4 Deep AI Items Today (8.4.23)

Emergence debate, election influence, UN

Happy Friday.

Four top stories in AI today, according to our weighted analysis of AI community engagement. These ones are deep. Any of them could have global consequences in the relatively near-term future.

Cheers,

Marshall, Editing @aiTTENTION

Language Learning Models Shaped, Not Discovered, Argues Suleyman

DeepMind co-founder, and now CEO of Inflection AI, Mustafa Suleyman posted a tweet yesterday arguing that LLMs are guided and shaped by design choices. "Capabilities don't arise on their own. There are no emergent properties here. This implies a level of autonomy or agency that simply isn't there." A lot of debate ensued. Meta's Yann LeCunn agreed with Suleyman. Many people disagreed. For example, cognitive scientist Joseph Lawrence disagreed, arguing that intelligent emergent behavior doesn't require agency, it's seen in ant colonies and cells in an immune system. [via @mustafasuleymn]

Altman Foresees AI's Potential Election Influence

OpenAI leader Sam Altman warns about AI's potential impact on elections, emphasizing personalized persuasion and high-quality media. Vinod Khosla said something similar on Kara Swisher's podcast last week: "I'd be shocked if China doesn't have an effort to have bots talk to every US voter in the next election to the best of its ability, one-on-one conversations manipulating their views. They'll definitely do it in Taiwan." [via @sama]

Fei-Fei Li Joins UN Secretary-General’s Scientific Advisory Board

Stanford Institute for Human-Centered AI Co-Director and Sequoia Professor in the Computer Science Department Fei-Fei Li has joined a new scientific advisory board created by United Nations Secretary-General Antonio Guterres that will advise UN leaders on the breakthroughs in science and technology and how to harness the benefits of these advances while mitigating potential risks. [Stanford.edu]

Unfolded — Meet the scientists using AlphaFold

Launched with great fanfare in 2021, AlphaFold is an AI system from DeepMind that the firm says has predicted 200M+ protein structures and has been used by over 1 million researchers globally to advance biological studies. The protien predictions are free for academic and commercial use under Creative Commons license. Elon Musk asked at 2:36 AM this morning "When will the first product that uses AlphaFold ship?" [unfolded.deepmind.com]

that's it!