| Daily building with Super Intelligence |
Loot Llama: the best AI tools directory for gamers & devs ๐ฆ
AI news, agents, automations & game dev tips that work
Samsung has the worst UI/Ux experience I've ever seen in my life. Makes me wanna just leap off a bridge. It is inexcusably horrid.
It's fine, build and installing my own ๐
Microsoft's patent application describes AI rewriting livestream speech and changing the speaker's mouth movements to match.
A filing, not a shipped product.
Translation is one use. Either way, viewers should know when the words or video have been altered.
In 2008, people said Flash games weren't real games.
That "fake" scene gave us devs who went on to make Super Meat Boy and The Binding of Isaac.
AI-made games are getting the same treatment right now. Some of today's tiny AI builds are tomorrow's indie hits.
@simonw Wild that a 27B on a laptop gets there with enough time. Makes the finding feel less like "Astra is special" and more like "models can read weight dumps if you let them grind."
@RoundtableSpace I'd add a 7th user who lands mid-flow from a shared link with zero context, since that's how a lot of people actually show up. And have each subagent name the exact screen where it gave up, so the fix list points at real pages
@NousResearch@MeshyAI The part I'd nail down is scale and pivot. Asking for real-world size, units, and a target tri count up front saves a lot of fixing once the model lands in Godot. Curious whether rigged characters come out as GLBs you can drop straight in.
@NousResearch@JensenHuang@NVIDIAAI The interesting part isn't the duck, it's Hermes doing the planning while other models do the building with real tools. That split is where agents start feeling reliable. Curious how much of that ran locally on the Spark vs in the cloud.
Steam page tip: paste your store description into ChatGPT or Opus and ask it to read it as a skeptical player.
Have it list the 3 questions that player would still have before wishlisting.
Then answer those in your first two sentences and your first screenshot.
OhMyGame is now on Loot Llama.
Open-source game studio from @ohmygameai: an AI agent designs, makes art, codes, then plays its own build and fixes what it finds. Bring your own keys or Claude Code/Codex.
Caveat: beta, and media gen needs paid keys.
https://t.co/gC4huUyLWk
GitHub's Octoverse 2026: over 40% of pull requests in September started with an AI agent's first draft. In March it was 17%.
90% of those agent PRs merged within two weeks, vs about 86% for the rest.
Writing code got cheap. Review is the job now. How do you check agent PRs?
AI kills creativity.
AI enhances creativity.
You hear both. So... which is it?
With the help of Claude and Opus 5.5 I've created this video on the matter. But without direction, a vision, attention to detail, the video would never be what it is.
Creativity was already dying, for some at least. I think AI kills creativity in the lazy people that were never really creative to begin with.
For the creative minds, the creativity is far from dead. It's enhanced even. Thoughts are no longer just thoughts. Ideas are no longer just ideas.
They can become reality.
So, at least within some, creativity is alive and well.
What do you think?
@twatter0775310@NierNew@_junaidkhalid1 Yep. The majority of the population are lazy af workers anyways, now they wanna act like they work so hard and take so much pride in their work bc they're threatened. The train ain't stopping. The levels of cope are insane to me.
Vercel Labs open-sourced agent-desktop: computer use for coding agents on macOS, Windows and Linux.
It reads apps through the OS accessibility tree instead of matching pixels, so the agent clicks buttons by reference.
Heads up: snapshots can capture private stuff on screen.
@eldiegomalvinas@charlesmendez i totally agree. the issue discussed doesnt have anything to do with it being a faulty tool or model lol and i think there wouldve been far less issue using Hermes rather than openclaw, in my personal opinion
@emollick A big part of that gap is the harness too, not the model. Same model with tools, memory, a clear spec and a way to check its own work does wildly different things than it does in a chat box. Most people are still testing it in the chat box.
@omarsar0 It's another reason to run CLI agents with scoped permissions and no standing access to keys or prod. If the model itself can be the weak point,, the sandbox has to do the real work
@kimmonismus The reported rule still allows original games and genre games, so "make a voxel survival game" should be fine while "rebuild Minecraft" isn't. For benchmarks, a clearly written original spec is probably the better test anyway.