My new indicator of AI-generated academic text is slightly random three-word groupings at the end of paragraphs.
Ex: "XYZ is important for legibility, visibility, and control."
Remote
Emmie Hine
@emmiehine@dair-community.social
mastodon 4.7.3Researching the ethics & governance of emerging tech, especially AI and XR.
PhD student in Law, Science, and Technology @Unibo & Master of the (social science of the) Internet @Oxford
ex-SWE
405 Followers
126 Following
21 Posts
Joined November 06, 2022
pronouns:
she/hers
newsletter:
personal site:
Replying to
The comparison was right there. So I had Claude re-write the US Constitution in the style of its own. The result? Kind of like the Declaration of Independence got an MFA.
Replying to
But it's a valuable thought experiment.
Madison assumed people are not angels and created design structures that still work when virtue fails.
Anthropic hopes for virtue (which they can, in a way, instill) and so leads with values, then applies judgment.
Replying to
@seldo @mmitchell_ai I can confidently say "no" and also that it is mostly my own fault 😅
Replying to
(Before we start, yes, I was snowed in.)
Context: Anthropic just dropped a ~30,000 word internal “constitution” for Claude with the rules & values shaping what it will and won't do.
Replying to
🧠 Grok scandal continues: Backlash over Grok’s ability to undress women in generated images hasn’t died down. xAI has *slightly* tightened limits, but with major loopholes. 2/
Replying to
Check it out (and get a sneak preview of the newsletter revamp!) here: https://ethicalreckoner.substack.com/p/wr-73-new-year-new-newsletter-plus
Replying to
It’s funny, but it’s also a real governance question:
Do we constrain power with hard rules, try to instill values we can trust, or both? And in what order? 🔗👇
Replying to
New longread + the full “Claudified Constitution” PDF is in today’s Tech Reckoner:
https://techreckoner.substack.com/p/er-41-the-us-constitution-claudified
Replying to
🚫 The safeguards xAI announced mostly apply to paid accounts or are vague about jurisdictions, meaning they may not offer actual protection. 3/
Replying to
🌫️ Environmental loss: xAI lost its EPA appeal to keep methane turbines powering its Memphis data center. A huge win for local communities concerned about pollution around Colossus data center operations. 4/
Replying to
📢 OpenAI goes ad-supported: Meanwhile, OpenAI is testing ads in its free and “ChatGPT Go” tiers. Overall they seem to be going about it in a thoughtful way, but it is a big shift in their business model. 5/
Replying to
🧠 AI economics stress: OpenAI’s move toward ads underlines industry pressure — AI companies still aren't profitable, and ads may become the new norm for subsidized access. 6/
Replying to
🤖 Big picture takeaway: A turbulent week for AI players.
📌 Extra reckoning: Meta also announced big layoffs in XR/Reality Labs--how does this relate to the AI bubble? Read more 🔗👇 7/
Issue 3 of the China AI Bulletin is out! This edition: China finalizes its Human-Like AI Interim Measures, DeepSeek and Moonshot launch new flagship models, and the NDRC blocks Meta's $2B Manus acquisition.
Read it here: https://chinaaibulletin.substack.com/p/china-ai-bulletin-3
There is no faster route to self-knowledge than measuring the pixel distance between your pupils for a Chinese visa photo.
Unfortunately, that knowledge is just a detailed inventory of all your minor facial asymmetries.
Slightly longer hiatus than planned 🫣 but the ER is back with a new edition on the Digital Vulnerabilities in the Age of AI (DIVAS) summit. Curious about how AI makes us vulnerable in new contexts? Check it out here: https://ethicalreckoner.substack.com/p/er-40-divas-dispatches
