The Daily Index — Monday, June 12, 2023 · TLDR Index

Across today’s newsletters

Today’s Index

TLDR AI

META LEADS OPEN-SOURCE AI WITH NEW TEXT-TO-MUSIC MODEL

CEO Mark Zuckerburg recently discussed his goals with open-sourcing significant expensive AI tools in a podcast. This collection of music generation code, model weights, and evaluations is an example of just that. It allows you to generate both melody conditioned and text conditioned music. The code is fully open sourced, but model weights are not commercially open.

ai.honu.io · 4 min

TLDR Web Dev

CS:GO: FROM ZERO TO 0-DAY

Neodyme identified three independent remote code execution (RCE) vulnerabilities in the popular Counter-Strike: Global Offensive game. Each vulnerability can be triggered when the game client connects to their malicious Python CS:GO server. This post details their journey through the CS:GO binary and conducts a technical deep dive into various identified bugs. They conclude by presenting a proof of concept (POC) exploit that leverages four different logic bugs into remote code execution in the game’s client, triggered when a client connects to the server.

neodyme.io · 16 min

TLDR Crypto

CRACKDOWN

As a result of the SEC's latest offensive against crypto exchanges, Robinhood is delisting Solana, Polygon, and Cardano after June 27. Users will be able to transfer these cryptocurrencies until the deadline, but post-deadline, Robinhood will sell the coins, crediting the corresponding USD value to users' accounts. The decision follows Robinhood's review of SEC lawsuits against Binance and Coinbase. Despite having its securities trading license, Robinhood officials acknowledged that they are not sure if this would allow them to trade the tokens the SEC is claiming are securities.

theblock.co · 1 min

TLDR Founders

WEB3 IN SAAS CLOTHING

Tomasz Tunguz argues due to a severe decline in funding and modest addressable market, web3 startups will be forced to become web2 SaaS companies imbued with the unique traits that web3 technology enables.

tomtunguz.com · 1 min

TLDR AI

STABLE DIFFUSION TAKES STEREOTYPES TO THE EXTREME

Bloomberg journalists found that Stable Diffusion takes racial and gender disparities to extremes that are even worse than those found in the real world, stroking fears that generative AI will perpetuate biases.

archive.ph · 15 min

TLDR AI

KHAN ACADEMY KHANMIGO

Khan Academy has introduced an AI-powered chatbot, Khanmigo, which aims to provide personalized tutoring to students. Powered by GPT-4, the bot offers guidance on a broad range of subjects and uses the Socratic method to encourage learners to solve their problems independently. While proponents believe such bots could revolutionize education, concerns about the technology's potential for error and impact on teachers' roles persist and its effectiveness is currently being studied ahead of a broader rollout in U.S. schools.

archive.ph · 6 min

TLDR AI

NEW MULTIMODAL TEXT/VISION MODEL FROM SINGAPORE

A lot of new acronyms are introduced in this work. The gist is that researchers collected a new dataset for visual instruction tuning, trained a model, and made some algorithmic tweaks along the way. The demo is really impressive, especially after the AR reveal from Apple. An exciting peak at what AI assistants might look like in the future.

otter-ntu.github.io · 7 min

TLDR AI

MAKING LANGUAGE MODELS MORE TRUTHFUL AT TEST TIME

Language models are well known for confidently confabulating (sometimes called hallucinating). It turns out that these models have an internal representation for truthfulness and by tweaking the model activations you can elicit a truthful response. The researchers found that by using this technique on the instruction-tuned Alpaca model, they improved TruthfulQA performance from 32.5% to 65.1%.

arxiv.org · 28 min

TLDR AI

SPQR: A TECHNIQUE FOR HIGH-EFFICIENCY LLM COMPRESSION

The paper presents Sparse-Quantized Representation (SpQR), a new format and technique that allows almost lossless compression of large language models (LLMs), overcoming the usual accuracy losses from quantization. SpQR enables powerful LLMs to run on common devices like laptops and mobile phones without performance degradation, offering over 4x memory compression gains and faster inference than traditional methods.

arxiv.org · 19 min