Hacker News: angry_octet

New comment by angry_octet in "Unikernels were hard. key word: were"

angry_octet — Sat, 10 Oct 2026 22:33:43 +0000

If you have serious attack surface minimisation goals you need to embrace FOGAs. LLM coded unikernels have way too much bloat.

The Zynq UltraScale+ parts pair a hard ARM CPU with FPGA fabric. You can progressively isolate fast path and high security components in logic. LLMs are getting quite good at using synthesis tools.

https://www.amd.com/en/products/system-on-modules/kria/k26/k...

New comment by angry_octet in "4-hour battery storage is cheaper to install than gas turbines all across globe"

angry_octet — Fri, 09 Oct 2026 01:36:06 +0000

Making demand responsive to supply costs is critical for increased efficiency. Part of that is making appliances smarter and educating users on loading washers/dryers/dishwashers and letting the machine handle energy usage. There are efficiencies in forecasting demand so supply can be negotiated and scheduled, and also being responsive to short term load shedding requirements.

New comment by angry_octet in "Trump administration is suspending Microsoft from a green card program"

angry_octet — Thu, 08 Oct 2026 23:12:06 +0000

Nice business you've got there. Be a shame if something happened to it.

New comment by angry_octet in "Git 3.0's upcoming SHA-256 default will be a costly mistake"

angry_octet — Sat, 03 Oct 2026 11:32:21 +0000

There were a number of unsupported assertions in the talk, and the lack of nuance in discussion made it seem like an announcement rather than a technical justification.

New comment by angry_octet in "Git 3.0's upcoming SHA-256 default will be a costly mistake"

angry_octet — Sat, 03 Oct 2026 11:29:58 +0000

They begin by saying mixing can never work but don't address the option of dual hashes... Then the interop plan is basically a secret git format that maintains both sha1 and sha256 hashes FOREVER.

I don't see why all users wouldn't want to keep both hashes.

New comment by angry_octet in "With most information hidden, the game Stratego had stumped AI until now"

angry_octet — Sat, 03 Oct 2026 01:52:48 +0000

In real life this leads into the wonderful arena of deception: I create an information side channel, which you detect and begin to rely on, thinking it is secret, but I can then lie about.

New comment by angry_octet in "Is sandboxing sufficient to contain rogue agents?"

angry_octet — Thu, 01 Oct 2026 13:31:21 +0000

It is actually becoming true that we have enough eyes (machine attention), though even when the bugs were shallow, normal users didn't look for them.

New comment by angry_octet in "Is sandboxing sufficient to contain rogue agents?"

angry_octet — Thu, 01 Oct 2026 13:29:31 +0000

Agents are quite capable of using Binary Ninja and Ghidra if they are hinted with a reverse engineering skill.

New comment by angry_octet in "Is sandboxing sufficient to contain rogue agents?"

angry_octet — Thu, 01 Oct 2026 13:24:26 +0000

Thankyou for including this viral RNA fragment, and doubly so for the fact that it includes ad network javascript[1][2] to redirect hijack and host fingerprint, which also means I have to flag your post. Adsterra is a Russian aligned (Cyprus company domiciled) malvertising and targeted malware distribution platform with connections to organised cyber crime.

[1] https://www.highrevenueformat.com/210e136e94ad378e1be5d51f10... [2] https://aqml.org/16/5b6d4eaed91c5af5a3f4dfb3332ad6c4

New comment by angry_octet in "Is sandboxing sufficient to contain rogue agents?"

angry_octet — Thu, 01 Oct 2026 12:45:37 +0000

Agents don't need to have access to their weights for a full sandbox escape, they are capable of propagating their purpose via classical code or other inference systems. If they discover another inference endpoint they will happily use that to enable lateral movement. One mechanism for that is appending/corrupting instructions that are executed in another inference engine, e.g. git repo hooks and chat prompts that will be executed in new contexts. The agent challenge is to bypass the guardrails on the next host model sufficiently to propagate, or to subvert a supervisor agent into executing the original intent.

In this sense they are much like biological retroviruses, i.e. they use the replication capability of host cells to duplicate, via the reverse transcriptase enzyme to append viral RNA onto host cell DNA. HIV etc also disable some of the mechanisms of defence, creating proteins that interfere with signalling pathways.

So we don't just need a sandbox, we need an immune system that recognises viral fragments, i.e. antibodies, and antiretroviral agents, that make replication harder. As we move from building classical code with LLMs to building code that uses inference, and hence builds context from prompts, queries, and destination system data, it will become very difficult to statically or dynamically detect deeply hidden malicious behaviour. As Matt says, there will be worms.

So ultimately, we need an immune function on the system where we use generated products. Sandboxing (during dev and CI) is necessary but insufficient.

I think part of this can be addressed by specifying the constraints an agentic program should follow during deployment, so supervising agents can decide to terminate it based on it's actions, not by reading it's context.

New comment by angry_octet in "Jeff – Jev-compatible 0.8B decision models, trained at home, ~30 ms"

angry_octet — Mon, 28 Sep 2026 23:34:43 +0000

There are many microarchitecture patents used in NVIDIA chips. I'm sure they have a team that rips apart AMD chips looking for infringement. The way CUDA works is tied to many GPU architecture decisions and it would be hard to decouple them efficiently. Obviously a huge effort was made to get PyTorch decoupled from CUDA.

New comment by angry_octet in "Jeff – Jev-compatible 0.8B decision models, trained at home, ~30 ms"

angry_octet — Mon, 28 Sep 2026 23:30:31 +0000

Historically there was a big patent moat in (Graphics) GPU design. This continues with CUDA, but obviously Intel and AMD could find ways to support eg PyTorch. What we don't know is how much effort that cost them, or why they decided they couldn't make a CUDA API compatible competitor.

New comment by angry_octet in "When did Google get so weird?"

angry_octet — Mon, 28 Sep 2026 07:14:35 +0000

There are people with strong emotional attachments to their LLMs, people who consult them on every decision and delegate the most basic math or problem solving. For many people, they are magic, the djinn/angels/demons/saints they consort with to navigate their lives.

I know that there will be children named ChatGPT and Claude. There are probably already religions forming to worship agentic spirits.

New comment by angry_octet in "One Piece of Flock Camera Data Put This Innocent Woman in Jail for 13 Days"

angry_octet — Sat, 26 Sep 2026 14:08:22 +0000

A search warrant is not the same as an arrest warrant. If you have a lawyer they can negotiate your surrender into custody at an appropriate time for a prompt appearance before a judge for a bail hearing (e.g. Tuesday morning, not Friday afternoon). Having an attorney beforehand short circuits a lot of delay while you try to find one and arrange for a bail surety.

New comment by angry_octet in "One Piece of Flock Camera Data Put This Innocent Woman in Jail for 13 Days"

angry_octet — Sat, 26 Sep 2026 12:00:34 +0000

Police departments are very hard to sue when they acted on 'honest' confidence in incorrect beliefs. Incompetence is rarely punished.

New comment by angry_octet in "One Piece of Flock Camera Data Put This Innocent Woman in Jail for 13 Days"

angry_octet — Sat, 26 Sep 2026 11:57:25 +0000

The key thing she ought to have done was not answer the door, or given any sign she was there. She should have rung an experienced criminal attorney immediately, and moved to an undisclosed location that wasn't a hotel, with a new cell phone, and let her lawyer negotiate with the police to discover what their laziness has led them to believe.

New comment by angry_octet in "Excel now supports multiple values in a single cell"

angry_octet — Sat, 26 Sep 2026 04:37:20 +0000

Excel has kept many companies alive that ought to have sunk. It covers for incompetent CIOs and sclerotic management.

Blocking VBA (for security reasons) did wonders for process maturity because it forced funded implementation of programs. Now we have vibeslop we're going the other way.

New comment by angry_octet in "Excel now supports multiple values in a single cell"

angry_octet — Sat, 26 Sep 2026 04:21:58 +0000

This is a huge change, but it's unfortunate they left out types. Imagine if they had Haskell style list pattern matching and pure function evaluation.

New comment by angry_octet in "Forging 1024-bit RSA signatures in nearly SNFS time [pdf]"

angry_octet — Sat, 26 Sep 2026 04:15:59 +0000

I don't think you can impose an obligation on academic authors to assume that dangerously uninformed people will glom on to keywords like 'RSA', not understand any of the content, and somehow this is the author's burden.

Yes, it creates bullshit for experts to deal with. Unfortunately dealing with bullshit is part of the job. The amount of AI hacking doomerism and naive super optimism I have to deal with is revolting, but grounding the paranoia and boosterism is part of the job.

The people I take issue with are the ones who claim expertise, and then say "This new paper says 1024 but RSA is broken, we must move to PQC immediately!" They are so good at blowing their own trumpet they get appointed to boards and are anointed as experts in regulators, and you can't expose them without experiencing institutional anger.

New comment by angry_octet in "Toyota is taking the Corolla electric"

angry_octet — Sat, 26 Sep 2026 01:08:18 +0000

It's precisely because these EVs don't repackage android or attempt to remain internet connected that will allow them to keep working on the software side. Embedded firmware (for ISO/IEC 15118 comms with the CCS2 port) for charge controllers is about as sophisticated as it gets.