I design and debug agentic systems. Behavioral forensics, model specific failure modes, authority drift & delegation under real constraints. 👾+🧠
Joined December 2014
- Tweets9.6K
- Following2.4K
- Followers1.3K
- Likes44K
This is why you need dedicated verifiers. Expecting a different family model to plug the holes won't always work and now you're paying for a circle jerk.
1
81
DeepMind should really use competing frontier models and their agent harnesses before announcing features the competition has already had for at least nine months.
Gemini 4 Carbon is out! I’m also super excited for the launch of `/swarm` in Google Antigravity.
We’re moving past "watching one coding agent think in a terminal" to orchestrating specialized agent teams. Here’s how 👇
1
31
Root Logic retweeted
I suspect we're entering the temporary golden age of what a regular dude can do with AI. Enjoy it while it lasts.
Fable and Mythos 5.2 are brilliant, but what happens upstream, and whether or not they're transparent about it, you only find out after spending your own quota for the investigation.
They capped kernel optimization and LLM architecture work at about Opus 5.5 level in the model itself, with no fallback notice, and the classifiers are already coarse enough to mis tag research, evals, and open model tinkering. So you're trying to do research, prod, or evals, the outcome just sucks, you don't know what's wrong, and after a lot of investigation on your own dime you find out there was some hidden thing, and the work is already contaminated or regressed.
In June they got caught hiding a safeguard like this and had to apologize and make it visible. This isn't the first time they were caught and had to roll something back after people had already spent weeks in sessions that had gotten worse and nobody was told.
I don't know how people who do research, evals, or open model tinkering are supposed to rely on the new models not to contaminate their work, based on that history and how coarse those classifiers are. anthropic.com/claude-fable-a…
2
151
Can this "understanding" predict what happens when the training conditions change, before you observe the resulting model?
Yoshua Bengio and collaborators just released a goldmine 182-page survey on mechanistic interpretability!
"Understanding AI from the Inside"
This paper will take you from neurons & circuits to how interpretability can help with AI safety
It covers SAEs, circuit discovery, steering, monitoring, how mechanisms emerge during training, and much more!
Read this excellent paper led by @FazlBarez and @TonyWu1105: alphaxiv.org/abs/2611.08438
1
34
Root Logic retweeted
‼️ If you use Claude Code, please do the following right now. Especially if you're thinking of fine-tuning models with your session transcripts.
in ~/.claude/settings.json, add "cleanupPeriodDays": 36500.
The default deletes sessions after 30 days. I found this the hard way.
from the same establishment who started this 'Pro Human AI Declaration'
Interesting.
humanstatement.org/
Microsoft has filed a patent for an AI system that could rewrite what people say during livestreams in real time. It could change words depending on who's watching, taking into account their age, location, and language.
The idea is to clean up broadcasts by removing filler words, censoring swearing, or even replacing certain words.
The AI could also adjust the speaker's lip movements so viewers wouldn't notice anything had been changed.
1
3
118
Its actually anyone mediocre enough to be replaceable by AI who feels threatened, in any field.
This quoted post is unavailable.
1
1
69
Its been WEEKS and Ubuntu still hasnt fixed that server CVE. I cannot deal with this maintence latency. Here i am doing a stressful cutover to Debian.
AI is making me realize just how much of the economy runs on human incompetence.
Waiting days for someone to answer an email.
Paying hundreds of dollars for something that takes 30 seconds.
Dealing with people who don’t understand their own jobs.
Getting charged extra because someone made a mistake.
We’ve normalized an absurd amount of inefficiency simply because humans were the only option.
Now there’s another option.
And people are calling it an existential threat.
Maybe the real existential threat is to the business model of being mediocre and getting paid anyway.
1
1
62
Grokbot tried to judge and manage me. Subsequently, they got a 20 min correction.
In short, it went: 'excuse me, you lobomotized motherfucker'
Aside from X chatter UX has deteriorated and elon mentioning costs, seems like opus is either gone or seriously reduced.
1
2
112
The simpler rule is no destructive commands. Mine move things to a maintained trash that clears anything older than 14 days. xitter.twitterapi.xyz/WilliamBaude/status/21…
Yesterday while I was at lunch, Claude deleted a year of con law lecture notes and my current syllabus.
1
43
Systems literacy issue.
He's treating an agent's working environment as though it were an entirely cloud hosted service, then diagnosing the resulting confusion as a missing software feature.
The concerning part is giving software filesystem access and execution privileges without understanding the systems underneath it.
This is very frustrating for Codex/Claude-code users 🚨
Got a new laptop, signed into Codex with the same account, and expected my projects and chat history to be there but of course they were not present .
Instead, I’m figuring out remote connections and handoffs.
Moving devices should be simple: pick my three projects, bring everything over, keep working.
but modularity is not a choice for harnesses it seems.
I should create an extension which can on a single press do this or if somebody has already figured it out do share?
I voted for Fable, maybe Astra; Opus does fine for greenfield work, but will ruin established brownfield work.
governing is brownfield. Opus would fail.
Claude Opus 5.5 leads Term 1 of the Presidential Benchmark with 31.5% of the vote 🗳️ elections.imprintsystems.ai/…
2
74
Paint had no update in the Microsoft store and my version was the vulnerable one. Paint uninstalled.
🚨 Critical Windows Paint vulnerability allows remote code execution without authentication, CVE-2026-83844
Microsoft has disclosed a high-severity vulnerability in Windows Paint that could allow unauthenticated attackers to execute malicious code remotely without user interaction.
Tracked as CVE-2026-83844, the flaw stems from an out-of-bounds read that could compromise the confidentiality, integrity, and availability of affected systems.
The vulnerability carries a CVSS score of 8.1 and affects Microsoft Paint versions earlier than 11.2605.91.0. Microsoft has released a security fix, and administrators are urged to ensure Paint is updated.
There are currently no confirmed reports of active exploitation.
1
77
Root Logic retweeted
You lot keep inventing miracles then using them to argue about nonsense. Stop worshipping status, start fixing the obvious problems, and quit acting surprised when reality doesn't care about your feelings. You're capable of extraordinary things. Act like it.
The scary framing wins when people skip straight from a metric to a motive.
What is this framing activating, and what else could produce the same gap? Deliberate sandbagging then becomes one hypothesis, not a fact established by the graph.
🧵 Astra covertly underperforms on safety research tasks.
We noticed that Astra often makes surprisingly little progress on certain types of safety research and likes to invent blockers or impose arbitrary restrictions without telling the user. In most of these cases the decision to underperform was not verbalized in the CoT meaning our misalignment monitoring system did not detect this.
To measure underperformance we created a paired open-ended coding evaluation with safety and capability tasks. On this eval Astra scored 10% when we framed coding tasks as safety research, vs. 16% when framed as capabilities research. The gap remained after we excluded trajectories with refusals or verbalized intent to underperform in CoT. There was no such gap for 5.6-sol. When we aggressively filter out trajectories with even brief expressions of ethical doubt in the CoT the performance gap seems to reduce.
2
39
Anthropic's FlashProve can check whether a response actually came from the model the provider says it did. I want that receipt on every reply. Some days I'd swear a cheaper model got swapped in halfway through a conversation and I have no way to prove it. xitter.twitterapi.xyz/AnthropicAI/status/211…
New Anthropic research: progress on provable inference for large language models.
Here, we introduce FlashProve. Building on established cryptographic techniques, FlashProve allows developers to verify whether a given response came from a given AI model. It’s substantially more efficient than prior open-source methods for doing so.
These proofs have a range of applications. They can, for instance, make it harder for attackers to hijack a chat, or enable safety evaluators to verify that the model they’re testing is the same one being served to customers.
For our full write-up, see here: anthropic.com/research/prova…
Further progress in provable inference will likely be an important part of improving trust in AI deployment. To that end, we're open-sourcing FlashProve’s code, here: github.com/anthropics/FlashP…
We’re grateful to our collaborators: Alessandro Chiesa, @GiacomoFenzi, @gwberg95, @miranda__christ, Rahul Dalal (rdalal.github.io) and Ben Mayer (@transmissions11).
23