Article:

Are LLM's the best that they will ever be? A guest post on Asimov Addendum by @rufusrock "The technology might continue to get better, but that doesn’t mean that the user experience will. Why? Primarily because no one knows how to make LLMs profitable." Read more below👇

Join @IlanStrauss and @srulyrosenblat TOMORROW at a presentation at @UW on: Methods to identify content in LLM pre-training data, with a focus on membership inference attacks (MIA). Zoom link in the retweet.

Link:

Anthropic: "tested the models using 1,350 pairs of prompts across 9 task types and 150 topics." n (sample size) per topic = 9!? No wonder they mostly pool across topics and tasks to come up with a single score. But then what about the risks lurking in a specific topic?

We support @AnthropicAI important initiative on evaluating for political bias. But the small error band -+.01 given us caution, given that we know how sensitive model outputs are to slight changes ( https://t.co/1PRgLcnRGk). 1/2 🧵

Tech data giants, like Google, Meta, and Microsoft, currently control much of the world’s development and deployment of AI. @IlanStrauss, director of @AIdisclosures, will moderate a panel on Monday, 11/17 on limiting data giants’ dominance of AI. Register:

Amazon's Monetization of it's API: https://t.co/ZMu662pe1H Algorithmic Attention Rents: https://t.co/jMltvGCrmb Digital advertising as rent extraction in law and behavioral economics - a history:

Amazon charging for API usage now - opening up a new branch for rent extraction from it's third-party firms: Read our co-director's previous work on how Amazon uses its algorithm to extract rents from all sides - "algorithmic attention rents": 👇 @superwuster @linamkhan

Full article:

In today's Asimov Addendum: @timoreilly has an excellent piece on AI as a "jetplane for the mind": AI Is a Tool, Not a Worker -- like all software. As AI becomes more autonomous, it must be shaped as that jetplane, not a displacement device, Tim argues Link below📷

Full post by @timoreilly on attribution norms that market-enforced and self-enforcing: https://t.co/TtJ9gA5EE9

If “code is law,” then norms can follow from a well designed market. Norms Tim proposes are: respect do-not-crawl & licenses, attribute community platforms, link for depth, and pay fairly for high-value data. Build an architecture of participation, not extraction. 2/3

AI overviews shouldn’t be “one size fits all", our co-director @timoreilly argues in a new Asimov Addendum article. Canonical facts (e.g., airport parking) are not the same as Wikipedia, GitHub, or paywalled journalism. The responses -- and responsibilities -- must differ. 1/3

Looking for a 10 min weekend read? "If Humans Can See It, Bots Can Steal It" @timoreilly @isobelmoure and @IlanStrauss discuss new architectures for content in the age of AI. Let us know what you think💬 Full piece: https://t.co/sB1bx13OWT

- The need for "code" -- as the law of the internet -- to establish norms online by incentivizing good behavior. Read the full piece here:

"If Humans Can See It, Bots Can Steal It" In our latest Asimov Addendum post @timoreilly, @isobelmoure, & @IlanStrauss explore: - AI Bots running wild - The evolving internet under http - How business models from new tech arise in response to information being freed 1/2

Post: https://t.co/DReMyf3UfS Working Paper:

AI is not the messiah, its a very naughty market in need of disclosure. Read our most recent Asimov Addendum // @techpolicypress piece for more by Co-Directors @IlanStrauss and @timoreilly We also have a new working paper on SEC Disclosures for AI. Links below 👇

Look forward to discussing @doctorow’s terrific new book at the Brooklyn Public Library on Thursday evening.