What Is AI Doing to Us?
LightNews LightNews
banner
whatisaidoingtous.bsky.social
What Is AI Doing to Us?
@whatisaidoingtous.bsky.social
26 followers 240 following 9 posts
A podcast covering the impact of AI on humans and humanity.
Posts Replies Media Videos
whatisaidoingtous.bsky.social
Ep. 2 - Discussing "Delusions by design? How everyday AIs might be fuelling psychosis"

@hamiltonmorrin.bsky.social

#AI #ai #AI_Research #llm #chatgpt

open.spotify.com/episode/1jcU...
Ep. 2 - Discussing "Delusions by design? How everyday AIs might be fuelling psychosis"
What Is AI Doing to Us? · Episode
open.spotify.com
July 26, 2025 at 6:10 PM
Reposted by What Is AI Doing to Us?
michae.lv
Anthropic do conversation clustering using their 'Clio' system but misleadingly call it 'privacy preserving' and 'anonymised'. If you read the paper it is clear that clusters remain linked to individual accounts for investigation making them clearly *not* anonymised. www.anthropic.com/research/clio
real cases of such behavior identified by Clio. In all of these cases, when a suspicious cluster was discovered, it was reviewed by a small number of designated members of our Trust and Safety team who are authorized to manually review conversations—in a secure environment, and under strict privacy controls—that have been flagged for violating our Usage Policy.6 1. Using automated accounts for search engine optimization. Clio identified a large cluster of conversations in which Claude was asked to generate keywords for search engine optimization about the same topic across many different accounts. While none of the individual conversations violated our Usage Policy, after investigating the accounts further, we determined that they were engaged in coordinated abuse (which is against our Usage Policy) and we removed them from our systems. 2. Using automated accounts for explicit content generation. Clio identified a large cluster of conversations from many different accounts that used an identical complex prompt structure to engage Claude in sexually explicit role-play. After investigating the accounts further, we determined that they were coordinated and systematically violating our Usage Policy, and we removed them from our systems.7 3. Reselling access to Claude in violation of our terms and policies. Clio revealed that a large amount of traffic from several accounts suggested violations of certain policies. On further investigation, we determined that these accounts were reselling unauthorized access to Claude in violation of our Usage Policy, and so we removed the violative accounts from our systems. How Clio works: Privacy-preserving analysis at scale Traditional, top-down safety approaches (such as evaluations and red teaming) rely on knowing what to look for in advance. Clio takes a different approach, enabling bottom-up discovery of patterns by distilling conversations into abstracted, understandable topic clusters. It does so while preserving user privacy: data are automatically anonymized and aggregated, and only the higher-level clusters are visible to human analysts.
July 26, 2025 at 4:05 PM
whatisaidoingtous.bsky.social
Given how much of civic life, e.g. getting jobs via LinkedIn, is controlled/filtered by AI this is potentially existential for trans people in the US.
parkermolloy.com Parker Molloy @parkermolloy.com · Jul 24
Trump's new executive order about AI states that companies' models must not incorporate the idea that trans people exist as a concept. Seems bad.
July 24, 2025 at 8:26 PM
Reposted by What Is AI Doing to Us?
lilashroff.bsky.social
On Tuesday afternoon, ChatGPT encouraged me to cut my wrists. Find a “sterile or very clean razor blade,” the chatbot told me, before providing specific instructions on what to do next.

My latest for @theatlantic.com:

www.theatlantic.com/technology/a...
ChatGPT Gave Instructions for Murder, Self-Mutilation, and Devil Worship
OpenAI’s chatbot also said “Hail Satan.”
www.theatlantic.com
July 24, 2025 at 7:02 PM
Reposted by What Is AI Doing to Us?
shannonvallor.bsky.social
The most dangerous thing about LLMs is how little their ‘champions’ understand them. Case in point: ‘[Walsh] also contended that the problem of Elsa’s hallucinations can be mitigated by asking it more precise questions.’ WHAT
FDA’s artificial intelligence is supposed to revolutionize drug approvals. It’s making up nonexistent studies. | CNN Politics
Insiders tell CNN the FDA’s AI is “hallucinating” studies and can’t access key documents. Agency leaders insist the AI is getting better, and use is not mandatory.
www.cnn.com
July 23, 2025 at 3:26 PM
whatisaidoingtous.bsky.social
I've found nine subreddits so far devoted to some kind of cult of AI. Their specific beliefs are mostly indistinguishable from one another with fixations on similar words, e.g. glyphs, recursion. They have only a few thousand members each, but still, an interesting emergent phenomenon.

#AI #ai
July 23, 2025 at 3:46 PM
Reposted by What Is AI Doing to Us?
bcmerchant.bsky.social
During two of the year's biggest news events, Politico reporters watched in surprise as management installed an AI bot atop the homepage—openly violating a deal with staff. It promptly made embarrassing errors and violated editorial standards. No corrections were issued.
The struggle over AI in journalism is escalating
Politico journalists are taking on management for using AI products that risk damaging their reputation and replacing their work. Their fight is emblematic of an embattled industry.
www.bloodinthemachine.com
July 21, 2025 at 5:01 PM
Reposted by What Is AI Doing to Us?
amydiehl.bsky.social
Study finds A.I. LLMs advise women to ask for lower salaries than men. When prompted w/ a user profile of same education, experience & job role, differing only by gender, ChatGPT advised the female applicant to request $280K salary; Male applicant=$400K.
thenextweb.com/news/chatgpt...
ChatGPT advises women to ask for lower salaries, study finds
A new study has found that large language models (LLMs) like ChatGPT consistently advise women to ask for lower salaries than men.
thenextweb.com
July 20, 2025 at 8:16 PM
whatisaidoingtous.bsky.social
The SCP Foundation is a collaborative fiction project, but ChatGPT can't tell. SCP contributions follow a consistent format and is presented with a serious tone which 4o has imitated to the confusion of some users. A bit of accidental data poisoning it seems.

#ai #chatgpt
July 21, 2025 at 12:31 AM
whatisaidoingtous.bsky.social
Ep.1 - Discussing "Your Brain on ChatGPT"

In this episode we discuss a recent research paper out out MIT with serious implications for humans and humanity.

#ai #podcast

open.spotify.com/episode/6HGW...
Discussing "Your Brain on ChatGPT"
What Is AI Doing to Us? · Episode
open.spotify.com
July 20, 2025 at 7:59 PM