🥟 Chao-Down #306 Brands are concerned about how they are perceived by ChatGPT, Microsoft reports China is targeting US voters with AI disinformation, A look at tools to detect audio deepfakes

What are the risks for asking AI to help you code?

Researchers found that AI bots often invent fake software packages when asked for coding help. The finding represents a major security risk as bad actors could create malware with the same invented package name, then trick developers relying on the AI's generated code into downloading and installing it.

Security researcher Bar Lanyado from Lasso Security revealed that generative AI models, including GPT-3.5 and GPT-4, have inadvertently recommended non-existent software packages when prompted to help with coding tasks. These hallucinations have led to thousands of downloads of fictitious software with major companies like Alibaba duped into using these fake packages in their code.

The research highlights a concerning trend where AI’s influence can extend beyond mere suggestions, but actually represent widespread cybersecurity threats if not addressed promptly.

-Alex, your resident Chaos Coordinator.

What happened in AI? 📰

How Tech Giants Cut Corners to Harvest Data for A.I. (The New York Times)

Meta plans to more broadly label AI-generated content (Engadget)

Brands are concerned about how they're perceived by ChatGPT (Fast Company)

China Is Targeting U.S. Voters and Taiwan With AI-Powered Disinformation (WSJ)

Tools to detect audio deepfakes are in a race with technology (NPR)

AI: Inside the shadowy global battle to tame the world's most dangerous technology (POLITICO)

Always be Learnin’ 📕 📖

Why is observability so expensive? (mattklein123.dev)

The state of global AI safety research – Emerging Technology Observatory (eto.tech)

How Stack Overflow replaced Experts Exchange (graphite.dev)

Projects to Keep an Eye On 🛠

Universal-1 - Robust and accurate multilingual speech-to-text (Assembly AI)

zylon-ai/private-gpt: Interact with your documents using the power of GPT, 100% privately, no data leaks (Github)

nashsu/FreeAskInternet: FreeAskInternet is a completely free, private and locally running search aggregator & answer generate using LLM, without GPU needed. (Github)

The Latest in AI Research 💡

Great, Now Write an Article About That: The Crescendo Multi-Turn LLM Jailbreak Attack (crescendo-the-multiturn-jailbreak.github.io)

Mixture-of-Depths: Dynamically allocating compute in transformer-based language models (arxiv)

Language Models as Compilers: Simulating Pseudocode Execution Improves Algorithmic Reasoning in Language Models (arxiv)

The World Outside of AI 🌎

Shhh, I’m on vacation. Travelers are on a quest for the quiet (Bloomberg)

Exclusive: NYT to soon offer most articles via automated voice (Axios)

Brands Paid for Ads on Forbes.com. Some Ran on a Copycat Site Instead. (WSJ)

Scientists Use NASA Data to Predict Solar Corona Before Eclipse (NASA)

Scientists are deep-freezing corals to repopulate the ocean (Free Think)

Why loneliness is bad for your health (Nature)

One Last Bite 😋

2024 Eclipse: Airbnb bookings perfectly paint the path of totality (Axios)