#170 - new Sora rival, OpenAI robotics, understanding GPT4, AGI by 2027?

33 snips

Jun 9, 2024

Get ready for the latest buzz in AI! A new video generator, Cling, is poised to challenge OpenAI's Sora. Apple is debuting its 'Apple Intelligence' focusing on privacy in AI. There's exciting news in music generation with Udio 130, and Perplexity AI is turning searches into shareable pages. Meanwhile, OpenAI revives its robotics group. Big players like Google and Microsoft are strategizing against Nvidia, while Zoox tests self-driving cars in major cities. Plus, discussions about AGI's potential arrival by 2027 spark intriguing debates!

Ask episode

AI Snips

Chapters

Books

Transcript

Episode notes

ANECDOTE

Kling vs. Sora

Kling, from Kuaishou, rivals OpenAI's Sora in AI video generation.
It produces 2-minute HD videos at 30fps, a significant advancement in public demos.

INSIGHT

Kuaishou and State Control

Kuaishou, Kling's creator, is a Chinese company similar to TikTok, receiving state funding.
China's "golden share" in Kuaishou implies significant state control over this AI player.

INSIGHT

Apple's AI Strategy

Apple Intelligence, an upcoming AI system, will choose between on-device and cloud AI.
This system will feature a ChatGPT-like chatbot, prioritizing privacy and security, unlike Microsoft.

Get the Snipd Podcast app to discover more snips from this episode

Get the app

Automatic Data Curation for Self-Supervised Learning: A Clustering-Based Approach

Vasil Khalidov

Piotr Bojanowski

Timothée Darcet

Herve Jegou

Théo Moutakanni

Armand Joulin

Camille Couprie

Patrick Labatut

Hugo Touvron

Marc Szafraniec

Nikita Smetanin

Huy V. Vo

Maxime Oquab

This paper presents a method for automating data curation for self-supervised learning by using successive and hierarchical k-means clustering. The approach aims to create datasets that are large, diverse, and balanced, which can outperform uncurated data and match or surpass manually curated data in performance. The technique has been tested across various data domains, including web-based images, satellite images, and text.

Testing and mitigating elections-related risks

Not specified

The topic involves strategies and processes for identifying and mitigating risks associated with elections, particularly in the digital age. This includes efforts by organizations like Anthropic to test and improve their models for handling election-related queries and ensuring the integrity of electoral processes.

Situational Awareness: The Decade Ahead

Leopold Aschenbrenner

This essay series by Leopold Aschenbrenner delves into the future of artificial intelligence, focusing on the potential development of artificial general intelligence (AGI) by 2027. It discusses the implications of such advancements on national security, global power dynamics, and the need for robust governance to manage superintelligence. The work emphasizes the urgency of preventing hostile actors from acquiring these technologies and advocates for a coordinated international effort to ensure responsible AI development.

Teams of LLM Agents can Exploit Zero-Day Vulnerabilities

Research on AI-Enabled Cybersecurity Exploits

Richard Fang

Akul Gupta

Daniel Kang

Rohan Bindu

This research explores the capability of teams of large language model (LLM) agents to exploit real-world zero-day vulnerabilities. The study introduces a framework called Hierarchical Planning and Task-Specific Agents (HPTSA), which significantly improves the effectiveness of exploiting such vulnerabilities compared to individual LLMs.

GLM-4-9B-Chat-1M

A Large Language Model

THUDM

The GLM-4-9B-Chat-1M model is a powerful AI tool designed for tasks like multi-turn conversations, web browsing, code execution, and long text reasoning. It supports 26 languages and can handle contexts up to 1M tokens, making it suitable for complex content processing.

42: The Wildly Improbable Ideas of Douglas Adams

None

Kevin Jon Davies

Douglas Adams

This book compiles hundreds of Douglas Adams' personal artefacts, including notebooks, letters, and scripts, offering insights into his creative process and technological foresight. It covers his early work, collaborations, and projects that never came to fruition.

Stable Audio Open 1.0

AI Model for Text-to-Audio Generation

Stability AI

Stable Audio Open 1.0 is a latent diffusion model based on a transformer architecture, capable of producing high-quality audio up to 47 seconds long at 44.1kHz. It is particularly effective for generating sound effects and field recordings, and can be fine-tuned with custom audio data.

Zeta

Valentin Menendez

The Pill

Microsoft

Bill Gates