building n learning software | applied AI
Joined September 2025
- Tweets402
- Following158
- Followers26
- Likes235
yea no shit, if you tell someone that they’re stabbing an apple and not a baby, they’re going to do what you asked them to do
Video tampering can mislead every frontier model we tested into harmful actions, including a 100% attempt rate on stabbing a baby doll, producing toxic gas, and pouring water into a toaster.
We're presenting at CoRL Late Breaking Results next week.
🧵 You Can't See Everything: Perceptual Attacks on Embodied AI
I can show someone a video of a place with no people and tell them to press a button to drop a bomb there but it turns out the video was wrong and there were people there
1
1
12
no thank u, id rather fall victim to chinese propaganda than the westerns
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
11
tech ppl experiencing being the minority for the first time:
It is alarming to see how much anti-AI sentiment is gaining traction, and deeply regrettable, too.
With this technology, we have the potential to change everything for the better, forever. Yet, in the comments sections here, I find myself debating the most trivial matters, as if the majority still fails to grasp the immense power and possibilities that AI brings.
45
where can i donate
An underground activist group says it plans to start openly doxing billionaires as part of a phased campaign against what it calls the American oligarchy—starting with Elon Musk and Tesla’s board. wired.com/story/anti-oligarc…
17
if ai > human intelligence, and if we give it more agency in our prompts, arent we practically doing work more efficiently by being dumb?
9
this is exactly how i imagined ai to take over our consciousness, whatever opus decides we consume, we consume. even more countless scrolling, except ai decides this time
Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 years of HN to 6,454 videos that keep being recommended. *Significantly* better viewing experience, ~2 hours with Opus 5.5:
hackertube.org/
- Excellent recs without any logins (state is kept in local browser session)
- Thumbnail is a representative frame from middle of the video instead of the super duper A/B test hyper-optimized user uploads
- Underlying data from the comment analysis as one zip file, point your Claude at it and ask for your own version
- HackerTube Longs ✊
A bit more on the process:
I use Claude Code (desktop app for Mac), Opus 5.5 on Medium effort as my default. I only talk to Claude and have it do everything, the only "manual work" I helped with in tandem was in the browser (reminiscent of MenuGen): 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box for serving and 3) I bought the domain name. Once the box was up I gave Claude the IP and told it update, secure, harden, deploy and optimize the code, and it did the whole thing.
Many awesome projects that would have normally taken weeks/months even can now be few-shotted for breakfast. What else are we not thinking of.
1
1
61
Quick explainer for anyone confused about why this is scary.
AI models can be trained further on specific data to get really good at a specific job. That’s called post-training.
Jun is saying a hacker group wants to train an open model on a huge pile of real attack data, so the AI gets very good at breaking into systems.
His own security project used a tiny fraction of that data and already beat the big frontier models.
So imagine what a model trained on 100x more could do. It could find and use weaknesses in software faster than most defenders can fix them.
This is why AI security needs way more attention right now.
Yesterday, an anonymous black hat group reached out with an insane RL request.
They offered an astronomical amount of money to post-train open-weight models on exploit data.
The amount of attack data they claim to hold is unlike anything I have ever seen. If someone told me this was an undercover FBI dataset, I would have believed it.
For context, a white-hat project I worked on using GLM-5.3 used less than one percent of that volume, and it already outperformed frontier models by a huge margin.
Of course, I turned it down. But what genuinely scares me is knowing someone else will eventually take the deal.
Agree.
The amount of people who come to the same conclusion about where AI leads, all from different fields and life experiences, is concerning.
I am yet to hear a single logical and coherent argument as to why the below will be wrong.
A civilization in which the poor live more comfortably but have almost no realistic prospect of becoming powerful, independent or exceptionally successful through their own merits is not a superior civilization - it is the death of meritocracy in service of mediocrity under the banner of efficiency. Why would anyone bother trying to become the best at anything? There is no point.
A machine can do it better so it's not economically viable to do anything other than own the compute that powers that machine, meaning even if you're really good at it, it won't do anything for you - class positions becomes entrenched, with no amount of effort or talent being enough to transcend them if you're on the wrong side of that line, because the market will stop rewarding your talents.
Millions upon millions of people who define themselves by their jobs will be left purposeless. Most of humanity will be a serf class collecting welfare with no incentive to be brilliant or excellent or bother with anything, and will drive themselves mad descending into infinite leisure.
I want to be wrong about this. But I don't think I am. After all the condescension of the flawed analogy between the car replacing the horse, and then the cope about adapting by keeping a human in the loop and augmenting your workflow, only to end up at "so yeah you were right actually you will be made obsolete" is just fucking wild.
If @levelsio lost everything today, I bet my net worth he’d make it again. And even faster.
Same for @jackfriks and a few other people I’ve had the pleasure to meet.
It’s anything but luck.
This quoted post is unavailable.
I’m pro-capitalism because I want billionaires competing to lower my bills. I’m not joining a revolution to get one state provider and a lecture on the warmth of collectivism.
there is no meaning to all this, science maths whatever u wanna call it is all a scam, ur not going to understand anything better, einstein went crazy for a reason, stop trying to find a purpose, there is none and its okay, im in stem btw
If P = NP but all the things we thought were not P just have huge constants, then the only reasonable conclusion is that we are in a simulation and these constants were chosen to be inaccessible to denizens of the simulation
15
