A few of these went out to 200k+ users. Some raised millions. All built by a small team that ships fast. Happy to break down how we did any of them, just ask.
@haider1 the professor saying "by now it's hard to be surprised anymore" is the real headline. When experts stop being shocked, the benchmarks become a formality.
@ErnestoSOFTWARE the retention jump is the real stat. 35% to 60% is not a marketing win, that's a product win. AI scaled the content volume but something in the content itself got better too.
@alexcooldev real UGC ads are cooked if this keeps compounding. The gap between "AI image" and "shot on iPhone" is basically gone now for skin and lighting.
@haider1 two weeks of RL restarts and they're already on a second millennium problem. The 88 hours / 10k agents thing is what gets me. That's not a research team, that's a compute campaign.
@trikcode DeepSeek V4.1-Flash hitting 88.1 on CyberGym while GPT5.6-Sol sits at 84.5 is the chart doing the heavy lifting here. Frontier pricing is becoming a harder sell every cycle.