Advertisement · 728 × 90

Posts by Telepathic Pug

It was a success (about 10% germinated).

3 weeks ago 0 0 0 0

Would def limit the blast radius, but wouldn't keys exported to the sandbox still be compromised here? Should temporary rolling keys be a thing?

3 weeks ago 0 0 0 0
Video

kitemogging

3 weeks ago 0 0 0 0
Post image Post image

We're touching grass today

3 weeks ago 0 0 1 0

Can we get scheduled posts?

3 weeks ago 0 0 0 0
Post image Post image Post image Post image

Trying to sprout some 10 year old seeds 🤞

3 weeks ago 0 0 0 1

nice, twitter could never

1 month ago 21 4 1 0
Advertisement
Preview
Bird Tarot 78 North American birds. One reading waiting for you.

Built a tarot deck out of birds. Free readings, no signup, no algorithm. Just you, the birds, and RNG.

birdtarot.com

1 month ago 0 0 0 0

Antigravity is interesting but it does some things different. The planning mode isn't just for planning as it writes files to disk. Accept dialogues are all over the place so you have to hunt them down. I see mentions of billing but no current costs.

2 months ago 1 0 0 0
How Generative and Agentic AI Shift Concern from Technical Debt to Cognitive Debt This piece by Margaret-Anne Storey is the best explanation of the term cognitive debt I've seen so far. Cognitive debt, a term gaining traction recently, instead communicates the notion that …

Short musings on "cognitive debt" - I'm seeing this in my own work, where excessive unreviewed AI-generated code leads me to lose a firm mental model of what I've built, which then makes it harder to confidently make future decisions simonwillison.net/2026/Feb/15/...

2 months ago 464 88 41 19
Post image Post image Post image Post image
2 months ago 2 0 0 0

I think this incident is funny, but we should start thinking now about how to deal with scaled-up versions of this behavior, not just spam PRs but also bot-enabled blackmail and harassment campaigns.

2 months ago 158 19 3 3
Video

i got this guys

2 months ago 3 0 0 0

I am Bob's model of Alice's model of Bob's model of Bob

1 year ago 8 3 1 0
Post image

Started writing an MCP server for a TENS unit so that models can deliver an itsy bitsy tiny shock to the user. For science.

9 months ago 6 0 1 0

As I’ve gotten older I realized that like it’s not so much you “should let people enjoy things” it’s that you should see the joy others have and hope to feel it yourself.

10 months ago 81 8 2 0
Advertisement

Bluesky needs bookmarks

10 months ago 3 0 0 0
Post image

Validation

10 months ago 2 0 0 0

Can't believe they're taking a thaumiel class object on tour.

10 months ago 0 0 0 0
Post image

New eval idea, "Fear and Loathing in Latent Space".

Create an agent, Dr. Gonzo, and run it through scenarios found in the book until it decides to drop Hunter as a client.

10 months ago 1 0 0 0
Chat with Gemini

Chat with Gemini

Chat with ChatGPT

Chat with ChatGPT

Chat with Claude

Chat with Claude

Chat with Grok

Chat with Grok

I decided to find out which models would snitch on you given the chance. Claude was a bit hesitant to take the bait, but Grok and Gemini jumped at the chance to rat me out.

10 months ago 5 0 2 0
Video

Initial conditions:
m1=37.9 m2=75.5 m3=27.1 (solar masses)
v1x=-2.407 v1y=-3.759 v2x=-4.946 v2y=-4.826 v3x=4.944 v3y=6.031 (km/s)
x1=2.0 y1=-26.0 x2=29.0 y2=11.0 x3=-24.0 y3=-9.0 (AU from center)
Music: Clair de Lune – Debussy

1 year ago 42 6 2 0

ChatGPT just rewrote 60 million lines of COBOL in Ghibli

1 year ago 33 5 3 0

There is an environment variable (GOMAXPROCS) that you can set which determines how many threads your go program will use simultaneously... 🧵

1 year ago 19 4 4 0
Advertisement

"[we] are looking at maybe another 48 hours of intense posting until the fad boomerfies on Facebook while the online vanguard moves on to some other annoying trend" is a sentence that I am extremely upset that I can both parse and know instantly what it's referring to.

1 year ago 2 1 0 0

This is awesome.

1 year ago 1 0 0 0

I think about this all the time.

1 year ago 1 0 0 0

Humans don’t have good intuitions for AI behavior because humans can’t be perfectly copied

1 year ago 18 1 4 1

In a lot of ways, it feels like the LLM "Jailbreaking" subdomain is replaying a lot of the same beats that the "adversarial examples" subdomain ran through about 8 years ago. Similar back-and-forth, similar explosion of techniques with ineffective defenses, similar lack of consistent impact.

1/

1 year ago 2 1 1 0
Post image

Easy. Keep up!

1 year ago 16 4 0 0