@jedbrown.org
Prof developing fast algorithms, reliable software, and healthy communities for computational science. Opinions my own. https://hachyderm.io/@jedbrown https://PhyPID.org | aspiring killjoy | against epistemicide | he/they
Inevitability frame is insidious, yet we had 150 participants at our teach-in last week and our faculty assembly has been unanimous. Many assume that criticism is shallow, but upon learning what these products are and the power structures they support, turn strongly negative about their imposition.
After revelations that companies like Ring have been selling home security footage to ICE, the most common defense of keeping a camera up seems to be, βBut what if Iβm stalked?β
And as a violence researcher, Iβm just here to state for the record that police do not care about your security footage.
The truth is that some engineers found what they perceive to be a short-cut to KPIs and executives have investors to impress, but the industry has no idea what is required to maintain QA standards and manage tacit knowledge/tech debt/agility through the lifecycle of a software product.
Hereβs a better way of putting it: if youβd like to know why faculty are so upset about CU unilaterally signing a $2 million contract with Open AI here are a few reasons π
The Inspector General for NSF must launch an investigation into the Trump administrationβs alleged attempt to sell-off parts of NCAR β immediately!
ππ½ππ½ππ½
This puts me at 10 people released. I filed my 1st habeas petition 29 days ago. I am an incredibly small-time, nobody lawyer who knew nothing about immigration law the day I submitted that case. I barely know any more today. I had never sued the federal government and now I've beaten them 10 times.
Yes, that whole incident should be understood as clever PR by Amodei. @heidykhlaaf.bsky.social has deep expertise on this subject and a finger on the pulse.
Absolutely, and it is so telling that even the use cases touted here as good fall apart when examining construct validity.
> If you accept some A.I. tools, like to summarize a thousand pages of patient records
An LLM "summary" is a purely linguistic artifact, not an epistemic product of the source
It was great to join @aljazeera.com's podcast "The Take" to discuss the details of the DoW's use of Claude in Iran, as well as the stand-off between DoW and Anthropic that was largely safety theatre.
www.youtube.com/watch?v=skyI...
Even the military is not entirely composed of pure loyalists. Having people do the target analysis creates ample chance for evidence, objection, and people who can testify. The fetish for chatbots generating targets is the impunity that comes from scuttling the paths for liability and inhibitions.
Writing (prose and code) *is* thinking and learning. I think you'll enjoy this short video of Ward Cunningham on coining the term "technical debt" and how it wasn't meant to be about sloppy code, but misalignment of the code with the (evolving, through learning) mental model of the problem domain.
It hardly matters if a person gives final approval. They have been provided with a convincing counterfeit of an intelligence report. Automation bias and institutional culture discourages rigorous independent checking. It's an inevitable result of systemic choices. Choices that are being celebrated.
Media will keep drooling over the "surgical precision" of "AI" targeting. Nobody involved is behaving as though they will face accountability. The chatbot creates permission; it grants plausible deniability. Discourse often gets hung up on human intent, but this is the impact.
It's easily lost, but the US military does not need a chatbot to *find* or execute a double-tap bombing of a girls elementary school. The chatbot provides something shaped like a target analysis without the diligence, cognitive processes, and accountability of a human conducting a target analysis.
Nine, mostly white dudes on stage in a manel.
Everyone wants to sign letters The Future of Life Institute puts out every few years, it seems.
Take a look at this manel which happened around their first letter π
The billionaires and eugenicists on this manel are the actual existential risks to humanity we should worry about.
Yes, banned from Bluesky. Blacksky recently moved to its own appview so Bluesky-banned users like Εink are available. This is obviously still limiting, but there could be a future in which Bluesky is not a singular power in the network. (I moved in the fall; can recommend.)
Another day, another chatbot convincing someone to kill themselves. But yeah let's keep talking about how to put these in our classrooms!
The data from your Meta Ray Bans is used to train Meta's AI, which most people don't understand means that humans are looking at the most intimate details of their lives. www.svd.se/a/K8nrV4/met...
Only "shocking" in how these products are marketed and widespread failure to reflect on well-known cognitive biases going back to the ELIZA Effect (1966). These problems are inherent to the tech because LLMs are purely linguistic devices. Convincing (at times) counterfeiters, always epistemic void.
The fraudulent citations to real sources is a Grammarly feature (not even ironically). See it (and more counterfeiting) in action in this video:
www.carleton.edu/ai/blog/gram...
CFP, AI Resistance, Refusal, Reclamation & Reimagining: Ethical Imperatives and Emerging Practices! Note "This collection is focused ...on the strategies & actions of individuals, communities, organisations & collectives to actively resist, refuse, reimagine & reclaim 'artificial intelligence."
Also note that many developers are subject to coercive measures to maximize "AI" use and self-report success (performance eval), quality be damned. And contrast the rosy claims of the Potemkin performance for investors to what developers say when able to speak freely.
Excellent analysis of that study.
Evidence for the first case is surprisingly flimsy, even contradicted by Anthropic's own study (as well as the 2025 METR report showing strong cognitive bias in self-assessment). And people switch to maladaptive use modes thinking it's helping even when it causes a 30% drop in understanding.
Good thread on the psychological violence and emptiness of wrangling synthetic text.
Still, note that "hallucination" is a misnomer; the epistemic nihilism is always. Meaning is strictly in the eye of the beholder and correctness or falsehood can only ever be incidental, linguistic serendipity.
They are making money inside companies by automating reconciliation workflows. By qualifying inbound leads. By generating compliance documentation. By reducing customer support overhead. By stitching together painful operational tasks that humans hate doing.
These people tell on themselves like clockwork. It's great at creating plausible deniability for financial fraud, regulatory compliance fraud, and deceptive responses to customers (unless a court says we have to honor its promises). Just cuz we believe these jobs don't deserve to be done correctly.
Try now? Εink has been working on staging.blacksky.community since mid-Jan, and only today on main (not the first time I tried on the new appview, but it started working sometime within the past few hours for me).
Fun that it's been up over 14 hours with an identical quote printed twice. The quote itself so generic it could be slop. Totally normal and intentional editorial standards.
AI is a a lack of consent machine that is designed to move us all toward slavery
