Werd I/O

Utterly ghoulish. #AI

[Link]

I strongly disagree with this stance. Allowing your work to be mined by AI models should be opt-in only - otherwise there is no possible way for a publisher or author to apply a license or grant rights. #AI

[Link]

Different AI models have different political biases. Google's tend to be more socially conservative - possibly in part because they were trained on books rather than the wider internet. Regardless of the cause, this is proof, again, that AI models are not objective. #AI

[Link]

Black faces are overrepresented in databases used to train AI for law enforcement - and some facial recognition software used in this context fails 96% of the time. This practice is an accelerant for already deeply harmful inequities. Time to ban it. #AI

[Link]

This is a really comprehensive history and overview of LLMs. Simon has been bringing the goods, and this talk is no exception. #AI

[Link]

I don’t know that it’s fair to count AI startups as media startups. Given the (justified) labor disputes going on right now, I’d offer that they’re closer to anti-media, and I’m not sure that I’d think of them as a bright spot. There’s plenty of room for AI to assist creatives, but of course the real money is in replacing them or devaluing their work. #AI

[Link]

This caught my eye: an example of OpenAI licensing content from a publisher in order to make its models better. Other publishers should now know that they can make similar deals rather than letting their work be scraped up for free. #AI

[Link]

I respect Bruce Schneier a great deal, but I hate this proposal. For one thing, what about people outside the US whose data was used? On the internet, the public is global. Wherever the tools are used, the rights infringed by AI tools are everyone's, from everywhere. Paying at the point of use rather than at the point of scraping cannot be the way. #AI

[Link]

It's important that lawsuits like this center on the use, not the act of scraping itself - the latter does need to be protected. One to watch. #AI

[Link]

I think this is a legal challenge waiting to happen. While people who publish publicly online have a reasonable expectation that anyone can read their content, they don't have a similar expectation about content being modeled and analyzed. There's no de facto license to do this. #AI

[Link]

““Language skill indicates intelligence,” and its logical inverse, “lack of language skill indicates non-intelligence,” is a common heuristic with a long history. It is also a terrible one, inaccurate in a way that ruinously injures disabled people. Now, with recent advances in computing technology, we’re watching this heuristic fail in ways that will harm almost everyone.” #AI

[Link]

“To assess the effectiveness of current A.I.-detection technology, The New York Times tested five new services using more than 100 synthetic images and real photos. The results show that the services are advancing rapidly, but at times fall short.” #AI

[Link]

“AI-generated misinformation is insidious because it’s often invisible. It’s fluent but not grounded in real-world experience, and so it takes time and expertise to unpick. If machine-generated content supplants human authorship, it would be hard — impossible, even — to fully map the damage.” #AI

[Link]

“Auto-generating text based on other people’s discoveries and then automatically summarising that text by finding commonalities with existing text creates a loop of mechanised nonsense. It’s a prayer wheel for capitalism.” #AI

[Link]

“Human reviewers may read the chats, and researchers found that similar AI could reproduce the data it absorbed during training, creating a leak risk. Alphabet also alerted its engineers to avoid direct use of computer code that chatbots can generate, some of the people said.” #AI

[Link]

“When tested, "Over 90% of 1,008 generated jokes were the same 25 jokes.”” We have a lot in common. #AI

[Link]

“The new policy, establishing that AI-generated content is de facto allowed on the network, is harmful in both what it allows on the platform and in how it was implemented.” #AI

[Link]

“More and more prominent tech figures are voicing concerns about superintelligent AI and risks to the future of humanity. But as leading AI ethicist Timnit Gebru and researcher Émile P Torres point out, these ideologies have deeply racist foundations. #AI

[Link]

“Just yesterday, I asked an AI program to write an entire sci-fi novel for me, and [as someone who will die an empty shell of a man who wasted his life doing nothing for the world and, perhaps, should never have been born] I was super impressed.” #AI

[Link]

““Every single thing Tessa suggested were things that led to the development of my eating disorder,” Maxwell wrote in her Instagram post. “This robot causes harm.”” #AI

[Link]

“A free resource that will help you develop an AI-bullshit detector.” #AI

[Link]

“If Google's AI is going to mulch up original work and provide a distilled version of it to users at scale, without ever connecting them to the original work, how will publishers continue to monetize their work?” #AI

[Link]

“ChatGPT (via Plugins) can access YouTube transcripts. Which is pretty neat. However, as expected (and predicted by many researches) all these quickly built tools and integrations introduce Indirect Prompt Injection vulnerabilities.” Neat demo! #AI

[Link]

“Rather than pointing to some future utopia (or robots vs. humans dystopia), what we face in dealing with programs like ChatGPT is the further relentless corrosiveness of late-stage capitalism, in which authorship is of no value. All that matters is content.” #AI

[Link]

“Moreover, researchers have also discovered that it’s probably mathematically impossible to secure the training data for a large language model like GPT-4 or PaLM 2. This was outlined in a research paper that Google themselves tried to censor, an act that eventually led the Google-employed author, El Mahdi El Mhamdi, to leave the company. The paper has now been updated to say what the authors wanted it to say all along, and it’s a doozy.” #AI

[Link]