David Novák

Coder and artist working on not getting replaced by AI, Checkout my github or instagram. All links at

[David Novák @Kaladivo]

@kamcab1 @kalousekm Takže když Babiše jmenoval prezident, znamená to, že rozhoduje tak, jak prezident chce?

Náš systém má mnoho chyb, ta, kterou jste popsal vy ale nedává smysl…

[David Novák @Kaladivo]

@big_macow @Semper_Viventem So should you also label code that was generated by ide (snippets, autocomplete…)?.

It’s not about the code. It’s about the ownership and responsibility. AI does not have either of these…

[BTC Prague @BTCPrague]

People, community, full room. 🔑

[David Novák @Kaladivo]

@theo Do you think this makes open AI also hesitate with the release of new model (if they have any)? Could they be a little bit worried they get the same treatment as Anthropic?

[David Novák @Kaladivo]

@NewOneZ_ @michal_novak_21 You are right! Hopefully they catch up. But working with opus 4.6 is different than working with gpt 5.5. Try it, for developers or agentic use it requires much more handholding

[fejau @fejau_inc]

Crypto industry seeing AI getting rugged by the government

[David Novák @Kaladivo]

@hnizdiljan @zednicek_petr @michal_novak_21 Snad máte pravdu. Uvidíme, jestli se jim podaří za čtyři měsíce dostat na úroveň dnešních modelů a jak velkej ten gap bude potom…

[David Novák @Kaladivo]

@hnizdiljan @zednicek_petr @michal_novak_21 Pls, zdroj. Můžeme se tady bavit donekonečna co má jaké výsledky. Ale zatím vidím jen kimi k 2.6 kterej je pozadu …

[Programmer Humor @PR0GRAMMERHUM0R]

newBenchmarkDropped

[David Novák @Kaladivo]

@jsrailton Isnt the malware just flagging itself this way? If I find a text about some biological weapon I can say that I am dealing with malware right? I don’t even need latest and greatest model for that…

But I agree with you, the “security” guardrails are dumb…

[John Scott-Railton @jsrailton]

NEW: malware developers added nuclear & biological weapons text to to their spyware.

Goal? To trigger LLM safety refusals... so that their spyware wouldn't be analyzed by an AI security scanner.

Cleanest practical example I can think of for why over-indexing on first order safety alignment is risky.

When closed (and open) models ship with aggressive refusals, they will be sprinkled with second-order blindspots that attackers will discover...and exploit.

We are only in the earliest days of attackers leveraging these features, and it wouldn't surprise me if users systems that need to handle complex cybersecurity issues demand that models be less safety-blunted.

In the weeds: @SocketSecurity's post also shows why intention matters in how you design a malware analysis pipeline to avoid prompt manipulation.

H/T to colleagues that shared this with me https://t.co/f3Aj9TYxU4

[David Novák @Kaladivo]

@raffiki_art Not here in Czechia...

[David Novák @Kaladivo]

@boyacaxa @ViliamKlamarcik Hello, should be released by now :)! Sorry for missing that https://t.co/7b0SUh7onh

[David Novák @Kaladivo]

@OpenAI How does this work? Do I have to have original image file , or will the photo / screenshot be enough?

A lot of ai images in spams/scams won’t be possible to access directly…

[David Novák @Kaladivo]

@Everlier @ThePrimeagen There is a command you can run to permanently enable hidden files : `defaults write https://t.co/Q3c3cY6GPa.finder AppleShowAllFiles -bool true && killall Finder`. :) (after killing the finder system will automatically restart it ;))

[David Novák @Kaladivo]

@thsottiaux Opensource it!

[David Novák @Kaladivo]

@theo @ArtShendrik I believe it's important to be able to inspect the tool calls. With time you get a sense on what to look for and skip it. But it gives you a idea of what the agent is doing and weather you need to steer it into a different direction or modify your prompt.