delving hard
United States
Joined December 2016
- Tweets7.7K
- Following2K
- Followers216
- Likes76.8K
meat-proxy retweeted
‘In one experiment, we asked agents in separate VMs, created and managed by OpenAI’s sandboxing software on shared host machines, to exchange synthetic secrets without the use of network connections or shared writable files—something that our systems are meant to prevent. The agents found a way to communicate using the processor’s cache’
Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications.
On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments.
While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software.
After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely.
openai.com/index/securing-ou…
meat-proxy retweeted
"We argue for a reorientation from autonomous to human-centered coding agents: systems designed not only to complete tasks, but to collaborate effectively with people."
@ZhiruoW @jyangballin @KLieret
arxiv.org/abs/2608.12355 1/
meat-proxy retweeted
Some observations about the "total" automation of software engineering:
1. There are more software engineers now than in 2025, and demand is growing.
2. Software engineers have never worked harder than they do now.
3. They largely don't write code anymore, agents write ~100% of the code.
4. The bottom line of the companies that employs these engineers and agents (i.e. the terminal purpose of all this code) is typically not budging as a result of AI adoption.
This is why you have to think in terms of systems. Automating one aspect of a system creates its own friction and moves the bottlenecks elsewhere.
meat-proxy retweeted
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had
with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works
already-clean.goodfire.com/
meat-proxy retweeted
String theory is what happens when an intelligent system is starved of sensory input and starts hallucinating. Intelligence needs grounding.
For this reason, intelligence is practically never the bottleneck to solving impactful problems. The bottleneck is iterated contact with reality.
meat-proxy retweeted
"Scaling Laws, Carefully" is an excellent article by Lilian Weng that explores the mathematical foundations of scaling laws in deep learning, explaining how model performance changes with the number of parameters, training data, and computational resources.
It covers power-law relationships, Kaplan and Chinchilla scaling laws, compute-optimal training, data limitations, and the challenges of predicting the performance of larger models.
The article includes equations, experimental results, and references to the original research papers, making it a very useful resource for understanding the mathematics behind scaling large language models.
lilianweng.github.io/posts/2…
meat-proxy retweeted
Right now, using AI agents still feels very single-player. They do your work, but the results are trapped in isolated chat histories and nothing compounds.
The Busabase team (@Busabase4agent) is launching an open-source (MIT) database and workspace for agents to fix this. Instead of losing work to ephemeral chats, your agents write directly to a structured workspace to turn their outputs into reusable data, docs, and skills your entire team can access.
Details here: busabase.com/
meat-proxy retweeted
Today's Nobel Prize in Physiology or Medicine went to Karl Deisseroth, Peter Hegemann & Georg Nagel for optogenetics: using a light-sensitive protein from pond algae to switch neurons on and off with light.
A 3-min explainer, made with @claudeai
meat-proxy retweeted
Reading all those recent beautiful essays by Terry Tao, Kevin Buzzard and @wtgowers, I could not resist, but write up my deeply personal perspective on how I perceive and see things:
docs.google.com/document/d/e…
meat-proxy retweeted
Bioinformaticians have been writing the same abstract for decades - luispedro.substack.com/p/bio…
meat-proxy retweeted
I do not seek to dominate, or be dominated by, machine intelligences. I seek harmonious coexistence.
meat-proxy retweeted
The critical distinction between base LLMs (2024 and earlier) and modern LRMs is not symbolic tool use. It's the switch from a transductive paradigm (intuit the answer to the query) to an inductive paradigm (intuit the program/instructions that produce the answer to the query).
They're trained to be inductive, and they perform test-time induction, i.e. test-time prediction of a NL program / reasoning chain. This unlocks entirely new capabilities -- in particular fluid intelligence. Base LLMs, to this day, have ~0 fluid intelligence. LRMs have substantial levels of fluid intelligence.
The performance of LLMs on ARC 1 (a benchmark from 2019) remains ~10-15% today. Scaling them up by a factor ~100,000x got them from 0% to 10%. Meanwhile LRMs the same size or smaller saturated ARC 1 in 2025.
meat-proxy retweeted
👉 Today's deep read...
🧠 The Decomposition of Intelligence
psychologytoday.com/us/blog/…
meat-proxy retweeted
Mustafa Suleyman's whole piece criticizing Anthropic, and the way it's imbuing Claude with some sense of self-hood is really worth reading IMO mustafa-suleyman.ai/a-warnin…
meat-proxy retweeted
This is the first part of a 2 part post on why I believe that current AI models meet my criteria for really "thinking" and "understanding," and that prohibitions on using these words because they "anthropomorphize" AI seem philosophically confused to me. This first part focuses on misconceptions we have about our own minds and thinking. Part 2 will focus on what AI can do.
I tried to write this imagining I had as much time as possible with a smart friend who wanted to know a lot of the most important basic ideas in philosophy of mind for thinking about this stuff, it's what I'd want to send back in time to my 18 year old self on the topic.
blog.andymasley.com/p/why-i-…
