fleyta
@fleyta88 • 2,125 subscribers
ai researcher / dm open
Shorts
Videos
THIS AI CLIP MAKES A 40-PERSON VFX SET LOOK LIKE A 30-SECOND PROMPT what sells it is the sequence, not the monster. you get practical-looking water, blue-screen props, miniatures and crew-style framing before anything impossible happens. 3 different “production” worlds show up in under 40 seconds, but they all feel like they belong to the same fake movie shoot instead of random AI shots stitched together. that opens a much better format than another cinematic clip: build fake BTS around scenes that would normally need stunt teams, creature rigs, miniature sets and days of post-production. with Picsart, i can push the same concept through image, motion and edit while keeping the visual language consistent enough to feel like one production. 1 idea can become 4-6 impossible set pieces, one fake behind-the-scenes reel and multiple cuts for x, reels and tiktok. that is a much stronger use of AI video than just generating one pretty shot.
fleyta5,524,974 views • 18 days ago
![holy sh*t. Grok Bot has a 12x cost trap hiding inside one stupid setting... you can build the smartest agent team on earth and still waste hundreds of dollars waking it up when absolutely nothing changed. here's what happens when you run the same routine on different clocks: → every 5 min: 8,640 runs = ~$397/month → every hour: 720 runs = ~$33/month → same model, same prompt, 12x the estimated cost [the fix is actually pretty simple] instead of waking the entire Grok Bot crew every 5 minutes, give each part one job: → scheduler uses event triggers instead of blindly polling → researcher checks one hash, count or timestamp → chief starts the full workflow only when something changes → risky actions stop at a draft instead of waiting for you at 3 a.m. → every run gets logged so failures don't disappear the crazy part? with a cheap change check and 6 real changes a day, my API-rate estimate drops from $397 to $45 (-89%). you still check every 5 minutes. you just stop paying for full runs that accomplish nothing. and remember: unattended approvals expire after roughly 10 minutes. your bot won't wait until breakfast. these are example API costs, not actual Grok Bot subscription bills. stop optimizing the model when the clock is eating your budget. ↳](https://image.24vids.com/tw-2109580244155781616/amplify_video_thumb/2109580199503265792/img/hmmle5Evi6v0s-xo.jpg)
holy sh*t. Grok Bot has a 12x cost trap hiding inside one stupid setting... you can build the smartest agent team on earth and still waste hundreds of dollars waking it up when absolutely nothing changed. here's what happens when you run the same routine on different clocks: → every 5 min: 8,640 runs = ~$397/month → every hour: 720 runs = ~$33/month → same model, same prompt, 12x the estimated cost [the fix is actually pretty simple] instead of waking the entire Grok Bot crew every 5 minutes, give each part one job: → scheduler uses event triggers instead of blindly polling → researcher checks one hash, count or timestamp → chief starts the full workflow only when something changes → risky actions stop at a draft instead of waiting for you at 3 a.m. → every run gets logged so failures don't disappear the crazy part? with a cheap change check and 6 real changes a day, my API-rate estimate drops from $397 to $45 (-89%). you still check every 5 minutes. you just stop paying for full runs that accomplish nothing. and remember: unattended approvals expire after roughly 10 minutes. your bot won't wait until breakfast. these are example API costs, not actual Grok Bot subscription bills. stop optimizing the model when the clock is eating your budget. ↳
fleyta41,138 views • 2 days ago
this is certified f**king gold Andrej Karpathy just showed a cleaner way to build research agents, and it kills the dumbest loop in AI: search → answer → forget → search again he basically turns memory into a real system: raw sources stay raw the agent reads them once turns the useful parts into linked wiki pages then checks the wiki first every time a new question comes in so the loop changes: > question comes in > wiki gets searched first > gaps send the agent back to source material > the answer gets written back into the wiki > the system leaves the next run smarter than the last one that’s the trick the “memory” is not buried in some giant chat transcript anymore it lives in files the agent can actually work with: > pages it can read > links it can follow > notes it can update > gaps it can spot > stale claims it can fix after enough runs, you stop getting one-off answers you get an actual knowledge base that’s why this feels bigger than just “better RAG” a normal research agent can handle 100 questions and leave behind 100 dead chats Karpathy’s setup can handle the same 100 questions and leave behind a research system that compounds the architecture is stupidly simple: > raw/ = source of truth > wiki/ = maintained knowledge > schema = rules for how the whole thing gets updated search still matters it just stops being the whole product that’s the part I’d steal immediately: don’t make the agent remember harder make it write something worth remembering
fleyta21,324 views • 5 days ago
No more content to load