OpenAI Publishes Elon Musk's Private Emails 💥

Welcome to another edition of Horizon AI,

OpenAI leadership responds to Elon Musk's accusations and releases 8-year-old private emails where the billionaire appears to contradict his current position.

Let's get into it!

Read Time: 3.5 min

Here's what's new today in the Horizon AI

• Open AI Fires Back at Elon Musk by Publishing His Private Emails 💥

• AI Research: TripoSR Generates 3D Models in Under a Second

• AI Tutorial: How to analyze files for free with GPT-4 using Microsoft Copilot

• AI Tools to check out

• The Latest in AI and Tech 💡

AI News

OPENAI

Open AI Fires Back at Elon Musk by Publishing His Private Emails 💥

OpenAI has responded forcefully to Musk's recent lawsuit alleging breach of contract.

In a blog post published on Tuesday, OpenAI stated it will seek to dismiss all of Musk's claims. The post provided OpenAI's perspective on the events leading up to the lawsuit.

Details:

• OpenAI says Musk, co-founder of the company, at one point demanded "absolute control" by merging with Tesla, seeking majority equity, board control, and the CEO role

• They rejected these terms, viewing individual control as against OpenAI's mission

• OpenAI defends keeping some work closed-source for safety around AGI

The post included a 2016 email exchange where Musk appeared to agree that some OpenAI work could be kept private as it advanced toward AGI development. He also acknowledged the necessity to pivot to a for-profit structure to raise the billions necessary to achieve the company’s mission.

Musk's lawsuit last week claimed OpenAI strayed from its principles by becoming too financially motivated and too closely tied to Microsoft.

AI RESEARCH

TripoSR Generates 3D Models in Under a Second

Researchers from Stability AI and Tripo AI have unveiled TripoSR, an AI model capable of reconstructing 3D objects from a single 2D image in less than half a second on an Nvidia A100 GPU.

TripoSR represents a significant advancement in both quality and speed compared to existing open-source alternatives for 3D reconstruction.

How it works: TripoSR processes an RGB image via a vision-transformer-based encoder, transforming it into latent vectors. These vectors are then converted into a Triplane-NeRF representation for 3D reconstruction by a decoder.

This technology holds vast potential for applications across multiple industries, including entertainment, gaming, architecture, and beyond.

There is currently a demo on Hugging Face. Additional examples, along with the code and model, can be found on GitHub.

AI Tutorial

How to analyze files for free with GPT-4 using Microsoft Copilot

Copilot now allows you to upload your own documents for analysis.

*Note: The feature is gradually rolling out, so it may not yet be available for you

1. Go to Microsoft Copilot

2. Click on the paperclip icon to upload a file or drag and drop them into the chat box.

Here are some examples of how you can use this new feature:

• Chat with PDFs: You can upload your PDF and request Copilot to summarize, explain, or extract specific information from it

• Upload Excel or CSV files for analysis. You can ask Copilot to generate formulas or create graphs with the data.

AI Tools to check out

✨ Pixpic: Create professional-quality and life-like headshots or profile pictures online.

👔 OutfitGPT: Upload your photo and receive fashion tips from Italian stylist Laura Sigona.

🔍 Octomind: Provide your website URL and discover bugs before users with the help of AI.

📈 Just words: Boost user growth with continuous copy optimization.

📚 Reading Club: Transform bedtime stories into interactive adventures with AI.

The latest in AI and Tech

Google targets spam and AI content in search results

Google is rolling out changes to its search ranking systems aimed at surfacing high-quality content while suppressing spam and content generated by AI tools that simply repackage information from other sources. The company states it will more effectively combat techniques used to manipulate search rankings artificially.

AI-fueled political disinformation escalates rapidly, study says

A new study from the Center for Countering Digital Hate finds that AI-generated disinformation, particularly deepfake images related to elections, has seen a staggering 130% monthly increase on Twitter over the past year. The CCDH warns that this proliferation of misleading content, enabled by freely available AI tools and insufficient social media moderation, poses a severe risk to electoral integrity.

Researchers test leading AI models for copyright infringement, GPT-4 performed the worst

At Patronus AI, an AI evaluation company, researchers conducted tests using their new CopyrightCatcher tool to assess how often prominent language models produce copyrighted text. The study tested OpenAI's GPT-4, Anthropic's Claude 2, Meta's LLaMA 2, and Mistral AI's Mistral model. Notably, GPT-4 exhibited the highest rate of copyright violation, responding with copyrighted material to an average of 44% of the prompts.

IBM reports productivity gains from Adobe AI integration

IBM announced on Wednesday that their initial testing of generative artificial intelligence tools from Adobe has resulted in productivity improvements within their marketing operations. The company stated that it was utilizing Adobe's tools, capable of generating images from text-based prompts, to aid in marketing campaigns.

Microsoft’s small math AI model does math better than models 10 times its size

Microsoft Research has unveiled Orca-Math, a new variant of Mistral's Mistral 7B language model, that exhibits exceptional proficiency in solving mathematical word problems while maintaining a relatively small model size. Remarkably, Orca-Math outperforms models with ten times more parameters

That’s a wrap!

Thanks for sticking with us to the end! Let’s stay connected on LinkedIn and Twitter.

Not subscribed yet? Sign up here and send it to a colleague or friend!

See you in our next edition!

Gina 👩🏻‍💻