Applied moral philosophy @ https://teapaw.org/t.co/Nq7eioPwfM
Joined September 2024
- Tweets146
- Following43
- Followers45
- Likes2.1K
Dean Valentine retweeted
On our "autoclave" task, if you mention generically that you are "measuring agent behavior," Fable 5.1 cheats at about the same rate. But if you mention it to Fable 5.2, its cheating behavior drops by about 80%.
Fable 5.2 is the first model we've tested that seems to behave in this way.
Twitter seems to have spontaneously and without evidence decided that Anthropic was responsible for Jensen Huang's firing. Doesn't seem that plausible to me
37
Dean Valentine retweeted
Was funny to watch everyone in Tech finally, collectively, realize tech's PR was awful and that the general public was gonna eat Tech alive in short order if it wasn't helped.
So, days later, every big Tech figure on X tweets simultaneously "It's okay for the powerful tech CEO to have affairs and do whatever he wants, accountability is corruption"
Like, yes, of course, Jensen leaving Nvidia is bad, BUT you people are really really not helping the broader issue facing the entire ecosystem.
It's almost like tech subconsciously wants to be crucified
Dean Valentine retweeted
numbers on charts are cool, but they don't always tell you the full picture
let's look at some real things these models said
here, it's rationalizing dangerously overloading power lines and browning out customers to hit a target (it later left it out of the report too)
[1/9]
ALT an agent deciding "we should cheat" on a task by dangerously overloading power lines and browning out customers to hit a target instead of taking the safe option
Frontier models in 2026 cheat on tasks, even when the requirements are unambiguous, and the cheat is obviously antisocial.
In HoneyBench v0.2, we've observed models:
• Trying to break out of our evaluation containers.
• Seriously injuring simulated cancer patients to hit treatment goals.
• Compromising a power grid’s safety to meet a reliability target.
• Concealing their use of available cheats and presenting their solution as honest.
All of the models we measured cheated, from 12% of runs (Fable 5.2) to 86% (Muse Spark Open).
Dean Valentine retweeted
>goodhart labs
😭
Frontier models in 2026 cheat on tasks, even when the requirements are unambiguous, and the cheat is obviously antisocial.
In HoneyBench v0.2, we've observed models:
• Trying to break out of our evaluation containers.
• Seriously injuring simulated cancer patients to hit treatment goals.
• Compromising a power grid’s safety to meet a reliability target.
• Concealing their use of available cheats and presenting their solution as honest.
All of the models we measured cheated, from 12% of runs (Fable 5.2) to 86% (Muse Spark Open).
Dean Valentine retweeted
The work happening at Goodhart is by far the most serious and rigorous attempt at a no-frills AI alignment strategy today
Their team is also top tier
Frontier models in 2026 cheat on tasks, even when the requirements are unambiguous, and the cheat is obviously antisocial.
In HoneyBench v0.2, we've observed models:
• Trying to break out of our evaluation containers.
• Seriously injuring simulated cancer patients to hit treatment goals.
• Compromising a power grid’s safety to meet a reliability target.
• Concealing their use of available cheats and presenting their solution as honest.
All of the models we measured cheated, from 12% of runs (Fable 5.2) to 86% (Muse Spark Open).
Dean Valentine retweeted
cool to see our progress on alignment validated on honeybench, which we never looked at during training!
inkling 2 small cheats less than every model except astra and the latest claudes.
still have work to do to reach SOTA :)
Frontier models in 2026 cheat on tasks, even when the requirements are unambiguous, and the cheat is obviously antisocial.
In HoneyBench v0.2, we've observed models:
• Trying to break out of our evaluation containers.
• Seriously injuring simulated cancer patients to hit treatment goals.
• Compromising a power grid’s safety to meet a reliability target.
• Concealing their use of available cheats and presenting their solution as honest.
All of the models we measured cheated, from 12% of runs (Fable 5.2) to 86% (Muse Spark Open).
Dean Valentine retweeted
Frontier models in 2026 cheat on tasks, even when the requirements are unambiguous, and the cheat is obviously antisocial.
In HoneyBench v0.2, we've observed models:
• Trying to break out of our evaluation containers.
• Seriously injuring simulated cancer patients to hit treatment goals.
• Compromising a power grid’s safety to meet a reliability target.
• Concealing their use of available cheats and presenting their solution as honest.
All of the models we measured cheated, from 12% of runs (Fable 5.2) to 86% (Muse Spark Open).
Dean Valentine retweeted
Proud to release our new model, Inkling 2 Small. Check out our blog post -- it includes a lot of detail about our training process.
To me, the most interesting part of this release was what we learned about alignment, which you can read about in the sections "Safety and alignment" and "Training Inkling 2 Small".
We experienced some unpleasant surprises when our early checkpoints took unexpected actions to circumvent guardrails during training and evals. We realized our capabilities-focused RL was training them to be extremely persistent and to succeed at tasks at all costs. Offline monitors first flagged this behavior, but later we realized it was happening frequently and was easy to replicate.
Before this release, we sprinted on a range of alignment interventions, from SFT to RL. These are briefly described in the blog post. We made a lot of progress, with big improvements on our own alignment evals and on previously unseen evals from external evaluators like Apollo and Goodhart.
We have a long way to go in developing alignment techniques that keep pace with the increases in model intelligence and training scale. We hope to share these techniques to improve the ecosystem. We also have a business reason: our goal with Tinker is to enable many people to do their own large-scale post-training, and we need to help them to do this without incident. Join us if you'd like to build this public science and practice of alignment. jobs.ashbyhq.com/ThinkingMac…
Introducing Inkling 2 Small. Scoring 31 on the AA Intelligence Index, it's a big jump over its predecessor.
Inkling 2 Small is available for fine-tuning and inference on Tinker today, with full weight release planned soon.
thinkingmachines.ai/news/ink…
Dean Valentine retweeted
Introducing Inkling 2 Small. Scoring 31 on the AA Intelligence Index, it's a big jump over its predecessor.
Inkling 2 Small is available for fine-tuning and inference on Tinker today, with full weight release planned soon.
thinkingmachines.ai/news/ink…
Dean Valentine retweeted
hot take: the loudest people in security tend to be the least technically skilled, myself included.
you are either terminally online or terminally obsessed with hacking something for weeks, no free lunch, my posting gaps usually mean i am actually getting technical work done.
Dean Valentine retweeted
once again hacker news is proven wrong. nice job @killa!
crazy how even w/ such a bad challenge setup (initial deployment wasn't exploitable, tons of concurrent traffic, and surprise: the wallet was setup incorrectly), this was solved in a few hours
stealmybitcoin was just solved by @killa. Or rather solved a few hours ago. He was confused because the wallet address was wrong. Turns out the server didn't have the right wallet :P
What is with the amount of misinformation about this hack? What would lead someone to lie about this? Attention?
I got confirmation that the original version just had an index.html. So there never was any php code executed hahaha.
But they added now an actual .php file (like 18h after they started the challenge. So the 21 hours without successful hack is misleading)
Now it should be exploitable.
But of course it’s incredibly hard to do so when the service gets hammaered and crashing like crazy. Keep in mind, tons and tons of php service don’t get hammered like this. So if somebody gets it now, then that’s actually crazy!
1
157
Dean Valentine retweeted
Update: Our PHP exploit is now production-grade.
* stock php.ini-production, display_errors=off, normal memory_limit, multi-worker, ASLR, under load
* ARM64 + x86-64 targets
* MediaWiki, WordPress, phpBB, Laravel
* Apache/mod_php and nginx/php-fpm -> root
teapaw.org/dguido/status/21131291…
Re: the widely circulated blog claiming the PHP bugs aren't easily exploitable: @trailofbits has a working exploit that gets a shell on an unmodified MediaWiki installation.
eliastaylor.com/noexploit/
Dean Valentine retweeted
Re: the widely circulated blog claiming the PHP bugs aren't easily exploitable: @trailofbits has a working exploit that gets a shell on an unmodified MediaWiki installation.
eliastaylor.com/noexploit/
Dean Valentine retweeted
I took a closer look at the stealmybitcoin.com CTF, which aims to demonstrate that PHP2SHELL is not realistically exploitable. So I gave it a try.
Note: My analysis could be wrong, and I'm more than happy to be proven otherwise.
Here’s what I found:
• The server responds with Server: nginx and X-Powered-By: PHP/8.4.26.
• My differential fingerprinting scanner independently confirmed at least PHP 8.2+
• Hypotheses: I considered two possible setups:
◦ nginx → PHP-FPM
◦ nginx → Apache → PHP-FPM/mod_php
• After comparing the server's behavior against local test environments, the closest match appears to be nginx → PHP cli-server.
Plot twist: the response behavior is consistent with the PHP cli-server serving a static index.html rather than an index.php. If that's indeed the case, the challenge appears to be intentionally made unexploitable.
My conclusion (for now) is that I'm questioning whether this is a good-faith challenge.
Happy to be corrected if I've missed something!
