Igor

Nice

Used ChatGPT Work on the go to create and refine 3D toy models for printing. It helped me work through shapes, and small design details directly from my phone. Disclosure: Submitted for OpenAI’s ChatGPT Work promotion; approved posts may receive credits. #ChatGPTWork #3DPrinting

Cursor made Composer cheaper

Built a Gmail-like client for @resend. You paste your API key locally and manage all your mail from there. For when you want to reply to some emails manually, but don’t have all the mail stuff set up. github.com/IgorShadurin/rese…

When a feature keeps bouncing

I killed my Pro. Burned every limit. 4 days until reset. I’m not an ML researcher, but I used Codex /goal like one. Gave it a task, it spun up a $$$ GPU, and started doing the kind of stuff I’d normally avoid because it felt like months of pain. Now I guess I’m forced to rest... aka finally do marketing.

Codex /goal is absolutely insane! I asked it to speed up LivePortrait on Mac for video generation. In 4 hours, it found all the slow parts and rewrote them for Mac. It found similar mathematical approaches, did the research, tested different optimizations, and basically doubled the speed with ease. Honestly, in human mode, I’m not sure I could have done this in 2 months. @gdb @thsottiaux

Trying to achieve my goal

Brainrot will become the foundation of YumCut. I’m pivoting YumCut toward creating brainrot videos. There is strong demand for this, and I’m already getting search traffic on this topic. I’ll try to satisfy that demand. To strengthen my position in Google, I’m creating an open-source tool that I can make content about.

If I can’t write an article about a feature, I won’t build it. I’ll try working in weekly cycles. 2 days building the feature, 3 days talking about it. If I can’t talk about it, I won’t take it on. The development part is clear, but here’s what happens on the other days: 1 day for writing and publishing an article, 1 day for a long video, and Friday for creating experimental short videos. Right now, most of my traffic comes from articles about YumCut and its development. These articles also keep getting stolen, which gives me more backlinks. That leads to more traffic from Google. Videos bring weak results compared to the time spent, but I enjoy making them. So I’d like to find a working system where videos also bring traffic.

Me brekfust

Not just Nvidia: the AI boom has reached SanDisk For the average consumer, SanDisk is a brand that makes USB flash drives. But the company also produces memory that is highly needed by data centers for AI workloads. Over the past year, SanDisk shares have risen by 3000%, and its market capitalization has reached $154 billion.

YouTube just killed a dozen startups. Now you can analyze any video for free and understand why it is losing views and what needs to be changed. YouTube added Gemini, and now you can chat about any of your videos. You can even ask for a second-by-second breakdown and have your script rewritten to improve the video. The main takeaway for me is that YouTube wants short videos that bait comments. And yes, it even suggested a few solid ways to do that.

GitHub brings traffic, but there’s a problem. Most new users come from GitHub because YumCut is open source. I get around 30 registrations per day. In my case, the GitHub page looks more attractive to Google than the actual website. But there’s one problem - you don’t know how they found you: no refs, no access to search queries. So the only way to understand why they signed up is by talking to them over email. Although maybe I should ask this during registration too?

How can I make TikTok fullscreen so I don’t have to open my phone while vibe coding?

Just wanted to show this off a bit. I’m feeding all my crazy ideas into Codex while I don’t have technical tasks. Here’s what I have built right now: * A browser-based CapCut alternative controlled by voice through an OpenAI subscription * An indexer for my footage. The script uses Codex to extract detailed info so any video is easy to find * Voice control lets me generate images, find footage, and place it into a video in a single request * Analysis of other people’s TikTok videos to create templates and fill them with my own footage Over the last few days, I’ve made Codex understand video, copy other people’s ideas, and adapt them using my footage. I haven’t used it in practice yet, but I’m really enjoying this. Should I open-source this?

Check it

Different types of developers