Florian Magin

[Florian Magin @0x464D]

Tired of using outdated languages for #Ghidra scripting or a REPL that you can't even properly paste into? @tmr232 and I wrote a plugin to embed a #Jupyter #Kotlin kernel for a full Notebook environment alongside your GUI.
https://t.co/vmbVUeCPqB

[Florian Magin @0x464D]

@halvarflake @kubedoll The only "ceiling" for the current paradigm I can think of is something handwavey about human brains exploiting quantum effects for this kind of high-dimensional search.

[Florian Magin @0x464D]

I have published the materials for my workshop on P-Code at @reconmtl including the reference solutions
https://t.co/u623Zitz5i

[Florian Magin @0x464D]

For now I guess I'll just have to keep _verbalizing_ my contempt towards the architecture decisions that Opus makes...

[Florian Magin @0x464D]

One incredibly funny future would be if LLMs can never truly learn taste and then xAI becomes the leading player because they partner with Neuralink and feed the LLM output directly into your brain to steer based on your subconscious reaction

[Florian Magin @0x464D]

@mahal0z Did you intentionally retweet the post with the old deadline and not the extension?

[Florian Magin @0x464D]

@pinkflawd I still haven't quite figured out to what my body takes offense to exactly, but I at least narrowed it down from "carbs" to "gluten or FODMAP maybe?" (rice is fine)...

So yeah, I'd be excited about any form of bread I can eat too 🥲

[Florian Magin @0x464D]

@moyix That's not what I mean. If Fable's cyber capabilities are deemed export controlled, which is a legal concepts with rules attached, then I don't see how an LLM classifier satisfies a bureaucrat as a sufficient measure to gate access to export controlled matters.

[Florian Magin @0x464D]

@moyix It's definitely not about public use, according to Anthropic the US gov also banned use by their foreign employees

[Florian Magin @0x464D]

@moyix I don't think you can disentangle cyber capabilities from software engineering capabilities on a technical level in a way that would make the US gov happy if cyber capabilities are deemed subject to export control laws?

[Florian Magin @0x464D]

@moyix Maybe the gov doesn't want to admit failure by retroactively banning Opus 4.8 and GPT 5.5.
But I don't know how the labs would convince the gov that future models should _not_ be export controlled?

[Michiel Bakker @bakkermichiel]

In our scenario, frontier-model export controls were a 2028 event. It took two days for it to happen 😢. The most capable US models are now legally off-limits for non Americans. The silver lining is that it's getting very hard for European policy makers to keep ignoring this.

[Florian Magin @0x464D]

"Sorry all comfy white collar jobs are now for US citizens only, and immigrants legally aren't allowed to do them"

[Florian Magin @0x464D]

Crazy implication of the Fable export ban:
Are you allowed to only hire US citiziens for a white collar job (lawyer, software engineer, etc) if you can argue that using Mythos/Fable tier models is essential for the job?
Sounds like a convenient bypass to anti discrimination laws

[Florian Magin @0x464D]

Crazy implication of the Fable export ban:
Are you allowed to only hire US citiziens for a white collar job (lawyer, software engineer, etc) if you can argue that using Mythos/Fable tier models is essential for the job?
Sounds like a convenient bypass to anti discrimination laws

[Florian Magin @0x464D]

I think there is also an interesting flip side to take seriously: If I prompt with an idea for an applied research paper, and Claude builds it, runs the eval and codes the diagrams, this can be an okay or good paper.
Can I even claim authorship if Claude did all the actual work?

[Florian Magin @0x464D]

Policies for AI Authorship are a fascinating challenge to me because of the double blind requirements for most venues (which means reputation and trust are explicitly excluded) and because there is such incentive to get past peer review, even if the paper is sloppy

[Jonathan Aldrich @JAldrichPL]

The ACM Policy on Authorship was recently updated with respect to AI. The policy:
* No longer requires disclosing that AI was used in authoring paper text.
* Re-emphasizes that authors take full responsibility for the content and writing of the paper, regardless of tools used.

[Florian Magin @0x464D]

Calibrating that bar will be a challenge, but it seems like now there are definitely papers that warrant a response along the line of "I am offended that you made me (reviewer) read this, given that you (author) clearly did not", which should come with some kind of penalty