The Rundown AI
Posts
OpenAI's new AI safety plans

OpenAI's new AI safety plans

PLUS: AI predicts early death risks from life history

Rowan Cheung
December 19, 2023

Sign Up | Advertise | Jobs | Tools

Welcome, AI enthusiasts.

OpenAI just released a new framework for AI safety — including enhanced vetting and granting the new board power for models it deems too risky.

Are these proactive moves for the future, or a guardrail for a potentially imminent AGI? Let’s get into it…

In today’s AI rundown:

OpenAI empowers board to block risky AI
AI trained on life histories predicts early death risk
9 new AI tools to try
How to transform your video style with DomoAI
Deloitte using AI to PREVENT layoffs
Imran Khan’s AI voice clone rallies from jail
Study: Bots now beating humans at CAPTCHAs

Read time: 4 minutes

LATEST DEVELOPMENTS

OPENAI

🤖 OpenAI empowers board to block risky AI

Image source: OpenAI

The Rundown: OpenAI just published a new safety preparedness framework for managing AI, including granting the (newly reconstructed) board power to reverse executive decisions on releasing concerning models.

The details:

A dedicated "Preparedness" team will continually probe capabilities and risks then issue reports to advise leadership and directors.
The board can now override judged-safe rollouts if still uncomfortable following OpenAI's tumultuous leadership crisis.
The framework formalizes existing processes, bringing heightened accountability and increased rigorous vetting across security, weapons, and persuasion threats.
An OpenAI employee quote-tweeted the announcement saying (perhaps jokingly), “brace yourselves, agi is coming”.

Our thoughts: While the OpenAI board drama turned out to be less about safety and more about power, the issue is still at the forefront of AI researchers’ minds. This emphasis on safety is a positive step towards transparently steering AI ethically — especially with AGI on the horizon.

TOGETHER WITH AE STUDIO

🔥 Don’t lose the AI race, hire AE Studio

The Rundown: AE Studio is the key to industry dominance, securing AI talent from Harvard, Stanford, and MIT to streamline operations.

What AE Studio will do for you:

Turbocharge your business, save hundreds of hours, and WIN!
Have AE build custom your custom software and AI solutions
Hire AE to pinpoint where and why you should be building NOW!

Stop waiting and start winning by scheduling your FREE strategy session with AE Studio today to take control of your future in the AI race. Book now!

AI RESEARCH

🔮 AI trained on life histories predicts early death risk

Image source: Midjourney

The Rundown: An AI model trained on the life events of 6 million Danish citizens proved highly accurate at predicting mortality — outperforming tools traditionally used by insurers in assessing risk.

The details:

By ingesting health, work, and finance records, the system can estimate individual outcomes — including personality and chance of dying by a certain age.
It bested other models (including ones typically used by insurance companies) by 11% in predicting life expectancy over four years.
Researchers believe doctors could use it to flag health issues sooner for intervention or by governments to help reduce inequality.
The lead scientist warns the tech could empower profit-seeking predictions by big business, though basic versions already drive insurance decisions.

Why it matters: Research like this is both exciting and terrifying — with prediction models getting scary accurate when fed massive amounts of data. The dystopian scenario of reducing lives down to statistics is one that is becoming all too real.

TRENDING TOOLS

🐇 CodeRabbit- AI-powered code reviewer that provides context-aware feedback, line-by-line code suggestions, and real-time chat that gets smarter with usage (link)*

🌄 Pikzels AI- Create AI-generated thumbnails effortlessly (link)

📅 Agenda Hero Magic- Creates AI-generated calendar events (link)

💌 PONS.ai- Impress your network with an AI-generated festive greeting card (link)

🧠 Mental- A mental health app designed specifically for men (link)

💼 EarnBetter- Optimize your job hunt with a complimentary AI-powered assistant (link)

🧠 Humata- Ask questions and chat with any file instantly using AI (link)

👩‍🎨 Looka- AI-powered platform to design a logo and build a brand in seconds (link)

📝 PDF AI- Chat with any document (link)

Browse the most popular tools ever featured with our tool database.

* = This is sponsored content

AI TRAINING

🎥 Transform your video style with DomoAI

Want to take an existing video and transform it into a completely different style, all with the click of a button? A new platform called DomoAI recently launched a video-to-video feature, with some pretty amazing results.

Try it yourself by following the steps below:

Join the Discord: Like Midjourney and other AI generators, Domo is based in Discord. Join here to access the video-to-video tool.

Start generating: Navigate to the ‘generate-video’ channel, and type /video. Drop in the video file you’d like to stylize, then type a prompt that describes your video.

Pick your style: Once you hit enter, a message from the DomoAI bot will pop up, where you can select different styles and additional settings for the video output. Choose your desired selections and hit start.

Your video should be generated in a few minutes within the channel! Re-generate and alter the prompts/settings as needed to get the perfect output!

QUESTION OF THE DAY

What AI training tutorials would you like to see more of?

Vote below (click for live results)

QUICK HITS

💼 Deloitte using AI to PREVENT layoffs

While the common fear is AI taking jobs, consulting giant Deloitte is leveraging the tech to prevent layoffs — utilizing the tools to assess skills and move workers into more in-demand roles. Part of a wider AI push, Deloitte hopes predictive analytics can help moderate hiring swings and prevent cuts.

🗣️ Imran Khan’s AI voice clone rallies from jail

Jailed opposition leader Imran Khan just tapped AI to virtually campaign for Pakistan's election, with his party using text-to-speech to mimic Khan's voice. The 4-minute message was part of a “virtual rally” hosted on social media, utilizing ElevenLabs’ AI voice tech for the clone.

🤖 Study: Bots now beating humans at CAPTCHAs

We might need a new form of bot detection. A new study found that automated bots can now outperform humans at all tested types of CAPCHAS in both solving time and accuracy. The harder problems didn’t trick the bots — but did lead to many frustrated users abandoning the task.