omg TIL you can actually fork conversations in @opencode!
hit ctrl-x g to open the message selector, then when pick an earlier messsage you get an option to fork the convo
@iannuttall spot on. codex does pretty well with background tasks if you make it run them in the bg using tmux @steipete style https://t.co/wHEQGFjMyP, but yea would be great if they supported them natively.
wdyt of sst/opencode? i've found it a lot more enjoyable than factory
Today, I let Claude Haiku run an army of codex subagents to fix a myriad of new linter issues that popped up after adopting next/eslint-plugin-next.
All you need is tmux and some creativity.
one overlooked downside of hosting things on heztner boxes: their ip ranges get blocked all the time because lots of people use hetzner boxes for nefarious activities
My take with codex vs claude code
- Use codex if you have a detailed plan and want to walk away for 30 min. If you try and interrupt and iterate codex, you might as well restart from scratch
- Use claude code if youre not 100% sure where you're going and want to iterate
@MaximeRivest it does! but dangerous, that's just begging to get lethal trifecta-ed (h/t @simonw)
can try this to keep that smart and fast feeling but stay a bit safer https://t.co/ViQt3bOl1k
just paid a crying woman $500 for her mental breakdown video
it's about to do $300k in sales
here's why raw ugliness beats polished influencers:
we used to pay influencers $10k
for perfect videos
that got 1.5x ROAS
now we search tiktok for emotional breakdowns
about our problem
and license them for $500
the process:
search: "[your niche] ruined my life"
find: people genuinely suffering
license: their existing rants
add: "there's a better way" + product
it's 10x more believable
because it's real pain
examples printing:
hair loss guy having breakdown in car
licensed for $400
added our solution at end
$180k in sales
mom crying about stretchmarks
licensed for $600
added our cream at end
$240k in sales
dad furious about sleepless nights
licensed for $500
added our sleep solution
$320k in sales
the difference:
influencer video feels like ad
breakdown video feels like documentary
one is selling
other is sharing
tested across campaigns:
polished influencer content:
- 0.8% CTR
- 1.2% conversion
- 1.5x ROAS
raw emotional content:
- 4.2% CTR
- 3.8% conversion
- 5.7x ROAS
finding the gold:
search terms that work:
- "[problem] ruined my life"
- "crying about [problem]"
- "rant about [problem]"
- "[problem] breakdown"
look for:
- genuine emotion
- specific pain points
- relatable person
- decent video quality
avoid:
- obvious actors
- multiple takes
- perfect lighting
- any polish
the licensing process:
DM: "your video perfectly captures what our customers feel. can we license it for our awareness campaign? $500 for 90 days"
most say yes immediately
they made it for free
$500 is pure profit
then we add:
- simple intro slate
- "there's a better way" transition
- product demonstration
- clear CTA
the magic:
viewer thinks: "that's exactly how i feel"
then sees solution immediately
while emotions are high
it's not manipulation
it's meeting them where they are
you're manufacturing fake authenticity
i'm amplifying real pain
guess which one converts
stop paying for perfect
start paying for pain
the rawer the better
the uglier the realer
the realer the money
dm me "ads" and i'll setup your ads account using the same method that generated 10m+ in rev for me
@0xKiryoko@donvito not easily but you can set it up using a proxy tool, check this out. i was able to use it through the proxy tool with my claude max plan. glm took more work, had to fork the proxy server to suppport zai but can share if you're interested
if anthropic wants to get back in the race they need to go full gamification mode on claude code. if i got badges like "clauding at 2am" and "hit 5 hour usage limit twice in 1 day" every time i did i'd share them everywhere.
i'd even buy merch
@thsottiaux ah gotcha thanks, i missed that distinction. so the tldr for getting the most out of the model is short system prompts, but provide as much context in user messages as possible?
@thsottiaux now i'm confused, the openai gpt-5-codex cookbook consistently suggests taking a "less is more" approach to prompting codex which seems to contradict this. is there a goldilocks style just right amount of context to give it? how do you know out what the right amount is?