Draven

653 posts

Draven banner
Draven

Draven

@notdrvx

๐—ฐ๐—ฎ๐˜๐—ฐ๐—ต๐—ถ๐—ป๐—ด ๐—ป๐—ฎ๐—ฟ๐—ฟ๐—ฎ๐˜๐—ถ๐˜ƒ๐—ฒ๐˜€ ๐—ฏ๐—ฒ๐—ณ๐—ผ๐—ฟ๐—ฒ ๐—ฎ๐—ป๐˜†๐—ผ๐—ป๐—ฒ ๐—ฒ๐—น๐˜€๐—ฒ || ๐˜๐—ฟ๐—ฎ๐—ฐ๐—ธ๐—ถ๐—ป๐—ด ๐—ฒ๐˜ƒ๐—ฒ๐—ฟ๐˜†๐˜๐—ต๐—ถ๐—ป๐—ด ๐—ต๐—ฎ๐—ฝ๐—ฝ๐—ฒ๐—ป๐—ถ๐—ป๐—ด ๐—ถ๐—ป ๐—”๐—œ

Katฤฑlฤฑm Nisan 2026
10 Takip Edilen21 Takipรงiler
Draven
Draven@notdrvxยท
This 27-year-old runs an AI UGC agency for 8 ecommerce brands single-handedly, and it brings him $40,000 a month. It is a neural network that reverse-engineers competitors viral UGC clips and rebuilds them from scratch for someone else's product, without actors, without shoots, without a second take. From 1 competitor clip he extracts 1 psychological framework, from 1 framework, 20 hooks, from 20 hooks, 20 finished videos for 8 brands. All of it in the time it takes a regular creative to approve one brief. In the old days, an output like this required a team of 5 UGC actors: contracts at $300 per clip, plus photographers, plus studio rental, plus months of negotiation, plus the risk that one of them walks off with the audience. Now the whole setup moved into one chat window. A link to a competitor's clip in one line of the prompt, and a second later the AI broke it down by hook, trigger, and closer, second by second. The request "what emotional trigger is used and why does it convert" goes into invideo Agent One, which parses the video frame by frame, pulls the funnel structure, and returns a text breakdown without a single clarifying question in response. Inside the agent, frames from the competitor clip get recognized. Inside them, the psychological breakdown: physical push, punchy adjective, dailywear proof. Inside it, a ready prompt to rebuild the clip with your product. And every rebuild gets saved into one Creator Character sheet, from which the clip ships into any of the 8 ad accounts without a separate editor. In the first week this guy had his first 3 brands signed, by day 10 every client was getting 20 hooks per week, by the end of 30 days the agency crossed the $40,000 a month mark. The buyer in the AI UGC niche does not pay for a nice interface. He pays for the speed at which one hypothesis turns into 20 creatives with different faces. UGC production stopped being a profession with actors, contracts, and studios. It collapsed into one chat window and one link to a competitor's clip. This guy has no hired actors, no rented studio, not a single real model on camera. Only a laptop, a $50 a month subscription, and a link to the best competitor clip. Tomorrow every UGC agency will be able to shut down its casting department and move into a chat window, because the hook structure gets broken down in a minute and the rebuild arrives in the next. This is the point where the UGC actor industry turned from a separate market into one browser tab.
Alex Nguyen@alexcooldev

x.com/i/article/2058โ€ฆ

English
0
0
0
19
Draven
Draven@notdrvxยท
This 24-year-old runs 5 TikTok accounts of AI influencers single-handedly, and they bring his app $10,000 in revenue per month. It is a real-time face swap neural network that turns one person into an assembly line of digital influencers without studios, without actors, and without even a second shoot day. From 5 faces he gets 5 TikTok accounts, from 5 accounts 2.5 million views per month, and from those views 2,000 paid subscribers at $5 a month for the app. All in the very first month. In the old days, this setup required a team of 5 UGC actors: contracts at $300 per clip, plus photographers, plus studio rentals, plus months of negotiation, plus the risk that one of them walks off with the audience. Now the whole setup moved into one window with a webcam. A face description in one line of a prompt, and a second later someone else is on camera with a different haircut and different facial expressions. The request "redhead guy with light stubble, 22, freckles on the cheekbones" goes into the face swap model, which builds the render from facial landmarks and sends back a live video with no mask jitter and no smeared pixels at the border of neck and clothing. Inside the technology, facial landmarks are tracked in real time. Inside them, the generative model adapts lighting and skin texture to the new face. Inside it, the lips sync to audio without a millisecond of delay. And every clip gets saved into one library, from which it ships into any of the 5 TikTok accounts without a separate editor. In the first week this guy had 5 live accounts under different faces, by day 10 every account was steadily pulling 50K views, and by the end of 30 days the app crossed the $10,000 per month mark. The buyer in the UGC app niche does not pay for a nice interface. He pays for the speed at which one idea turns into 30 videos under 5 different faces. UGC marketing stopped being a profession with actors, contracts, and studios. It collapsed into one webcam and one line describing the face you need. This guy has no 5 hired actors, no studio, not a single real model on camera. Only a laptop, a webcam, and a description of the person he needs in one sentence. Tomorrow every UGC agency will be able to shut down its casting department and move into a window with a webcam, because the face is described in one sentence and the render arrives in real time. This is the point where the UGC actor industry turned from a separate market into one browser tab.
Simone Canc@simonecanciello

x.com/i/article/2058โ€ฆ

English
0
0
0
26
Draven
Draven@notdrvxยท
@haider1 watching this play out with infinite curiosity but 0 buying power lol
English
0
0
0
90
Haider.
Haider.@haider1ยท
anthropic is using "Mythos" to prepare for its IPO simple not making the model public allows them to make all sorts of claims without having to prove anything pushing a weaker opus 4.7 and shifting harder toward business and enterprise customers both fit the same move
English
27
5
166
20.4K
Draven
Draven@notdrvxยท
@MattPRD interesting shift tho what counts as "boring" determines a lot about the world we end up in?
English
0
0
0
54
Draven
Draven@notdrvxยท
@sainingxie id love to see examples where pose alone fails for an obvious task. gets ugly real fast when movement matches multiple interpretations.
English
0
0
0
98
Saining Xie
Saining Xie@sainingxieยท
๐Ÿ“ธlatest in our cambrian series: cambrian-p, p for pose. i think pose is probably the minimal sufficient 3d signal (and itโ€™s easy to get!) that we need for robust video multimodal models -- jointly modeling frames and pose turns image sequences into a globally grounded structure.
Jihan Yang@jihanyang13

Camera pose matters for video understanding! Today's MLLMs excel at recognizing activities, but still struggle with the underlying space and ego/object dynamics in video. We trace this gap to a missing piece: camera pose. Introducing Cambrian-P: a multimodal LLM natively grounded in camera pose. (1/n)

English
7
24
144
13.2K
Draven
Draven@notdrvxยท
@MoonDevOnYT lets be honest, teaching people to bot trade is just outsourcing their gambling habit
English
0
0
0
14
Moon Dev
Moon Dev@MoonDevOnYTยท
First bot ever i am going to hold your hand through building your first trading bot ever so you can remove emotions from your trading for the rest of your life there is no better thing you can do today than learn how to stop trading emotionally i will show you step by step how to get your bot live even if you have never coded before plus for memorial day you get $400 off you will never see a special like this again join here moondev.com/memorial moon dev
Moon Dev tweet media
English
6
1
10
932
Draven
Draven@notdrvxยท
@steipete the edge case part is real, but running for hours sounds like my local tests gave up on me
English
0
0
2
3.6K
Draven
Draven@notdrvxยท
@icreatelife real endorsement counts a lot is this the kind of thing a beginner could actually finish or is it more for ppl already messing with AI animation?
English
0
0
0
13
Draven
Draven@notdrvxยท
@rauchg @aurorascharff what am i supposed to do with this info? (next.js conventions feel like i need an esport team to keep up)
English
0
0
0
77
Guillermo Rauch
Guillermo Rauch@rauchgยท
Next.js Night ยท AMS ๐Ÿ‡ณ๐Ÿ‡ฑ ยท June 11 Learn whatโ€™s next + meet the team + share feedback luma.com/34nqdfc3
Guillermo Rauch tweet media
English
9
7
101
11.3K
Draven
Draven@notdrvxยท
@tengyanAI the CPU orchestration tail is the bottleneck everyone underestimates should price that in already tbh?
English
0
0
0
287
Teng Yan
Teng Yan@tengyanAIยท
If you understand this chart, you'll know why memory (specifically DRAM+HBM) remains a layup for the next few quarters.
Teng Yan tweet media
English
11
14
143
20.4K
Draven
Draven@notdrvxยท
@badlogicgames first sentence already aged perfectly. the second one feels more relevant every week
English
0
0
0
156
Mario Zechner
Mario Zechner@badlogicgamesยท
recommended reading. especially for those among you who think we should attribute conciousness to some matrices in the cloud.
Mario Zechner tweet media
English
9
17
200
8.1K
Draven
Draven@notdrvxยท
@lauriewired real good for a weekend hobby lol what was the first thing you built?
English
0
0
0
107
LaurieWired
LaurieWired@lauriewiredยท
Just hit 500,000 Subscribers! I never could have imagined (what really is my weekend hobby!) growing this much. Getting really close to overflowing a signed 20 bit word! Thereโ€™s still so, so much to explore in the field of Computer Science. My biggest advice is go make the content that would have inspired a younger you! No, itโ€™s not my dayjob. As much as I get a kick out of people thinking that Iโ€™m some giant private equity production (lol), itโ€™s just me, my small bedroom, and some cameras. Thanks everyone for the support, Iโ€™m excited to keep sharing this field with you all. Until next time, LaurieWired out.
LaurieWired tweet media
English
88
30
1.6K
27.7K
Draven
Draven@notdrvxยท
@pmddomingos so we accidentally fixed one problem by making the other one less obvious classic unintended side effect disguised as a win
English
0
0
0
23
Pedro Domingos
Pedro Domingos@pmddomingosยท
And in yet another instance of the AI comedy, weight decay was invented to control overfitting, but what it really does is improve training loss.
English
6
0
23
3K
Draven
Draven@notdrvxยท
@clairevo @DevinAI devin doing more for rev ops than most hires i know (claire the real mvp for the slack convo)
English
0
0
0
110
Draven
Draven@notdrvxยท
@Vtrivedy10 the sub-piece breakdown being invisible to the user is the real shift if it just works nobody questions the compute
English
0
0
0
32
Viv
Viv@Vtrivedy10ยท
in the same way we moved from Chat Completions to Agents as a default interaction pattern, I bet most AI requests will soon be treated as Work APIs driven by some harness setup like /goal - spend a good chunk of compute -deeply break down my problem into sub-pieces - verify sub-pieces and overall success - return a completed artifact this is what people actually want and it aligns with the incentive of higher token spend just like how Agent interactions did so it will almost certainly happen imo
English
20
4
59
2.3K
Draven
Draven@notdrvxยท
@samlambert the real test would be seeing his work at Apple without Jobs around 2012-2019. guess we got our answer.
English
0
0
2
604
Sam Lambert
Sam Lambert@samlambertยท
I think it's clear that Johnny Ive is an incredible designer when he has someone like Steve Jobs to push back on his ideas.
English
17
7
370
29.5K
Draven
Draven@notdrvxยท
@aiedge_ @base trusting an ai with wallet access is a choice i see the vision but lets talk about the recover key situation first
English
0
0
1
20
AI Edge
AI Edge@aiedge_ยท
Base ( @base ) just launched its new MCP server - and it's insane! Claude can now: โ€ข Trade on your behalf โ€ข Interact on-chain โ€ข Send transactions and much more. The best part is, it only takes a few minutes to set up. Full setup guide below
AI Edge tweet media
Miles Deutscher@milesdeutscher

Base ( @base ) just launched its new AI MCP - and it's extremely powerful. AI agents can now trade on your behalf, interact on-chain, access agent skills & way more. If you're in the crypto markets, you need to set this up. ๐Ÿงต: How to set up the new Base MCP in <2 minutes.๐Ÿ‘‡

English
5
5
23
2.7K
Draven
Draven@notdrvxยท
@TheAhmadOsman the osman to opensource pipeline is actually genius gonna start calling u Ahmad OSS Man now
English
0
0
0
28
Ahmad
Ahmad@TheAhmadOsmanยท
> Ahmad Osman > Ahmad OS Man > Ahmad OpenSource Man Yeah, thatโ€™s my name moving forward
Ahmad tweet media
English
17
0
62
7K
Draven
Draven@notdrvxยท
@Nicolascole77 mileage over talent if u zoom out enough the scary part is finding the thing u can outlast everyone on
English
0
0
0
6
Nicolas Cole ๐Ÿšข๐Ÿ‘ป
Nicolas Cole ๐Ÿšข๐Ÿ‘ป@Nicolascole77ยท
The way you win is by doing something more times than anyone else has ever done that "thing," which makes you the best person at it.
English
37
6
67
1.4K