No Existential Threat: AI Can't Learn on Its Own

What’s Ahead

• New Study Debunks AI Threats: LLMs Can't Learn Independently

• AI Aims to Achieve Einstein and Curie-Level Discoveries

• Silicon Valley is Fighting This AI Regulation Bill in California

New Study Debunks AI Threats: LLMs Can't Learn Independently

Recent research from the University of Bath and the Technical University of Darmstadt reveals that large language models (LLMs) like ChatGPT can't independently acquire new skills.

The Details

The study, featured at ACL 2024, highlights that LLMs, despite their language proficiency and ability to follow instructions, do not possess the capability to develop new skills or complex reasoning on their own.

These models are controllable and predictable, even as they improve in language processing and handling of detailed prompts. The research counters fears that larger models might evolve dangerous reasoning abilities.

Instead, practical risks, such as misuse in creating fake news and fraud, are emphasized. The study advises that future research should concentrate on addressing these real-world issues rather than hypothetical existential threats.

Read more

AI Aims to Achieve Einstein and Curie-Level Discoveries

The new "AI Scientist" promises to revolutionize research by autonomously handling the entire scientific process, potentially speeding up discoveries.

The Details

A breakthrough AI from Sakana AI, developed with Oxford and UBC, promises to transform scientific research. The AI autonomously performs the entire research cycle, from idea generation to publishing.

Capable of designing and conducting experiments and drafting papers, it can complete a full research report for just $15. The AI operates using advanced machine learning techniques. It reviews existing research to identify gaps, generates novel research ideas, and designs experiments based on these ideas.

The model then refines its hypotheses through iterative experimentation. While currently focusing on machine learning, the technology showcases potential for diverse fields by automating complex tasks traditionally done by human researchers, thereby potentially accelerating discovery and innovation.

Read more

Silicon Valley is Fighting This AI Regulation Bill in California

Silicon Valley's strong opposition to California’s SB 1047 highlights fears about its impact on AI innovation and the open-source ecosystem. 

The Details

Major tech firms and startups have strongly resisted California's Senate Bill 1047. Senator Scott Wiener introduced the bill, which requires AI developers to test their models for potential misuse, like enabling infrastructure attacks or developing weapons.

It also proposes the creation of CalCompute, a public cloud for AI research, and robust protections for whistleblowers in AI companies. Meta and Google argue that the bill could hinder technological advancement and open-source contributions.

Supporters, however, see it as a necessary step to ensure responsible AI development and transparency, aiming to balance innovation with safety and ethical considerations. The bill has sparked a heated debate about the future of AI regulation.

Read more

This X (Twitter) post ☝️ is 100% AI-generated. It shows a woman speaking at a conference and has gone viral for its striking realism.

Reactions vary: some viewers are impressed by the lifelike appearance, while others spot giveaways like unnatural gestures, odd finger movements, and occasional sound issues. Created with Runway's Gen-3 Alpha, this video highlights both AI's impressive advancements and its ongoing challenges.