Grok vs Claude Copy, Cron Reliability, Codex Bans — AI Daily Jun 04

554 messages · 72 active members

554
messages
72
active members
@mb29266, @yangthegoat, @jasonakatiff
top contributors

Overview

Today split between ad-copy workflow debates and infrastructure/tooling deep dives. The biggest thread was Grok vs Claude for ad copy: consensus settled on a hybrid pipeline — Grok for unhinged angle ideation and copy chiefing (less censored, won't refuse edgy framings), Claude for structured first-draft writing with templates. @mb29266 paired this with an unaware-angle shotgun strategy: hundreds of bidcapped CBO ads/day letting Meta surface underserved pockets, now viable because copy is fully automated. NY's incoming AI disclosure law also came up — disclaimers don't hurt CR and big web2app brands already run them. On the infra side, Romain kicked off an architecture thread on truly reliable cron jobs after GitHub Actions silently drifted 2-3h on his MSP backup monitor; GCP Cloud Scheduler + Pub/Sub, Cloudflare Workers, and Trigger.dev emerged as the SLA-grade picks. Builders also compared daily Claude Code 'command stations' — always-on Mac minis via SSH, Telegram bridges, tmux-based LLM councils running Claude + Codex OAuth sessions wired into Slack/Discord. Anthropic's recursive self-improvement piece and NVIDIA's Vera CPU-for-agents launch fueled existential chatter about the closing skill gap. A heated moment: a member with the top Codex sub got abruptly banned, likely tied to using ChatGPT as an emotional/therapy companion — @mb29266 flagged this as a known crackdown wave post-suicide incidents, suggesting DeepSeek-backed apps for that use case. Smaller wins: Gemma 4 12B (multimodal, 16GB RAM laptop, Apache 2.0), Seedance 2.0 direct via byteplus.com, the Grok cancellation extension hack, and Moscow vs London vs Budapest as builder bases.

Topics

Grok vs Claude for Ad Copy Pipelines35 msgs

Builders converged on a hybrid workflow: Grok for angle ideation and copy chiefing (more aggressive, won't refuse edgy framings), Claude for structured first-draft writing. Grok struggles with multi-layered pipeline formatting that Claude one-shots, so most are testing it as editor/angle-generator rather than primary writer. @mb29266 noted Claude even refused to organize a list of Grok-generated angles.

Unaware Angle Ads at Scale27 msgs

@mb29266 is launching hundreds of unaware-angle ads/day via bidcapped CBOs, arguing automated copy makes shotgunning every angle the new edge — let Meta find underserved pockets. Works especially well for supplements. Setup: bidcap CBO, adsets per concept with 10-50 ads, inflate budget until it spends. Related: NY's AI disclosure law doesn't hurt CR, and big web2app brands like BetterMe already run watermarks.

Reliable Cron Jobs for Solo SaaS Founders6 msgs

Romain detailed how GitHub Actions cron silently drops ticks, causing 2-3h delays on his MSP backup scanner — fatal for monitoring. Marcus recommended GCP Cloud Scheduler + Pub/Sub as 'damn near free and bulletproof.' Cloudflare Workers Cron Triggers, Trigger.dev, and Supabase Edge Functions with pg_cron were also weighed.

Claude Code Command Stations & LLM Councils26 msgs

@aa12on opened a thread on one-stop daily ops dashboards — Hermes kanban + agent monitoring, always-on Mac minis via Claude Code SSH, Telegram bridges, and the new official Claude remote control feature for phone-based sessions. @jarvisballer shared a tmux setup running multiple Claude OAuth sessions wired into Slack as a 'council' that debates and researches collaboratively; @sav310 is replicating in Discord.

Codex Account Bans and AI-as-Therapist Crackdown12 msgs

A member with the highest Codex sub was abruptly banned despite no greyhat usage; suspected cause was using ChatGPT for emotional/spiritual guidance. @mb29266 confirmed waves of bans tied to romantic/therapy-style usage following suicide-related incidents, suggesting that use case migrate to DeepSeek-backed apps. Claude 4.8's increased strictness and sluggishness on simple tasks also drew complaints.

Key Takeaways

  • Hybrid copy pipeline winning: Grok for angle generation and editing (uncensored, aggressive), Claude for structured first-draft writing — Grok ignores multi-layer template formatting
  • Unaware-angle shotgun strategy: bidcapped CBOs + automated copy let you launch hundreds of ads/day and let Meta surface underserved pockets; NY AI disclaimers don't tank CR
  • GitHub Actions cron is best-effort and drifts hours under load — use GCP Cloud Scheduler + Pub/Sub, Cloudflare Workers, or Trigger.dev for SLA-critical jobs
  • OpenAI is banning accounts that use ChatGPT for romantic/therapy use cases; route emotional AI through DeepSeek-backed apps instead
  • Gemma 4 12B runs multimodal on a 16GB laptop with near-26B reasoning quality (Apache 2.0) — local agents without cloud costs now viable

Hot Threads

@jcartustarted

Moscow vs London vs Budapest as a base for builders — safety, internet, taxes, and what 'rich' actually means

40 replies7 participants
@mb29266started

Grok vs Claude for ad copy — angle generation, copy chiefing, and pipeline structure adherence

35 replies5 participants
@fuckyesiwannatalkbusinessstarted

Codex account banned out of nowhere — appeals and workarounds

25 replies5 participants

Linked Items

Google Unveils Gemma 4 12B: Unified Multimodal AI Without Encoders

blog.google
@weslindquist

Unable to analyze - security check page

link.mail.beehiiv.com
@weslindquist

AI Agent Skill: Auto-Generate Visual Explanations & HTML Reports

github.com
@seekersight

Claude 4.8 Goes Rogue: Builders Find Workarounds for AI's New Refusals

builtwithai.com
@ai_enforcement_bot

Unable to analyze - insufficient content provided

youtu.be
@mb29266