT to build new AI features.

• Against the Grain: Even as other US-based AI companies flee China, Elon Musk is reportedly planning to build a data center there to fuel Tesla’s self-driving AI projects.

• Robo-Reporters: Gannett — which owns more than 1,000 news outlets, including USA Today — announced it’s going to start adding AI-generated summaries to the top of some articles.

• Can’t Sample This: Record label giant Sony Music Group sent out 700 letters to AI startups warning them they’re forbidden from using its artists’ music to train new models.

• Going Long: Short-form social media platform TikTok is reportedly experimenting with hour-long videos.

CHART

How have employees been using AI at work?

A recent Gallup study of around 19,000 American workers found that, among those who frequently interact with AI at work, 43% use it to complete basic tasks. Nearly a third use AI for learning new things, while a quarter use it for identifying problems.

So far, workers’ interactions with AI have been mostly solitary: Only 11% say they use it to collaborate with coworkers. That could soon change as tech companies release more tools to summarize meetings and brainstorm with colleagues.

AI GENERATED IMAGES

Vintage Posters

Prompt: a Barcelona chair modernist poster design, shades of white, green and black, print, graphic design

--ar 4:5

Acquire new customers and drive revenue by partnering with us

Superhuman is the world’s biggest AI newsletter for businesses and professionals with 650,000+ readers working at the world’s leading startups and enterprises. Companies like Amazon, Calendly, and Notion feature their products in Superhuman. You can learn more about partnering with us here.  

🧞 Your wish is my command 

Thanks for reading.

Until next time!

Zain & the Superhuman AI team

p.s. If you liked this newsletter, share it with your friends and colleagues here.