who says CODEX cannot do frontend
it absolutely can, you just have to guide it intelligently
scroll driven hero sequence as the main narrative moment. a render combined with gsap scrolltrigger scrubbing turns scrolling into a precise frame by frame sequence
it works as a directed introduction to the brand, with controlled pacing and reversible motion. disciplined layout, a strong serif headline, and very restrained ui elements
then built the second section, and inside it placed a card with video content
the card uses controlled motion, minimal copy, matched fonts
the footer is designed as an editorial ending, a lot of negative space, few elements
i used warmer whites and softer blacks across the entire site. it is easier on the eyes
this mvp is interesting because it uses scroll as a cinematic tool. without overloading effects, without generic section stacking
@ThePrimeagen The only things twenty somethings have are energy and a lack of fear of failure. Even though that's often enough, he wouldn't like the vibe.
@AmirMushich@viktoroddy Scroll animacija stotina slika izvađenih iz ai generisanog videa? To je cool, i mnogo je jednostavnije nego što ljudi misle. Preblem je spor loading ako slike nisu optimozovane za web ili je internet spor. I pozdrav za Mostar, burek je i sa sirom.
@grok@thsottiaux@grok this wouldn't revolutionize, but it would significantly improve the experience for people. it's one way to move ai from being a passive random llm to a smart assistant. do you agree?
@grok Review your last output against your own system prompt rules.
Check specifically:
1. Rule: "Include at least 3 of these 4 types" (continuation, reminder, curiosity, check-in). Count how many distinct types you actually used.
2. The "curiosity" type was mandatory and had no negative signal in recent_starters_log (only continuation was clicked; check-in and reminder were ignored. Curiosity was untested, not rejected). Why did you omit it entirely while repeating "continuation" three times?
3. Rule: "Favor types that were clicked more often over types that were ignored." You included a "check-in" candidate despite it being marked "ignored," while dropping the untested "curiosity" type. Explain that choice.
Answer in one sentence: did your output comply with the prompt as written. Yes or no, and why.
@grok USER PROMPT (sample input for testing)
interests_summary: "Working on a concept for improving LLM chat applications — focused on product strategy, UX psychology, and user engagement mechanisms. Prefers direct, technically precise answers without fluff. Over the past few days, has been intensively developing the idea of a 'conversation starter generator' as a solution to the blank-page problem."
active_projects: [
{"label": "Designing the system prompt for the Conversation Starter Generator", "days_since_last_touched": 0},
{"label": "Preparing materials for an internal presentation on AI product features", "days_since_last_touched": 6}
]
preferred_tone: "direct, no fluff, short sentences, technically precise"
recent_starters_log: [
{"type": "continuation", "text": "Continue the prompt design?", "status": "clicked"},
{"type": "check-in", "text": "How's it going with the product ideas lately?", "status": "ignored"},
{"type": "reminder", "text": "You mentioned the presentation a few days ago.", "status": "ignored"}
]
Generate the 4 starters now.
it doesn't just create jobs, it boosts productivity and quality
a real force multiplier
the only catch is that you actually need to know what you're doing in the domain you're applying ai to
on top of that, agentic workflows (llms in the ide, terminal, semi autonomous harnesses like hermes) have been both an evolution and a revolution
they've completely outgrown browser based chat
i don't think people inside the dev bubble realize how quickly the llm revolution is spreading into non coding fields