Kerrick Staley

hacker / climber / urbanism fanboi

[Kerrick Staley @Kerksta]

@StephenLCasper Can you give an abbreviated version at BKCircle? I’d be excited to watch this!

[Kerrick Staley @Kerksta]

@dfrsrchtwts the example Daniel is quoting is from Gemini, checks out

[Kerrick Staley @Kerksta]

@dfrsrchtwts I've seen the stuck-in-a-loop behavior with Gemini specifically, see https://t.co/ESKfXYwaP0 (click "Transcript") and https://t.co/MqslEGbEwj for broader context

[Kerrick Staley @Kerksta]

Do people have a workflow they like for designing floor plans with AI agents? I want to say "what if we put the staircase over here" and have it Just Work. Codex and Claude Code are both really bad at this right now.

[Cas (Stephen Casper) @StephenLCasper]

Apply by June 7 if you'd like to be part of MATS. This autumn, I'm excited to work with MATS scholars on cutting-edge AI safeguards and AI gov!

https://t.co/61pbgnQDHQ

[Kerrick Staley @Kerksta]

@inerati I prompt Claude / Codex to make 3D-printable CAD designs (for small/simple functional things) and I have a few of these around my apartment right now. I see this evolving towards what you want.

[Kerrick Staley @Kerksta]

I'm midway through the Astra fellowship and it's been an incredible learning opportunity, with excellent mentorship and a friendly and thoughtful community. I highly recommend applying!

[Henry @sleight_henry]

❗️Only two days left to apply to the Astra Fellowship!
Apps close EOD SUNDAY May 3rd, AoE. Astra's 5 months, fully funded, @ConstellOrg Berkeley

80%+ of our first cohort now work full-time in AI safety

Mentors include Redwood, AI Futures, TruthfulAI, CoG, IAPS, RAND & more ⏬

[Kerrick Staley @Kerksta]

We found that when one AI monitors another for suspicious behavior, there is an surprising relationship between the TPR and FPR of the monitor that seems to hold across diverse settings and monitor models.

https://t.co/jeSMrkTf25

[Kerrick Staley @Kerksta]

@nabla_theta I find it similar to how China has many "dialects". There was enough flow of people/information/commerce to keep the languages similar but enough people and geographical division for them to diverge (esp before modern telecom tech).

[Kerrick Staley @Kerksta]

I made a baby eval! LLMs can sorta do CAD now https://t.co/MqslEGcclR

[Kerrick Staley @Kerksta]

a confusion matrix is called that because it is confusing

[Kerrick Staley @Kerksta]

This directly follows from the definitions of variance and the formula for the Gaussian PDF, so maybe not that cool. But IDK I found it surprising.

[Kerrick Staley @Kerksta]

Fun stats fact: take any set of numbers, normalize them to mean=0 var=1, fit a Gaussian. That best-fit Gaussian will be mean=0 var=1. The fun part is that the log-loss of the fit will *always* be 1/2 * (log(2 * pi) + 1) = 1.419 regardless of how bell-shaped the data was.

[Kerrick Staley @Kerksta]

short post exploring some stats https://t.co/R8XdFQk8x3

[Kerrick Staley @Kerksta]

Totally cheesed the start of this V8 at Movement Gowanus, and the rest was V4 territory. We take those! Video 3x speed

[Kerrick Staley @Kerksta]

I'm going to try to climb 50 (gym) V6 boulder problems and write 10 blog posts in the next year! Documenting progress the climbs in this thread, and will tweet the blog posts in my main feed.

[Kerrick Staley @Kerksta]

V6 flash (the blue guy) at Movement Gowanus! White V6 has so far evaded capture.

[Kerrick Staley @Kerksta]

V6 2/50! No video, but it’s the black hold purple tag

[Kerrick Staley @Kerksta]

@Prigoose I think the real estate agency they hired is poorly run; doesn't seem like an outright scam but also they would have happily kept my money if I didn't keep making noise.

[Kerrick Staley @Kerksta]

Bonus @patio11 blog post: the "Presenting like a professional" section from https://t.co/1P5uQmKfER teaches a super valuable life skill.

[Kerrick Staley @Kerksta]

@Prigoose The unit is full of the owner's stuff (furniture plus personal effects) and the owner only wants renters who are OK with keeping the stuff there. Weird ask. I wanted to rent it and temporarily move the stuff to storage but we never reached a deal.

[Kerrick Staley @Kerksta]

@Prigoose The nice thing is that once I wrote this up I would be able to directly forward it to @REBNY and the @NYSDOS Department of Licensing if they continued to ignore me.

[Kerrick Staley @Kerksta]

@Prigoose It was wordy and tiresome to write (took over 30 minutes to get all the info together and write it up) but I just summarized all our past communication and asked them (again) to refund me.