bvy.

beep-boop developer

[bvy. @bvylabs]

@denysdovhan And node.js critical security releases "resolving multiple vulnerabilities, including HIGH-severity flaws that allow for remote code execution and security restriction bypasses"
🥱

[bvy. @bvylabs]

@wasmerio

[bvy. @bvylabs]

https://t.co/br4mc81Gkr

Там можна пре-заповнити дані, щоб швидше на НП пройти процес і впевнитися що все коректно.

Ну і хоча б якось можна це зробити з цими усіма тривогами. Маю на увазі у цнап там же талон-черга і з шахедами воно не дуже реально

Але маленька ще деталь на яку можна натрапити: КІТ-номер - в них в програмі чітко пробиті перші літери «КІТ», але зараз старлінки до мене, наприклад, прийшли з номерами, що починаються на «KFV». В такому випадку кажемо, щоб «KFV» не вписували, а одразу в себе в програмі після «КІТ» писали номер якій у нас після «KFV».

[bvy. @bvylabs]

@KausArtem @RobertZeltinsh А якщо не при отримані, то кількості відділень НП в рази більше ніж цнапів і можна пре-заповнити все вдома, ні?

[bvy. @bvylabs]

@KausArtem @RobertZeltinsh а навіщо в цнап, якщо можна відразу в НП при отримані?

[bvy. @bvylabs]

@forgebitz @ImSh4yy kinda like computer-use from codex/cc apps🧐

[bvy. @bvylabs]

"we have fixed opus 5.5 writing" = 3 level nested list ¯\_(ツ)_/¯

[bvy. @bvylabs]

Peak Claude Code experience 🫴

[bvy. @bvylabs]

and grok-build cli is nice as well

[bvy. @bvylabs]

It sounds wild, but I enjoy grok-4.6 and Heavy subscription.

After fable, slopus and gpt-sol it feels like the one model that can get shit done.

It understands me, answers as I want, and makes the decisions I would. No endless circling - just gets things done properly.

[bvy. @bvylabs]

Using Anthropic models = fighting their bullshit classifier 50% of the working time.

I'm so tired of it.

[bvy. @bvylabs]

@thsottiaux luna as subagents for sol/terra

[bvy. @bvylabs]

I don't know the reason, but Codex/CC feels so much worse than it did a year ago. You ask it to implement something, and it runs for like 2 hours without making any real progress. You start being more specific, literally tell it to write the code, but it still keeps going in circles?

[bvy. @bvylabs]

@thsottiaux finally fix v1/v2 backend, so it will be possible to spawn Luna subagents with Terra/Sol

[bvy. @bvylabs]

@thsottiaux brother, anyway to not make it braindead along the way?

gpt-5.6-sol high is unusable now, it feels like opus or even worse

[bvy. @bvylabs]

Why is Claude Code’s goal input still limited to 4k chrs?

It’s super annoying having to save everything to a file every time instead of just dumping it directly into the input.

[bvy. @bvylabs]

I'm tired of Opus 5 giving me its opinion instead of doing the work.

Jeez, just start working. I don't care what you think about the task and that you don't like it - just do what I asked.

[bvy. @bvylabs]

I'm starting to get flashbacks with codex subscription 🙄

[bvy. @bvylabs]

gpt-5.6-sol is absolutely unusable for the second day. I do not know what they changed, but the model has become brain-dead.

[bvy. @bvylabs]

@denysdovhan 1k+ токенів щоб текст підправити🥲

Рекомендую "use DS-STE100 Simplified Technical English" і все.

Можна ще 1-4 речення в якому порядку повинна бути інформація, що потрібно фокусувати на діях, або що потрібно відсіювати зайву.