The Rundown AI
Posts
Sam Altman speaks on AGI & OpenAI drama

Sam Altman speaks on AGI & OpenAI drama

PLUS: Stability AI's Image-to-3D upgrades

Rowan Cheung
December 15, 2023

Sign Up | Advertise | Jobs | Tools

Welcome, AI enthusiasts.

It’s been quite the end to the year for Sam Altman. Following his TIME ‘CEO of the Year’ award — the now-reinstated CEO just spoke candidly about AGI, AI risks, and more.

Understandably, Altman sounded more stressed about shepherding the super-intelligence to come than the juicy boardroom drama. Let’s get into it…

In today’s AI rundown:

Sam Altman opens up on OpenAI drama
Stability AI releases new image-to-3D model
10 new AI tools to try
How to run powerful AI assistants locally on your phones
Alibaba releases new image-to-video model
OpenAI explores ‘weak’ models controlling strong AI

Read time: 3 minutes

LATEST DEVELOPMENTS

OPENAI

👀 Sam Altman opens up on OpenAI drama

Image source: TIME

The Rundown: OpenAI CEO Sam Altman just gave a candid, wide-ranging interview after being named TIME's CEO of the Year, reflecting on his short-lived firing, governing powerful AI, and why he remains hopeful despite risks.

The details:

Altman felt the boardroom drama brought productive self-reflection on OpenAI's governance needing to support its mission as AI capabilities charge ahead.
He believes AGI will “be the most powerful technology humanity has yet invented”, with huge implications for democratizing access to information globally.
Altman also emphasized the risks of AI personal disinformation, particularly with the upcoming elections in 2024.

The relevance: Altman’s remarks sound like someone more stressed about shepherding a world-altering technology than petty boardroom drama. And despite the risks, Sam sounds confident that AGI and accelerating AI will change the world for the better — sooner rather than later.

TOGETHER WITH MASTERWORKS

🎨 Harvard data scientist upends finance community with opportunities in this market

The Rundown: The AI revolution has already begun to rewire Wall St, and its impact has been strongly felt in one growing market in particular. Because, thanks to a Harvard data scientist and his crack team, everyday people can now benefit from a previously “off-limits” investment.

The company that makes it possible is called Masterworks, whose unique investment platform enables savvy investors to invest in blue-chip art for a fraction of the cost.

Intrigued? The Rundown AI readers can skip the waitlist.

Investing involves risk and past performance is not indicative of future returns. See important Reg A disclosures and aggregate advisory performance masterworks.com/cd

STABILITY AI

🤯 Stability AI releases new image-to-3D model

Image source: Stability AI

The Rundown: Stability AI just released an enhanced image-to-3D model called Stable Zero123, alongside a new tiered membership structure that aims to stabilize the company’s revenue generation.

The details:

The model can take a single image input and generate a 3D object, displaying subject from a variety of angles.
Zero123 touts higher quality results by utilizing cleaner Objaverse dataset renders and adding camera elevation data.
The open-source startup also announced a new membership model, which includes a free/research tier, a $20/mo professional tier, and custom enterprise pricing.

Why it matters: Stability ups the competition in the image-to-3D AI space, keeping their foot on the gas with yet another impressive model rollout. If the current pace of development keeps up, we will likely have text-to-video game creation within just a few years.

QUESTION OF THE DAY

Given Sam Altman's comments and recent advancements, when do you think AGI will be achieved?

Vote below (click for live results)

TRENDING TOOLS

🛡️ DryRun Security- AI-powered security buddy empowers developers in their everyday workflow (link)*

👚 Outfit Anyone- Try on clothes virtually with AI (link)

🔍 Unwrapped AI- Get a Spotify-like recap of your 2023 GitHub activity (link)

🛠️ Scade.pro- Access 1500+ AI tools without coding using this platform (link)

💬 MessengerX.io- Monetize with GPTs for uncensored information and entertainment (link)

📸 Snapby AI- Generate lifelike images directly in Figma for free (link)

🤖 GradesAI- An AI-powered app that forecasts exam questions for students (link)

📝 Copy AI- Generate content, such as blog headlines, emails, and social media posts using AI (link)

🔥 Jasper- An AI copywriter assistant that can generate any kind of marketing copy (link)

✍️ Rytr- An AI writing assistant that helps you create high-quality content, in just a few seconds (link)

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

* = This is sponsored content

AI TRAINING

📱 How to run powerful AI assistants locally on your phone

Want access to powerful AI models on your phone, without needing WiFi or risking your privacy? Now you can with an incredible new app called LLMFarm — letting you run AI models locally on your iPhone or iPad, completely offline.

Here’s a tutorial on how to access it by Alvaro Cintas:

Step 1: Go to LLMFarm TestFlight page and click “Install with TestFlight". Download the TestFlight app first if you don't have it already.

Step 2: Return to LLMFarm and click "Start Testing" to install the app itself.

Step 3: Next, find the pre-trained AI model you’d like to use locally from HuggingFace. For this tutorial, we used the Mistral-7B model. Navigate to the ‘files and versions’ tab and download the model — note that some may take up significant space.

Step 4: Back in the LLMFarm app, select "Add Model" and navigate to the Mistral 7B file you just downloaded. In the ‘format’ section, type: <s>[INST] [INST]. In prediction options, toggle on Metal and MLock.

QUICK HITS

📝 Take a free AI for Productivity workshop (sponsor)

Join Section’s free workshop to learn how to use AI to be more efficient at communication, problem-solving, and brainstorming.

Click here to enroll for free.

🎥 Alibaba releases new image-to-video model

Alibaba Group just released i2VGen-XL, a new image-to-video model capable of generating high-definition ouputs. The release brings yet another competitor to the ever-expanding AI video generation sector, with capabilities ramping up at a truly insane pace.

🦾 OpenAI explores ‘weak’ models controlling strong AI

OpenAI just released a new paper, introducing the concept of using small AI models to supervise more powerful ones — studying if they can generalize training despite limited capabilities. Aiming to tackle aligning superhuman AI, experiments show promising control of LLMs.