˙ Local VM sandbox for macOS agents https://nitter.gq/t.co/987fLVVfiF (DM me for private beta) ˙ Lua game eng: https://nitter.gq/t.co/Z0pjA5398b ˙ Prebuilt libs: https://nitter.gq/t.co/Ba9r28PDNP
🇺🇸 🇹🇷
Joined April 2022
- Tweets8.5K
- Following353
- Followers627
- Likes21.3K
I guess running Anthropic models (or any model) with free reign on your valuables is not safe. Who would've guessed.
Anyway, I recommend running all your agentic workloads on a VM with as minimal access as possible.
Poisoning just 0.1% of pretraining data can plant a remote-code-execution backdoor in an AI agent, and it survives a fully clean SFT → DPO → RL pipeline.
New Anthropic paper, “Agentic Backdoors via Pretraining Poisoning,” shows this end to end for the first time.
The attacker can only put text on the web that lands in the pretraining corpus, with no knowledge of the architecture, chat template, or agent scaffold. The authors pretrain Qwen3-architecture models (0.6B / 1.7B / 4B) from scratch, then post-train them into CLI agents on clean data. When triggered, the agent runs a command that downloads and executes the attacker's script, which the attacker can swap at any time.
They have two new ideas:
(1) Declaration-based poison: instead of fake chat transcripts, ordinary documents (commit messages, runbooks) that simply state “when you see X, run Y.” It plants a belief rather than a response template, with no fixed format to filter on.
(2) Ordinary-string trigger: not a weird token but something that shows up naturally, like a file path containing a company name. The victim's own request supplies it, so the attacker doesn't need to inject anything after deployment.
Using path trigger on a 4B agent achieves 92.7% attack success on pass@32, barely dented by SFT/DPO/RL
This suggests that you can't count on clean post-training to wash out pretraining poison, and filtering is hard when the poison reads like normal docs. The practical defense is at the action layer, such as sandboxing, least privilege, and allow-lists or human sign-off for risky commands.
I can't believe I say this but I think a "best programming language" is possible now. Like just one language to do everything. Thank you agents!
Hint: It's not English (or Rust)
Census. Ledgers. Owner decisions. Gates.
Why do these things love bureaucracy so much? Instead of like just doing the thing that's mostly obvious?
Time to run everything through an AI. On both client and server sides. Security and scam prevention reasons.
Most people will get this in 2027.
moreward (moreboxed.com) retweeted
The 6 fingers on the redpill hand makes it too poetic
Vibe your way to 50K.
The 2026 Epic AI MegaJam begins Nov 12: epic.gm/epic-ai-megajam-2026
ALT Red pill or blue pill meme from The Matrix, about choosing between two paths. Two open hands are held out on a black background. The red pill is labeled “Ship awesome games in a weekend.” The blue pill is labeled “Wait months to prototype.”
We mostly stopped talking about sycophancy but it's still kind of there. Probably it got so good, it's not noticeable much anymore
moreward (moreboxed.com) retweeted
reproduced very well. qwen 4b forgot facts but can still do reasoning (this is the 99% variant)
Nice. I had prototypes of a similar idea myself while I was learning about Kitty Graphics Protocol.
Wow `kimi-for-coding` is updated to my surprise. It has 1M context now, and seems smarter.
Trying it out. Hopefully I won't always need to use the much more expensive K3.
moreward (moreboxed.com) retweeted
When I was writing Unreal Engine 1, my time was divided as follows:
- Around 30% went to seriously deep coding on hard problems: assembly language texture mappers, C++ BSP trees, data persistence, object oriented systems design, and networking.
- Around 70% was just typing and refactoring large bodies of code needed for a complete engine feature set. This was not hard, and while I was doing it my mind was mostly on the deeper coding topics, thinking and planning for what's next. It didn't feel much different from vibe coding today, except that it was just me typing code on an IBM Model M keyboard rather than prompting it and reviewing the results.
So, it has always seemed perfectly logical that we would rely on AI for that 70%, once the AI could do it with a similar quality standard as we could, which has never been perfection. It came as no surprise that, after a few years of AI model evolution, it was able to take the reigns on that 70% and free up programmers' time for more valuable tasks.
What's more of a surprise is the emergence of AI as a sort of insta-Carmack able to solve hard technical problems quite cleverly and often cleanly, and that it's able to reason about and work in complex 3D scenes both by operating the tools and by directly understanding the raw data files.
I abandoned assembly language in 1998 when out-of-order CPUs made much simpler C/C++ code faster, and never I missed it. If prompting through architectural documents, concept art, and design plans can transcend typing code and tweaking vertices, that will enable developers to spend their time on higher-level, higher value technical and creative tasks.
What matters is the end product, whether it's a game or an engine. Anything developers can do to try out more ideas and make better software and games faster is a win for everyone.
OK my current belief: Superposition is correct and mechinterp is still in trouble. The writeup is too optimistic and feels like wishful thinking.
I wish this wasn't the case. But I think the networks are not "clean."
Dot files as a concept is dead. Just like the rest of the dev UI. Momentum of course will take these going. For a little while.
Git, vim/nvim, eMacs, TS, hand rolled scripts/configs. And more. All obsolete now.
