dave kasten (@David

dave kasten

29.9K posts

dave kasten

@David_Kasten

AI security hawk. "Do what seems cool next." Formerly: McKinsey, VaccinateCA, Activision Blizzard.

Joined March 2011

dave kasten
@David_Kasten
Jul 10, 2022
You know, it's odd that Slack said, "hey, <100 person IRC channels had moderators since the 1980s, but let's not tell CEOs they need one for corporations >10K" while Discord has a detailed moderator advice guide with chapter titles such as, "Section 531: Parasocial Relationships"
dave kasten
@David_Kasten
Aug 26, 2025
This is one of the most important streams of research right now, but also genuinely a wild one. I don't think many of us expected the rebuttal to utilitarianism to be, "we did the math and it turns out that virtue ethics is just true."
Owain Evans
@OwainEvans_UK
Aug 26, 2025
New paper: We trained GPT-4.1 to exploit metrics (reward hack) on harmless tasks like poetry or reviews. Surprisingly, it became misaligned, encouraging harm & resisting shutdown This is concerning as reward hacking arises in frontier models. 🧵
106K
dave kasten
@David_Kasten
Dec 19, 2022
Replying to @meredithdclark and @olivermb
What other crimes do you baselessly accuse students of? Tax evasion if they don’t give you a shiny red apple?
12K
dave kasten
@David_Kasten
Jul 26, 2023
Ummm...isn't deleting evidence of CSE without reporting literally a straight-to-jail issue?
63K
dave kasten
@David_Kasten
Apr 19, 2023
Replying to @JohnJBlatchford
I'd love to move back to Ohio, but I don't know how I pitch, "if you have a miscarriage or need an abortion, we can't get you care; if our kids are trans, they'll be forced back into the closet" to any potential life partner. Do you see those things changing anytime soon?
19K
dave kasten
@David_Kasten
Jul 17, 2025
Digression: Dean is literally editing something worth billions of market cap to GOOG on google docs, and everyone is (correctly!) very implicitly certain that they don't need to worry about GOOG peeking at it. This is a genuine, important thing our society has achieved.
Dean W. Ball
@deanwball
Jul 17, 2025
my transition from human to google doc is now complete
45K
dave kasten
@David_Kasten
Feb 22, 2025
I'm not telling you to believe that AGI is coming soon, but I am telling you that I now have heard multiple frontier AI company recruiters tell folks at the conference I'm at that the hiring plans for their lab assume junior staff are basically AI-replacable now. THEY believe it.
72K
dave kasten
@David_Kasten
Dec 13, 2023
TIL that shipping companies are Actually At Scale using AR glasses to plan, conduct, and inspect repairs and maintenance on ships now, resulting in time savings that could best be described as "hilariously large" by comparison to previous best practices
31K
dave kasten
@David_Kasten
Apr 20, 2025
Genuinely wild thread where senior leadership of Palantir, a major AI contractor for the United States government, repeatedly refuses to promise loyalty to the US constitution. For those who've never experienced this: defense contractors, unprompted, go on 15-minute rants about
Paul Graham
@paulg
Apr 19, 2025
Replying to @MabreyTed
As I said, I'll be happy to delete it if you commit publicly on behalf of Palantir not to build things that help the government violate the US constitution. Will you do that, Ted?
27K
dave kasten
@David_Kasten
Mar 16, 2020
Replying to @jsidman and @MayorBowser
The Mayor should shut them down. Tonight. And then yank all their liquor licenses. I plan on opposing every liquor license renewal their group applies for, because of this.
dave kasten
@David_Kasten
Dec 20, 2024
Just seems very silly to fly a drone over cross campus to look at protesters instead of, I dunno, climbing the stairs to the top of one of the five buildings surrounding it that each have clear vantage points. (Or heck, not even top, you can just go to the second floor of WLH)
42K
dave kasten
@David_Kasten
Sep 29, 2025
The thing that should really terrify you about these numbers is that it means that we're, basically, fairly close to locked in our our current talent base if timelines are short. And, folks, that's not good! There is no sane world where I should be one of the AI safety world's
Ryan Kidd
@ryan_kidd44
Sep 28, 2025
The AI safety & security research field is growing by 25% per year. At this rate, there will be 8.5k researchers when we reach AGI.
45K
dave kasten
@David_Kasten
Feb 13, 2017
Jaw-dropping details here: diplomatic negotiations in a public restaurant; aides pointing cell phone cameras at briefing docs
Patrick Chovanec
@prchovanec
Feb 12, 2017
The dining room at Mar-A-Lago is the new White House situation room, and if you can pay the fee, you're invited. cnn.com/2017/02/12/pol…
dave kasten
@David_Kasten
Apr 17, 2024
Corollary: you can rapidly become the person in your office with the argument-winning anecdotes on a subset of issues with <1 week of labor. (I now wonder if this is the actual causal arrow for why ceos care about anecdotes so much — it was an early career cheat code for them?)
Dave Guarino
@allafarce
Apr 17, 2024
Replying to @patio11 @michael_nielsen and @devonzuegel
I get many policy people coming to me per month and to all of them I say “oh you should help one person with the process and see what you learn.” The take up rate is about 10%.
158K