THAT Fable 5 is clearly not back. What is back is something slightly better than opus 4.8 that uses the same name and consume way more tokens than before. Same prompt first release created magic, now it created a similar slop that opus 4.8 would produce
I think MR needs to implement something that makes username completely hidden, not only during the first minute, but the entire match, generating random names like Overwatch does. This will not solve the problem completely but at least it will sure difficult these people from targeting you guys that easy. Not a 100% solve but it could be a start.
@fayhecode I think he wants a few followers and likes. There is no actual evidence that this is true. That doesn't remove the fact that Opus is good but this seems a lot like several asset packs out there
@Pirat_Nation This game needed time, actual game dev time. they launched it raw, it was like a cake that you remove from oven before it's ready. seems like a cake but barely taste like one.
@JakeSucky At the time, devs complained about the engine and how hard it was to make things work. also the game was barely optimized, and the feedbacks were about content, there was not enough time to fix it. some devs said 6 months wouldnt fix much, so they gave up.
I think a good place for LLMs in games is to create placeholders so we can test things and use actors in the final product. For example this POC I put together, like, it's AI but I am able to test it before sending to real actors #gamedev
@Enscion25 Funny how the audio reveals the data that was used for training this. The audio has some sort of echo as it was from a speech in a large auditorium
I've been using Bolt since March, mostly for MVPs and prototypes for clients. The V2 is amazing but I noticed that the approach changed to be more vibe code oriented and it creates things that I don't want and it's really hard to guide the agent to not use the new features like database or auth. These features are really cool for vibe coders but I'm using legacy model for now because I have no option to turn things off. Even when I say "Don't" it says, "oh the system tells me that this feature is available, you know what? I will ignore your ask and use it anyway".
@petergostev One thing that I noticed while testing models is that they fail to calculate face flip correctly, specially when they try to create some voxel scenario. And seeing your video, I can see the issue still occurs. The colosseum for example, both messed up with the faces a bit
This morning we announced several upgrades to Claude Code.
We also launched two new features for managing context on the Claude Developer Platform.
Here’s what’s new:
Careful while building your dream app using AI tools.
Regardless of what you're prompting, always ask AI to revise architecture, look for logic flaws and security concerns.
Some models cut corners, specially when the codebase become bigger.
So, always double check the output. even if you just vibe coding.
A good practice is to ask one model to build and other to review. It's not 100%, but it's like a team, one can see things that other dont.