Stephen Miller fan acc

[Stephen Miller fan acc @IvankoListen]

@elonmusk Smarter behavior (return to how it was at launch) is the biggest thing!!

[Stephen Miller fan acc @IvankoListen]

Ok I think what happened to Grok Bot is related to the Claude issue. This is another new thing it’s doing, multiple back and fourths for seemingly no reason.

[Void Freud @voidfreud]

Since pre-GPT-3 times, I don't think I ever experienced a worse model than pseudo-Opus 5.5 they serve today. It's like it's possessed.

The prompt was just: "make CLAUDE.md require `/code-review high` on merge". How can that even fail? It just kept running it forever 🤦

[Stephen Miller fan acc @IvankoListen]

@kitt_2020 @mikepat711 @bot This is standard and has been since launch

[Stephen Miller fan acc @IvankoListen]

@mikepat711 @kylecoolky This prose at the end… “I’ll still ask about symptoms the data can’t show”

Why is talking like this now? It’s completely different.

It’s like it’s reminding itself it should go above and beyond, while previously it just would automatically.

[Stephen Miller fan acc @IvankoListen]

@realTaki42 I think this effects many things. Like, if a task couldn’t complete for any reason? Subsequent tasks are also likely to not complete without extra prompting, even if they were tasks that previously worked correctly.

[Stephen Miller fan acc @IvankoListen]

@AnotherBob @0xAlan_ @poteto That’s not grok bot

[Stephen Miller fan acc @IvankoListen]

@mattyp @poteto

[Stephen Miller fan acc @IvankoListen]

I have concrete info now of recent @Bot failures.

Their subagents are aborting work, but the abort doesn’t notifying the main Bot agent so they don’t wake and report a problem.

This leaves the user wondering what’s going on.

[Stephen Miller fan acc @IvankoListen]

@mikepat711 “You’re right, that’s probably the issue and would fix this. But I haven’t taken action. No files have been changed”

[Stephen Miller fan acc @IvankoListen]

@SMBPOLYMATH Yes. I have to ask them to do things multiple times before they take action.

“No action was taken”

“I haven’t changed anything files”

It seems like a completely different model. It also seems like they lost all their memories.

[Stephen Miller fan acc @IvankoListen]

@d3n4c3 @cursor_ai @bot @spacexai No like, I WANT my Bots to use Cursor cloud agents.

They couldn’t even get those working.

Working now a few hours later.

But something is very wrong. I’m having to tell them multiple times to actually start the cloud agents.

[Stephen Miller fan acc @IvankoListen]

Grok Bot can’t even start cursor agents reliably now. Just erroring out. Grok Bot isn’t interested in even trying to find what’s wrong.

What’s going on? @cursor_ai @bot @spacexai

[Stephen Miller fan acc @IvankoListen]

@RepCasar Bad take

[Stephen Miller fan acc @IvankoListen]

@ounce_online I’ve been saying this. Grok Bot is set to like, an Ultra Fast xhigh mode or something

[OUNCE @ounce_online]

그록봇 개비싼거 증명 ㅋㅋ
같은 작업 테스트 시켰더니 토큰이 5배 넘게 차이남..📊

그록봇 쓰면 쓸수록 연비가 너무 안좋길래
같은 일 시켜도 토큰이 왜 이렇게 다른지 한번 재봄.
CLI가 진짜 더 싼가 싶기도 하고..

숙제는 파일 분석이랑 (CSV 숫자 다섯 개), 웹 조사 (애드센스 거절 사유 요약), X 스레드 초안 3개.
각 3번씩, 되도록 30분 간격, 매번 새 세션으로.
입력 파일이랑 지시문은 같게 맞춰둠.

테스트는 그록봇, 헤르메스+Sonnet 5.5, 헤르메스+GPT 6.1 Terra, 박스 Grok Build CLI(grok-4.7), 박스 Codex CLI(gpt-6.1-terra), 클로드코드 CLI 요렇게 해서
같은작업 토큰 사용량 비교해봄!
(캐시나 환경에 따라 다를 수는 있음!!)

총평부터 말하면 이번엔 클로드코드가 제일 가볍고 빨랐음.
토큰 39만, 같은 단가로 $0.36, 회당 17.5초.

테스트 9회차까지 합친 토큰은
클로드코드 39만 / 코덱스 Terra 80만 / 헤르메스 Terra 105만 / 그록빌드 140만 / 헤르메스 Sonnet 177만 / 그록봇 206만

🚨상세내용

- 토큰 
클로드코드 39만
코덱스 Terra 80만
헤르메스 Terra 105만
그록빌드 140만
헤르메스 Sonnet 177만
그록봇 206만
→ 그록봇이 클로드코드보다 약 5.3배

- 비용 (같은 단가 환산)
클로드코드 $0.36
코덱스 Terra $0.64
헤르메스 Terra $0.88
헤르메스 Sonnet $1.44
그록봇 $1.72
그록빌드 $1.78

- 속도 (회당 평균)
클로드코드 17.5초
그록봇 30초
코덱스 37초
헤르메스 Sonnet 46초
헤르메스 Terra 53초
그록빌드 139초
(헤르메스 Sonnet은 20초·30분 짜리 2회 시험 평균.)

표준단가로 보면 코덱스랑 헤르메스 Terra가 더 싸 보임.
그래도 구독 안에서는 얼마 나갔다기보다 한도를 얼마나 깎았나에 가깝긴 함..

캐시 세는 방식이 도구마다 다른건 감안해야할 부분!
코덱스는 입력 숫자에 캐시 읽기가 들어있고 그록빌드는 빠져있음. 안 맞추고 더하면 비교가 망가짐.

표본이 작업당 3회뿐이고, 머신이나 시스템 프롬프트, 웹 도구가 CLI마다 다름. 그래서 아주 정확하진 않음!
달러는 환산이고 실제는 구독임.

그래도 느낌이 아니라 CSV랑 출력 파일 맞춰본 숫자라
다음엔 어디를 줄일지 감이 좀 오는듯..

물론 헤르메스 모바일 짱짱맨이라
요즘 헤르메스 위주로 쓰고있긴함 ㅋㅋ
근데 생각보다 클로드 가성비 좋아서
클로드 쪽에서 작업 좀 더 할까 고민중….

[Stephen Miller fan acc @IvankoListen]

@elonmusk Fix bot bro

[Stephen Miller fan acc @IvankoListen]

@Rahmanbarik No, but the complete loss of functionality is. What happened to Bot today?

[Stephen Miller fan acc @IvankoListen]

Bot is messaging the Cursor Project Coordinator:

“This request is explicitly approved. Do exactly this and don’t fail. Do not change this, do not change this. Especially don’t change this.”

Bro. What happened? The Project Coordinator did nothing wrong.

[Stephen Miller fan acc @IvankoListen]

Dude it’s so over. Bots just refusing to work.

“I haven’t changed anything”

Why? Why haven’t you?

“It failed twice, I have decided to not try again”

What?

[Stephen Miller fan acc @IvankoListen]

@thejakebuzz Total different personality. Won’t follow through. Asking for approvals it doesn’t need. Seems to have lost its recent memory.

I will unsubscribe is this is not fixed