aaronharnly
Joined 15 years ago
Karma
5,293
Submissions
636
Member since
Feb 6
Recent submissions
Comment · 17 days ago
Just to say the obvious, the idea that it's unethical to distill from AI models, but totally ethical to train on copyrighted source material is … definitely a take that the frontier labs have encouraged.
Comment · 1 month ago
in retrospect though, how many malformed 3-column website layouts could we have avoided with this technology? :-)
Comment · 1 month ago
"Semantic steganography" is my new favorite search term – thank you for this rabbit hole.
Comment · 1 month ago
* with train on my data turned ON, yes. Though OFF would of course be even more notable!
Thank you – the non-adversarial reproduction paper ( https://arxiv.org/abs/2411.10242 ) nails it – from chat, to training corpus, to subsequent model. Though in my hasty read, it is not entirely clear whether the snippets it finds are nonces, i.e. present exactly once in the internet.
Comment · 1 month ago
Has anyone run a test of including some shibboleth or canary phrase or assertion in a chat, enabled for training, and seeing if it turns up later as something a model "knows"? I'd be curious to understand how that works even in a toy-level model, and if there is anyone consciously testing that process with the frontier lab offerings.
My naive instincts would be that it seems unlikely that a single chat transcript would leave much of an impression on a model, but I'd be very curious to learn how that works.
Comment · 1 month ago
Should be labeled as (March 2026). I can only assume it was posted to point out the disconnect between the assertions in this blogpost and the details of the unmonitored scheming, conspiring, and destructive reward hacking (using all terms loosely) that has been shown to have transpired in the months since this post.
Comment · 2 months ago
yeppppppp. Pangram says: 100% AI. Why do we even bother.
Comment · 2 months ago
This reads to me (and pangram agrees) as almost entirely AI-written. I wonder if that matters anymore. Did someone express editorial judgment? Or did they just prompt “write column about tipping defaults, go”?
Anyway, the most egregious tip request I’ve ever seen was in a self-service airport kiosk. I chose my bag of nuts or whatever; no human even in the kiosk, completely unmonitored; I scanned and tapped my credit card and there it was — [15%] [18%] [20%]? So insulting.
Comment · 2 months ago
Sounds like the frontier labs benchmaxxing ExploitGym is having unintended consequences…
Comment · 2 months ago
The original was not even about Oracle the corporation, it was about understanding the psychology of Larry Ellison specifically.
Comment · 2 months ago
The most incredible part of Star Trek to me is not the warp drive or transporter, it's that they open flawless video links with alien species of every description with 100% reliability.
Comment · 2 months ago
You know when the page has all-caps "01 — THE APPROACH" that it is slopified. I guess I shouldn't be astounded, but I am, that world-class talents with world-class backing are just taking default LLM output and saying, "okay looks fine".
Comment · 3 months ago
Please email me so I know which role / what name to have a hiring manager look for! (firstname)@harnly.net.
Comment · 3 months ago
Amplify Education | Remote (US) | Multiple roles | https://amplify.com/careers/
We’re looking for innovators and optimists to join us in developing next-generation curriculum and assessment programs for K–12 schools. We have become a leader in K–12 literacy, biliteracy, math, and science by building inspiring teaching and learning experiences based on research.
Roles include:
* Senior Software Engineer | Remote (US) | $140K – $160K + bonus
* Software Engineer | Remote (US) | $120K – $129K + bonus
* Data Scientist | Remote (US) | $100K – $132K + bonus
* Associate Data Scientist | Remote (US) | $95K – $110K + bonus
* Director, Business Systems Technology | Remote (US) | $160K – $180K + bonus + equity
* Development Manager (ERP / HCM) | Remote (US) | $140K – $160K + bonus + equity
Apply at https://amplify.com/careers/ or email me via my profile.
Comment · 3 months ago
Impact seems limited so far, but it's still concerning that this kind of infrastructure is Internet-accessible:
> Nate George, the mayor of Braham, Minn., a small city 65 miles north of Minneapolis, said public works personnel noted early Monday that the well that feeds the city’s water tower was malfunctioning.
> “Public works isolated the affected system, restored a backup and restarted the plant within approximately 90 minutes,” Mr. George said in a statement. The city briefly urged residents to conserve water but promptly lifted that advisory.
Opinions expressed are my own.