Weekend Links: Foundational models, incumbents head-start, and WWII desinformation
Open vs. Closed Models
What are the risks of open foundation models? The answer depends on how you frame the question. Straight off the bat, it's not precise enough:
Today, we are releasing a paper on the societal impact of open foundation models. We make three main contributions. First, we diagnose that the disagreement on the impact of openness results from a lack of precision in claims about its societal impact. Second, we analyze the benefits of open foundation models such as transparency, distribution of power, and enabling scientific research (including when open model weights are not enough to realize some benefits). Third, we offer a risk assessment framework for assessing the marginal risk of open foundation models compared to closed models or existing technology like web search on the internet.
This quote is from a new post on the always worth reading AI Snake Oil. If you're mulling over AI safety, it's definitely worth a click to read the full text, summarizing a paper the authors at AI Snake Oil also contributed to.
Bruce Schneier ponders adjacent issues in a post on his blog. Also worth a read.
Is There Still Room for New Challengers?
And tied to foundation models, I leave Ben Thompson's piece Mistral, Microsoft's Investment, Generative AI and Customer Support (behind a paywall) with a question: Is there, after all, a chance for new challengers who don't already sit on massive data sets?
Mistral AI is challenging the conventional wisdom that the winners of the AI race will emerge from among the tech industry’s U.S. giants.
Disinformation During World War II
With 2024 being a "super election year," discussions about disinformation online, often linked to generative AI models, are plentiful. The Guardian has [a lengthy piece on disinformation during World War II](https://www.theguardian.com/books/2024/mar/02/the-man-who-tricked-nazi-germany-lessons- from-the-past-on-how-to-beat-disinformation). Though the methods of producing and spreading such content have changed, reading it feels like there are important underlying elements, like psychology, that remain as they were then.