ChatGPT Image 2 + Seedance AAA Game

DeepSeek-V4-Pro has been launched! My heart goes out to team Uber however as they could have lasted more than 4 months with using Deepseek v4 for their agentic operations. Nonetheless here are some outstanding features of the model: - Enhanced Agentic Capabilities: Open-source SOTA in Agentic Coding benchmarks. - Rich World Knowledge: Leads all current open models, trailing only Gemini-3.1-Pro. - World-Class Reasoning: Beats all current open models in Math/STEM/Coding, rivaling top closed-source models.

you shipped something cool, posted a screenshot of your terminal and... nobody stopped scrolling if we are being real, in 2026, screenshots are dead weight on a feed who reads monospace text in a 2-inch preview? so, i created a skill to show llms how to record my cli running in real time and hand me a video file ready to post it writes a script, captures every frame in a real terminal, and gives me a video try it here 👇 npx skills add jondoescoding/vhs-demo-videos

i had hours of content i wanted to repurpose but my ai couldn't watch any of it so, i built a tool that turns videos into text locally on my machine, then feeds the transcript straight to the llm now i describe what i need & it writes posts with the full context of what i actually said, not some hallucinated summary runs locally & 100% offline check it out 👇 github.com/jondoescoding/scribe

I have a son preparing for his grade 4 PEP exams, and I wanted a way to generate past-paper-like multiple-choice math questions for him to practice. Plain ChatGPT struggles at this. It cannot do math notation, it often provides incorrect or duplicate multiple-choice solutions, and the question quality is pretty low standard and uncreative. Not to mention the fact that almost half of the questions in the real exam come with diagrams and follow a certain question format. After about 3 cumulative days of coding and 25% of my $200 Cursor usage tokens, I'm able to produce a reasonable level of quality past paper. I am even able to use compact mode so I can print it on less sheets of paper to save ink. See attached paper and answer sheet as well as the codebase structure. It uses AI and prewritten templates to generate, evaluate, and validate questions before creating a paper. I plan to extend it to an online tool soon, right now it's a Python application with about 30 command-line interface commands.