Shun Lyu

Last month, Family Rewards was featured on @indieappsanta ! We saw 26x daily downloads during the campaign and maintained 2x our baseline after. Great ASO work too! Highly recommend for indie devs! #indieappsanta #buildinpublic

🌞0⃣4⃣🌞 Tired of the daily chore battle? 🧹💥 Family Rewards: Chore Chart makes it easy to assign tasks, track progress, and reward your kids with points, badges, and custom prizes! 🎯✨ From “should do” to “don’t do” rules, this smart chores tracker builds responsibility while reducing your stress. Create challenges, manage habits, and keep everyone motivated — all from your phone! 📱💪 📩 Questions or feedback? Reach us anytime at fa.shun.lyu@gmail.com Download Family Rewards: Chore Chart: 👉 apps.apple.com/US/app/id64598… To get the promotion, download the Indie App Santa in the App Store now! 👉 apps.apple.com/us/app/indie-a…

Pro tip: When going from any big model to a small one YOU NEED few shot examples. Claude 3.5 gpt4o/ gemini/ etc are really good at understanding what you mean with minimal instructions. gpt4 mini + haiku + flash are not like that - they need a lot more effort put into their prompts So things like 1) few shot examples 2) Clearly formatted instructions (so XML/ markdown) 3) Edge case handling Also never start with the small models in your system. Use the big guns, then once you understand where they are good/bad at, you can distill into smaller ones (or fine tune!) i do not recommend using smaller models for complex reasoning tasks - they're much better at narrow, scoped down things oh and don't write hand prompts, instead go to claude and explain what you're trying to do, and have it write a better prompt + examples.

@TheCustomerGuy @d4m1n @kensavage github.com/davabase/whisp… plus finding a way to get system audio piped into it

@levelsio It's like trying to bottle lightning

Discover Habit Tracker: Gamify & Reward! With over 300 templates for habits and rewards, setting up has never been easier. Download now and start your routine! #Productivity #HabitTracker

It’s been 6 months since I left my full-time job to pursue entrepreneurship. Runway is dwindling down, and mental stress is at an all-time high. Personal advice: Quit your job only if you have a huge war chest or a product with traction.

Recap of the first 2 years. Things really only started to move after 1.5 years of grinding and earning near-zero revenue. My bank account shrank down to almost nothing. bannerbear.com/journey-to-10k…

If you think you can ignore generative AI as a developer to keep your code pure, you're hamstringing yourself. In fact, it's just meta-coding. You still get to write instructions, just as a prompt. Top left 3 lines are mine. The rest is @AnthropicAI Claude. And the code works.

I would like to create an AI-based tool app, but I don't have any ideas at the moment. #buildinpublic

proof of life / updates: - left ny after a year to move back to topanga, ca (all i needed) - started freelancing for airbnb a couple of months ago - started a side project called cherries with one of my best friends - enjoying summer (favorite season)

Looks like the @TropicalMBA guys opened a French business and did not tell the rest of us.

A truly Hugh-nique specimen.

@d4m1n 🎆

👋 goodbye GPT 3.5-turbo GPT-4o mini is out 🎉 It's cheaper: 60% less vs GPT-3.5 Turbo $0.15 per 1M input tokens $0.60 per 1M output tokens. It's smarter: 82% on MMLU vs. 69.8% for GPT-3.5 Turbo, excelling in textual and multimodal reasoning. It's more capable: Supports text and vision, with future plans for audio and video; improved multilingual understanding, 128k context window, and up to 16k output tokens per request.

“AI knobs” is an excellent shorthand for what we should aim for as product builders. We’re tool-using creatures. When you wield a hammer, your proprioception extends to the hammer itself. Yet in most areas we still haven’t wrangled AI into something that feels like a tool.

Initial impressions of gpt4o Mini: - Surprisingly capable for its cost - Outperforms Haiku and Llama 3 in my tests It needs different prompting, but with a few beefy prompts it's nearly a 4o replacement at 10x cheaper Should I do a mini-guide on how to go from prompting large models to smaller ones?

On average, building habits takes around 66 days before the behavior change becomes automatic. Read more here! buff.ly/4fi8FZE #HabitTracker #habits

Hey everyone! I'm planning to develop a tool app based on the power of large language models. 🚀 Any suggestions? I'd love to hear your thoughts and ideas! #buildinpublic