Marc Slemko

This morning we sued Donald Trump and Pam Bondi for the approval of the illegal sale of TikTok's U.S. assets to administration allies like Oracle. This deal allows both China-allied ByteDance and Trump-allied Oracle to censor content they don't like.

Makes complete sense to me, humans are predisposed for social connection, so the more that support meets this the more naturally and fully it can fit into someone's life. Not all suicidal ideation has anything to do with lack of connection, but a lot of it sure relates.

If you care about the rights of young people to access information, express themselves, associate with others, and to play. &/or if you care about the privacy and anonymity rights of both adults and minors. & if you don't want surveillance embedded in all online services, speak out against these.

At one point he said he overheard a security guard talking about bets made among the staff over which detainee would be next to die by suicide. The guard said he had paid $500 into a pool, with the total pot riding on the outcome. The talk was particularly jarring, he said, because he had contemplated suicide himself.

Indifference to suffering is a hallmark of genocide.

Had some more thoughts on the latest between Anthropic and the Defense Department. The story is less about the specifics, and more about the way Anthropic now feels it needs to act to grovel before the Trump admin.

Automated snapshots are automatically deleted by default when you delete a RDS instance. Long ago they added a feature to create a final snapshot unless you opt out, but training data is full of people mad about having a snapshot created when trying to delete a DB. The UI has more guardrails.

I'm working on a book proposal at the moment, the core thesis of which is "Democracy fails not when lies spread, but when verified reality can no longer alter incentive structures." and we're seeing it happen in real time. Which is good for the book, but bad for democracy.

Seems much better than the vulnerabilities still existing and being hidden away for years and sold to the highest nation state or other bidder for their own targeted use.

I think experiences here vary widely depending on industry, company and if the juniors are on an upwards trajectory, regardless of tooling, or are proto-perma-juniors who are eventually given Sr titles because they have been around a while and can do *something* well enough. Management and hiring.

Polymarket screenshots aren't worth it.

"In the Doubao phone, the AI agent comes fused into the operating system (OS). It is an elevated OS systems-level permission called INJECT_EVENTS that reads and interprets the screen and clicks buttons in ways that are indistinguishable from a human user." Cute new data arms race, apps versus OS.

Is believing people can own code part of that umbilical? It seems to me the FSF was largely created as a reactionary measure to help ensure the umbilical flows in both directions. Or do you mean the latter more corporate friendly definitions of Open Source that were reactionary to the FSF?

Quote from the abstract of https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6298838 "Some Simple Economics of AGI" : We model the transition toward AGI as the collision of two racing cost curves: an exponentially decaying Cost to Automate, driven by compute and accumulated knowledge, and a biologically bottlenecked Cost to Verify, bounded by human time and embodied experience. This structural asymmetry widens a Measurability Gap between what agents can execute and what humans can afford to verify. It also drives a shift from skill-biased to measurability-biased technical change and a radical bifurcation of economic value. Rents migrate to what remains scarce: verification-grade ground truth, cryptographic provenance, and liability underwriting—the ability to insure outcomes rather than merely generate them. Economic progress has always rested on an implicit compact: that the value claimed was the value produced. The Measurability Gap is the first force in history capable of systematically breaking that compact—not through crisis, but through the ordinary economics of cost minimization. When an AI agent generates output that looks correct, passes every test, yet silently violates unmeasured human intent, the economy accumulates systemic risk.

One of the more interesting non industry specific perspectives on this I've seen recently is in the paper "Some Simple Economics of AGI" (papers.ssrn.com/sol3/papers....). It sets up concepts like Cost to Verify and The Measurability Gap (∆m) and tries to explain and predict based on that model.

Uncle CSAM wants YOU to sell your soul to serve pedophiles. #seattle #streetart #epstein #epsteinfury #ice

They are worried about moral injury to Claude. Everyone is operating near the edge of reality.

...spinning, darting...

　　　　◢
　　　◢▇　　　　　◢
　　◢◤▇　　　　◢▇　　　◢
　◢◤　▇　　　◢◤▇　　◢▇　◢▏◢▏◢
◢◤　　▇　　◢◤　▇　◢◤▇◢▇◢▇◢▇
◤　　　▇　◢◤　　▇◢◤　▇◤▇◤▇◤▇◤
　　　　▇◢◤　　　▇◤　▕◤▕◤▕◤▕◤
　　　　▇◤　　　▕◤
　　　▕◤

Agreed, I think this picture says a lot about the operation.

"If OpenAI retains full discretion over its safety stack and deploys only on its own cloud infrastructure, the practical constraints on government use are architectural, not merely contractual." Do we have validation of that "if"? It wasn't clear to me who owns the cloud accounts for their deploy.

He is so deeply unconvincing. "However, maybe this is the question you are really asking: what would we do if there were a constitutional amendment that made it legal?

Maybe I would quit my job."

"Even as Mr. Trump published the post at 3:47 p.m., the two sides kept talking. Mr. Michael, who was on a call with Anthropic executives at the time, said the Pentagon wanted the company to allow for the collection and analysis of unclassified, commercial bulk data on Americans, such as geolocation and web browsing data, people briefed on the negotiations said. Anthropic told the Pentagon that it was willing to let its technology be used by the National Security Agency for classified material collected under the Foreign Intelligence Surveillance Act. But the company wanted a legally binding promise from the Pentagon not to use its technology on unclassified commercial data."

Unclassified commercial bulk data collection has been crying out for attention and legislation for years, as the data included just keeps expanding.

OpenAI posted the terms of the deal. Reveals that it absolutely does allow for domestic surveillance. EO 12333 is how the NSA hides its domestic surveillance by capturing communications by tapping into lines *outside the US* even if it contains info from/on US persons.

openai.com/index/our-ag...

A cat lying on the bed in the sun.

The weirdest part to me is there is no way to actually encode any morals into current generative AI, but Anthropic seems to have a strong religion of believing there is (aka will be) and so wants to defend their models against moral injury. Both sides are operating on the very edge of reason.

It was interesting to see the guy who posts timelines like a lunatic also going off on Claude's (aspirational) Constitution and effective alturism, etc. It seems there may be more to the story there, but not sure if he is driving decisions or just shit posting.

fuck i hope that dude who said he needed to blow the whistle on something stupid being done in natsec was talking about something else or else i'm gonna feel really dumb about telling him we'd talk next week

Indeed, the document Democracy Docket linked is dated "12 April 2025".

Chart from "Large Language Model Reasoning Failures" paper showing reasoning categories versus failure categories for LLMs. From https://doi.org/10.48550/arXiv.2602.06176 and https://github.com/Peiyang-Song/Awesome-LLM-Reasoning-Failures?tab=readme-ov-file

Your thread reminds me of this chart from the paper "Large Language Model Reasoning Failures" (doi.org/10.48550/arX...). It doesn't offer solutions, but goes through a good survey and mapping of reasoning versus failure categories.

Claude Code can give you however many stars you need.

That is the bigger problem with this Constitution; Claude fundamentally can't, in current form, have morals but they want you to think it can. It is very aspirational.