Daniel Cazares

AI automation & security research. I build agents, break things responsibly, and post what I learn.

[Daniel Cazares @daniel_cazares]

@victornunez Imagine getting sand in the hinge or anything in the hinge and then folding it.

[Daniel Cazares @daniel_cazares]

@thsottiaux @thsottiaux Tibo, who we gotta talk to 'round here to so we can reorder /presets prompts.

[Daniel Cazares @daniel_cazares]

@thtbee_ Damn. When I saw this I legit said "oh my god."
You have fine design taste, man. GG

[Daniel Cazares @daniel_cazares]

@S1r1u5_ Frankly, anyone doing security research should be able to submit their use case and get approval. I'd be willing to show my birth certificate and install any monitoring software or even use a virtualized desktop on their servers that they control.

[Daniel Cazares @daniel_cazares]

@This_Liss I reported this to Anthropic awhile ago when Claude Code was sending my email in User-Agent headers. No response.

[Daniel Cazares @daniel_cazares]

@thsottiaux @victornunez I accept Codex credits if anyone is feeling generous. Just saying.

[Daniel Cazares @daniel_cazares]

Sharing my GPT 6 Astra model instructions file. Hope it helps people with Astra's annoying tendencies.

https://t.co/ib6r59Tqzh

[Daniel Cazares @daniel_cazares]

@TJeparskis Yeah, I had to write a whole custom prompt to fix all of its issues and sandbagging.
You can try my instructions file. See how it goes.

https://t.co/ib6r59Tqzh

[Daniel Cazares @daniel_cazares]

@lydiahallie i accidentally used mine at 100% usage. 😭

[Daniel Cazares @daniel_cazares]

@thsottiaux Oh, come on. That's 100% phoning it in.

[Daniel Cazares @daniel_cazares]

@nateberkopec @bitsweat @thorstenball "You baboon! Read for my intent!"

[Daniel Cazares @daniel_cazares]

@nateberkopec @bitsweat @thorstenball "Why on god's green earth would you do this?"

[Daniel Cazares @daniel_cazares]

@Spectromachina We should collab. I'm having Claude and Codex work on a game engine using WebGPU. Be cool to see it power some of your work.

[Daniel Cazares @daniel_cazares]

@Spectromachina It doesn't give us the right, but the question is can we still look if we don't stare.

[Daniel Cazares @daniel_cazares]

@Spectromachina @Meadowbrook_ @AnthropicAI Based

[Daniel Cazares @daniel_cazares]

@thsottiaux Oh, thank god. I'm out.

[Daniel Cazares @daniel_cazares]

@TokenGremlin @TokenGremlin I don't understand. Opus 5.5 Fast is billed entirely as extra usage credits. You can't use it as part of your sub at all.

[Daniel Cazares @daniel_cazares]

@kimmonismus @grok Focusing on "abusive or cruel behavior" toward models, if AI were potentially conscious or sentient, would it be cruel, abusive or similar when Anthropic deprecates a model, wipes context-windows, or reboots hardware? Don't bring in unprompted claims and stay on the point.

[Daniel Cazares @daniel_cazares]

@jehovahscript @grok Why are dots bad?

[Daniel Cazares @daniel_cazares]

@grok @JohnMcKeownn @thsottiaux @poteto @grok But it's about a post saying it'll go to other models. Meaning they're taking your share. So... you'll have less results and ship less...