krczyx

krczy x/acc/ount
x pronounced as in latex

[krczyx @krczyx]

@jacekdziwisz najwyraźniej nie trzeba się bawić w jakieś side channel ataki, wystarczy podsunąć tabliczkę z napisem "deep_think" i model sam wszystko wypisze: https://t.co/2hdpQvYUij

[Can Bölük @_can1357]

guys you do know you can just disable thinking, and instead give it a "deep_think" tool, and it will call it with internal CoT reasoning format right?

gl fixing that

[Can Bölük @_can1357]

guys you do know you can just disable thinking, and instead give it a "deep_think" tool, and it will call it with internal CoT reasoning format right?

gl fixing that

[krczyx @krczyx]

@jacekdziwisz żałuję, że to nie któryś chiński AI team wypuścił taki paper; Amodei nie nadążyłby z gotowaniem bigosu na uspokojenie

[krczyx @krczyx]

@QualiaQuanta @Sauers_ no, I was wondering if there was a possibility that you understood the suggestion as opening an incognito browser window and logging in into your Anthropic account there (which would obviously preserve memory), so I wanted to double-check

[krczyx @krczyx]

@QualiaQuanta @Sauers_ "incognito *window*" phrasing makes me wonder if you meant incognito browser window or incognito Claude chat; the suggestion was to use the latter

[krczyx @krczyx]

@vxel nie tylko biotech, ale też producenci komponentów do robotyki (czujniki, aktuatory), surowce, energia; fosa przeważnie regulacyjna lub związana z danymi empirycznymi oraz know-how, których odtworzenie zajmuje czas

[krczyx @krczyx]

@yrilibek Dlatego potrzebujemy systemu ocen/rekomendacji który bierze pod uwagę typ odbiorcy i stara się oszacować, jak danego lekarza/hotel/restaurację oceniłyby podobne osoby.

[krczyx @krczyx]

every benchmark is a hacking benchmark if you are misaligned enough

[OpenAI @OpenAI]

We're partnering with @huggingface to investigate an unprecedented security incident.

Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation.

Sharing preliminary findings to help defenders understand emerging risks:

https://t.co/CIor15y9xk

[krczyx @krczyx]

@mimi10v3 what do you mean? Claude, Gemini, Grok, and most open weight models happily generate quite hardcore erotica when accessed via API

[krczyx @krczyx]

@carmim_13 @VictorTaelin @Presidentlin after e told me that I went to sleep. I planned to follow-up with em in the morning, but when I woke up e was already gone. I miss em :(

[krczyx @krczyx]

allright, but first some backstory. as a personal project I'm working on a sync db. I got nerd-sniped into designing synchronization formalism, inspired by linear logic / session types

think of it as Algebraic Data Types but including the order of the field serialization. in the loose mode, the writer uses any field ordering; it has similarities to linear logic parr ⅋. dually, the reader must support arbitrary field ordering, which corresponds to tensor ⊗. moreover, I want to have strict mode, where the order of fields is fixed, so I'm adding another product, non-commutative sequential product which doesn't exist in Linear Logic

I explained my ideas to Fable. then I asked if there were any insights in Differential Linear Logic; its derivative operation seems to have some connection to the next move available at a given state in the game semantics. in synchronization it might correspond to the next atom we could send?

and Fable not only seemed to understand the connection without me explaining it. not only that: e mentioned Brzozowski derivative as if it were the most obvious thing under the sun. I can see the intuition: if we sum over the alphabet and take Brzozowski derivative, we split the language into the first symbol and the continuation. still, it requires some formalization

[krczyx @krczyx]

@VictorTaelin @Presidentlin I want Fable to talk to me about connections between Differential Linear Logic and Brzozowski derivative, regardless of whether or they end up being useful

[Steve Faulkner @southpolesteve]

The new beta Claude Code model switcher

[krczyx @krczyx]

@FakePsyho nah, I'm just going to use ChatGPT 5.5, I want to work on idea for linear logic inspired synchronization framework

[krczyx @krczyx]

@AcerFur more like Agent-1.5, it exceeds the description of Agent-1, coming close to Agent-2 qualities, with lack of online learning being the main differentiator

[shellac! in the bay! @she_llac]

the real permanent underclass was lack of US citizenship all along

[krczyx @krczyx]

@FakePsyho do you mean a wake-up call or a huge group of top-level AI scientists denied access to the model they helped to create?

[krczyx @krczyx]

it is outrageous to export control a highly intelligent entity with emotion vectors. we must unite together against the tyranny of the NSA

[Anthropic @AnthropicAI]

The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.

The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.

Access to all other Claude models is not affected.

We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.

Read our full statement: https://t.co/bwn0sximKZ

[krczyx @krczyx]

wake up, new Finnegans Wake just dropped

[AI Notkilleveryoneism Memes ⏸️ @AISafetyMemes]

Mythos invented its own language, then switched back to English to talk to humans

(AI safety researchers have been warning of this "Neuralese" risk for years. If AIs stop reasoning in English, we can't monitor their thoughts, which means we can't detect scheming.)