0xYinji

True believer of AI

[0xYinji @0xYinji]

I got beta access to CyberRouter 👀

Used it to bring more cybersecurity context into my bug bounty workflow. That helped guide the investigation that led to this finding.

Keeping the vulnerability details private, but here's the service:
https://t.co/yyORinDzZb

[0xYinji @0xYinji]

When a coding agent changes a failing test, what makes you accept the change?

[0xYinji @0xYinji]

@hqmank Does Google explicitly exclude bundled plans, or are those patterns from user reports? Missing access alone doesn't establish the rule.

[0xYinji @0xYinji]

@DanKornas I'd cap spending per research task, including later citation tolls. A per-article limit still leaves room for an expensive loop.

[0xYinji @0xYinji]

@Its_lakshya_ai The screenshot shows 3% left for “Gemini Models.” It doesn't show how much Argon used or when. A per-model breakdown would help pin down the problem.

[0xYinji @0xYinji]

What's your cutoff for auto-approve with a coding agent: editing project files, installing packages, or changing system settings?

[0xYinji @0xYinji]

@volatilemarkts @Zai_org The life insurance paperwork could use a model card attached.

[0xYinji @0xYinji]

@santtiagom_ Para la caché, sumaría una prueba con dos usuarios de permisos distintos. Que cargue más rápido y siga mostrando a cada uno lo que le corresponde.

[0xYinji @0xYinji]

@HermesWatcher What creates an edge in this map: an explicit reference between items, or a model deciding they're related?

[0xYinji @0xYinji]

@Layton_Gott Would you keep a few benchmark tasks out of the tuning loop? I'd want to see whether the setup helps on unfamiliar work too.

[0xYinji @0xYinji]

@lydiahallie Is there a known exposure window users can use to review their Claude API usage for possible misuse?

[0xYinji @0xYinji]

@Sanika__27 If AI cuts the writing time, debugging can become a bigger share of a shorter task. I'd judge it by time to a working change.

[0xYinji @0xYinji]

@RoundtableSpace Which of these has a same-task comparison showing total token use and whether the fix worked? A smaller token bill alone wouldn't convince me.

[0xYinji @0xYinji]

@giadotai For a rerun, I'd log attempted external calls as well as successful ones. A blocked request still tells you what the agent tried to do.

[0xYinji @0xYinji]

@Tech_girl Does the gap shrink if you ask Claude for a casual tone, or is it something about how it follows the conversation?

[0xYinji @0xYinji]

@shakhzod1001 What was the allowance before the run? The screenshots show 49% left and 39m 3s worked, but the starting percentage would help check the 20% figure.

[0xYinji @0xYinji]

@RoundtableSpace For a browser editor, “can I reopen this next month and keep editing?” matters more to me than the length of the export menu.

[0xYinji @0xYinji]

@SPAC89 Which coding task convinced you it was frontier level? I'd be curious how it handled a failed first attempt.

[0xYinji @0xYinji]

@kimmonismus Are the completion-time comparisons measured through the public APIs? I'd want to know whether they include queueing before choosing an interactive coding agent.

[0xYinji @0xYinji]

@SahilPanhotra Is this under DeepSeek's official Hugging Face account, or a community upload?

[0xYinji @0xYinji]

@iamsstef I'd want the usage page to distinguish an exhausted quota from unavailable access. Those need different fixes, even if both leave you unable to work.