This is amazing. I’m genuinely shocked. 8 hours in to my 20 hour bolt.new hackathon and I have a working prototype of my mobile app in expo go. In 4 hours I was able to get the backend working on Convex and get the AI badge production pipeline working among other AI

I’d like to officially share what I am wrapping up for the bolt.new hackathon: Leafari - A mobile app that makes learning about the plants all around you more fun. It’s addictive and fun to use, think Pokemon go, but you capture plants. Here is the story behind it: The

WOW! Some of these entry videos are super polished. I'm blown away by what you all have built. I just went vlog style with my beginner video editing skills. Hope its enough to get noticed!

Heres something it would be good to get guidance on from bolt.new on the worlds largest hackathon: There are no official rules on how to go about updates to an app now that the deadline has passed. For those of us that set out to build a real app for REAL users - putting

With just a dash of effort these people slinging AI responses can make them nearly undetectable. Peak lazy.

I’m ready for this “build an app / website / etc, with one prompt” fad to end. The winners that emerge from this will be process oriented and will guide you through all the phases of ideation, architecture and defining documentation before writing a single line of code.

"Draw me an SVG of a Pelican riding a Bicycle" The king of all benchmarks from Simon Willison Tells me everything I need to know about a model.

Prediction: Genie and other models like it will finally kick off the VR boom. Imagine using this thing with a high end VR rig?

Gpt-5 is the first to nail the bike frame in the pelican riding a bike svg benchmark!

What a difference a prompt can make: GPT-5 "Generate an SVG of a pelican riding a bicycle" vs "Draw an SVG of a pelican riding a bicycle. Make it high quality with good detail and quality shading and textures. Work really hard on this one. Make sure to nail the likeness of the

Seeing a lot of gpt-5 vs opus 4.1 posts that use the same prompt for both and declare opus the winner. This misses the fact that Claude models are very opinionated - they will fill in details that you did not prompt. This trait also makes it hard to tame. I have been able to

Gpt-5 might be the most misunderstood model. It’s really good for a certain prompting style, while coming off flat for those who lazy prompt and are used to models that make more assumptions.

🤣 everyone just like… chill, ok?

This entire thread just blew my mind. Shrimp are seriously OP, someone needs to rebalance the simulation.

Can confirm. Jack is very helpful for helping lock in b2c marketing strategy!

Theory: Every X algorithm change will revolve around rewarding content that can train Grok on knowledge not easily obtained through other training data. This aligns with what Nikita Bier is saying.