OpenAI's secret hack exposed

PLUS: China shows off major AI progress

Author

Rowan Cheung
July 08, 2024

Welcome, AI enthusiasts.

OpenAI has been increasing its safety efforts recently, but is the company’s AI secret sauce already out in the wild?

A 2023 hack of the AI giant was just uncovered, raising major questions about what info may have fallen into the wrong hands. Let’s investigate…

In today’s AI rundown:

  • OpenAI's secret AI details stolen in 2023 hack

  • Chinese AI firms flex muscles at WAIC

  • Control AI videos with Luma Labs’ Keyframes

  • DeepMind’s JEST supercharges AI training

  • 6 new AI tools & 4 new AI jobs

  • More AI & tech news

Read time: 4 minutes

LATEST DEVELOPMENTS

OPENAI

Image source: Midjourney

The Rundown: A new report from the New York Times just revealed that a hacker breached OpenAI’s internal messaging systems last year, stealing sensitive details about the company’s tech — with the event going unreported to the public or authorities.

The details:

  • The breach occurred in early 2023, with the hacker accessing an online forum where employees discussed OpenAI’s latest tech advances.

  • While core AI systems and customer data weren’t compromised, internal discussions about AI designs were exposed.

  • OpenAI informed employees and the board in April 2023, but did not disclose the incident publicly or to law enforcement.

  • Former researcher Leopold Aschenbrenner (later fired for allegedly leaking sensitive info) criticized OpenAI’s security in a memo following the hack.

  • OpenAI has since established a Safety and Security Committee, including the addition of former NSA head Paul Nakasone, to address future risks.

Why it matters: Is OpenAI’s secret sauce out in the wild? As other players continue to even the playing field in the AI race, it’s fair to wonder if leaks and hacks have played a role in the development. The report also adds new intrigue to Aschenbrenner’s firing — who has been adamant that his release was politically motivated.

TOGETHER WITH LOGITECH

The Rundown: Logitech’s Logi AI Prompt Builder streamlines your ChatGPT prompting process — providing a user-friendly interface for creating pre-defined, customizable prompts to save time and effort.

With Prompt Builder, you can:

  • Access commonly used prompt recipes with a single click

  • Tailor prompts by defining tone, complexity, and length

  • Gain access for free with supported Logitech mice and keyboards

Learn more and start building powerful prompts today.

WAIC

Image source: CGTN

The Rundown: The World Artificial Intelligence Conference (WAIC) took place this weekend in Shanghai, with Chinese companies showcasing significant advances in LLMs, robotics, and other AI-infused products despite U.S. sanctions on advanced chips.

The details:

  • SenseTime unveiled SenseNova 5.5 at the event, claiming the model outperforms GPT-4o in 5 out of 8 key metrics.

  • The company also released SenseNova 5o, a real-time multimodal model capable of processing audio, text, image, and video.

  • Alibaba's cloud unit reported its open-source Tongyi Qianwen models doubled downloads to over 20M in just two months.

  • iFlytek introduced SparkDesk V4.0, touting advances over GPT-4 Turbo in multiple domains.

  • Moore Threads showcased KUAE, an AI data center solution with GPUs performing at 60% of NVIDIA's restricted A100.

Why it matters: If China’s AI firms are being slowed down by U.S. restrictions, they certainly aren’t showing it. The models and tech continue to rival the leaders in the market — and while sanctions may have created hurdles, they may have also spurred Chinese innovation with workarounds to stay competitive.

AI TRAINING

The Rundown: Luma Labs’ new Keyframe feature allows you to add both starting and ending keyframes to guide AI video generation, increasing control over the final output.

Step-by-step:

  1. Head over to Luma Labs and log in, or sign up for free.

  2. Set starting and ending keyframes using the provided options.

  3. Craft a detailed prompt describing the desired video and transition.

  4. Generate your video and review the results 🎉

PRESENTED BY STACK AI

The Rundown: Stack AI is an AI automation platform developed by two MIT PhDs that lets companies harness the power of artificial intelligence without writing a single line of code.

With Stack AI, enterprises can:

  • Connect AI to their knowledge bases with governance and security.

  • Automate tasks like Staff training, RFP responses, auditing customer calls, and KYC due diligence.

  • Deploy LLMs with SOC 2, GDPR, and HIPAA compliance.

Create an account now and start automating your workflows!

GOOGLE DEEPMIND

Image source: Midjourney

The Rundown: Google DeepMind researchers published new research introducing JEST (Joint Example Selection Technique), a method that accelerates AI model training while significantly reducing computing requirements.

The details:

  • The technique leverages two AI models — a pre-trained reference model and a ‘learner’ model that is being trained to identify the most valuable data examples.

  • JEST intelligently selects the most instructive batches of data, making AI training up to 13x faster and 10x more efficient than current state-of the-art methods.

  • In benchmark tests, JEST achieved top-tier performance while only using 10% of the training data required by previous leading models.

  • The method enables ‘data quality bootstrapping’ — using small, curated datasets to guide learning on larger unstructured ones.

Why it matters: As AI’s massive energy consumption comes under increasing scrutiny, JEST’s ability to drastically cut computing requirements could be a game changer for training models in a more energy-efficient way. Plus, faster training capabilities mean the acceleration of advanced model releases is just getting started.

NEW TOOLS & JOBS

  • 🎨 xAI - Product Designer

  • 💻 Weights & Biases - Programmer Writer, Documentation

  • 📊 DeepL - Enterprise Customer Success Manager

  • 🛠️ Dataiku - Senior Infrastructure Engineer

QUICK HITS

Apple’s Siri upgrades are now expected to come in Spring 2025, with the voice assistant not likely to be part of the Apple Intelligence rollout this Fall.

Google reportedly explored limiting the ‘AI Overviews’ feature to its own app and excluding Safari users, as part of efforts to reduce reliance on Apple’s browser.

Stability AI announced new changes to its licensing model, making SD-3 free for commercial use up to $1M in annual revenue and addressing quality concerns.

YouTube introduced a new AI-powered eraser tool that allows creators to quickly remove copyrighted music from videos while preserving other audio elements.

OpenAI updated its ChatGPT macOS app to encrypt locally stored conversations after a security flaw was discovered that allowed plain text access to chat logs.

Kuaishou’s KLING video model is now available as a web app for generations up to 10 seconds, though a Chinese phone number is still required for access.

EU chief Margrethe Vesteger warned that Nvidia’s dominance in AI chip supply is a ‘huge bottleneck’ as the company continues to face scrutiny from regulators.

THAT’S A WRAP

SPONSOR US

Get your product in front of over 600k+ AI enthusiasts

Our newsletter is read by thousands of tech professionals, investors, engineers, managers, and business owners around the world. Get in touch today.

FEEDBACK

How would you rate today's newsletter?

Vote below to help us improve the newsletter for you.

Login or Subscribe to participate in polls.

If you have specific feedback or anything interesting you’d like to share, please let us know by replying to this email.

Reply

or to participate.