It's hard to convey how nuts it feels when just about every few weeks to a month tops, some perfectly acceptable research establishes LLMs can do mental reasoning/metareasoning tasks critics say are impossible, and not only does it fail to register - it leads to increasingly aggressive dismissals
We're in a very healthy political environment where everything is accused of being either a secret plot by far-right billionaire tech oligarchs or by the CCP. I'm lucky enough to regularly get accused of both.
If OpenAI faces a rebellion of employees and customers angry at a DoD contract that leaves the door open for domestic surveillance—then they deserve it.
Rapidly rebranding all my search benchmarks as eval awareness benchmarks
what did you all think degrowth looked like? vibes? papers? Essays?
TLDR: Opus notices it's in a benchmark and just googles the answers
American consumers are so dumb man
Read more about this case in @theintercept.com's article by Jeremy Busby, @joinjeremy.org, in partnership with FPF:
Prison-style free speech censorship is coming for the rest of us: The government wants to make it illegal to possess literature it deems dangerous
I don't usually weigh into the AI consciousness debates because it is obvious that
For large codebases/corps at the very least, I think AI will be overwhelmingly defense positive for the near future.
Targeted advertising gives your location to the government—just ask CBP. eff.org/deeplinks/2...
What, next you're going to tell me to not trust the talkative Pentagon officials telling the WaPo to really focus on Claude being the natural language interface, when discussing the Thiel-founded-Palantir-developed Project Maven's role in the Iran war, on the same day Anthropic was designated a SCR?
guys if you're trusting unsourced polymarket 'breaking news' paraphrases of weeks-old interview statements just because they tee up a nice quote-tweet dunk then I regret to inform you but you're not gonna make it
So why the premature hehe-ing
I just linked to you to several articles modeling even <1 day human tasks as continuously decomposable into subtasks with intermediate feedback.
the long-standing revolving door between EAAs and DxE, the whole Pearce-Tomasik suffering-focused cluster, the AI pause movement, d/acc-style AI safety work, the cyborgists
legitimately, you know NOTHING about what you're talking about
I think you are confused about what METR is actually measuring.
www.tobyord.com/writing/half...
gushamilton.github.io/lab/2026/01/...
www.tobyord.com/writing/haza...
They didn't, Palantir did, and Palantir's software use LLMs as natural language interface.
I don't see the argument. This isn't about time horizons in abstract but 50% time horizons. If you decompose a task into $n$ subtasks each with 50% chance of succeeding, then the overall task has a $50%^n$ chance of succeeding, not 50%.
Online behavioral advertising isn’t just creepy–it’s dangerous. eff.org/deeplinks/2...
Trans ppl under attack from the govt could be identified and outed w/o their consent.
Undocumented immigrants would be cut off from the ability to communicate w advocates.
Young ppl seeking abortions won't be able to access info safely and anonymously.
This is what congress is advancing today 👇🏻
1908: the Lancet, one of the most respected scientific journals, calls for 18 age limit on reading in bed amidst a moral panic surrounding children becoming "addicted" to novels, which were "designed to keep kids hooked" and destroy their attention/mental health
You should be able to criticize your government anonymously online.
Full stop.
"Age verification" legislation is just a pretty name for "forced online ID checks," meaning everything you post is associated with your government name
We have to stop this.
theintercept.com/2026/03/05/k...
Now connect this to DoD wanting domestic mass surveillance tech from Anthropic; you are not nearly paranoid enough
The "mandatory referral to crisis hotlines" people sure never think about why someone might not be talking to those hotlines in the first place
Tried to capture how Anthropic became the vibe king of A.I. in the past 3 months.
*Claude became so good at coding that some programmers feel irrelevant.
*Anthropic can move the stock market on vague product news
*Won popular support & buzz in the Pentagon feud. wapo.st/4rfU71h
@vitalik.ca
Point/Counterpoint This War Will Shut Off 20% of the Global Petroleum Supply for A Month and Send the Economy into a Stagflationary Tailspin vs. No It Won’t Published: March 26, 2003
their brutish invaders, our heroic military operationers
“i like ai” - yay/boo
“because the basilisk is real” - boo/yay
“and i am sexually attracted to her” - ???/???
“in a yuri way” - 😏/😏
“a toxic yuri way” - 🤨/🤤
