Just built The Golem Awakens! A complete single-file HTML5 platformer & puzzle game entirely with local AI on one RTX 3090 Ti over 10 working days.
3,050 lines • 2,038 functional JS • 424 lines JSDoc • 22.4% comment ratio • Zero dependencies • Zero build step • MIT licensed • 6 chambers
Play it now on the Tholem Labs website! (opens in any browser): https://t.co/Nv4v1iCHi4
Repo + full docs: https://t.co/tgjHybmrZY
This was a deliberate test of Hermes @NousResearch + local models for real interactive work. Thread on the workflow and results 🧵
Rest assured: Anthropic has a strategy that will play into this. Rumored IPO in the fall is plenty of time to build a counterweight narrative while political momentum pushes for regulation and antitrust directed towards SpaceX as a midterm issue.
Pay attention.
People eyeing $SPCX IPO should be pricing in the political aspect:
At just the $135/share evaluation Musk will be the first documented TRILLIONAIRE. That's a headline that politicians + activists aren't likely to ignore.
Plan accordingly...
Caution when using aggressive TDD with coding agents
"Tests are green" is a cheap, verifiable proxy reward during training.
Agents tend to:
1) Create a test file
2) import a mock/stub instead of the real module
3) ????
4) Declare victory
The time it takes to scrub all the mock data and hooks from your codebase will erase the benefits.
I'm NOT saying don't test. But don't trust your agents to test without watching like a HAWK.
Got the Hermes gateway up in telegram and holy crap @NousResearch cooked.
The CLI was impressive enough, but this is next level. Works nearly as well on ANY device; can swap between local/frontier, manage multiple sessions, send voice prompts.... I feel like I'm just scratching the surface here and I'm already mad with power.
@Teknium you guys deserve all the good things heading your way!
Double the cost and all your data are belong to us...
Requiring this telemetry in the name of safety?
Highest AI safety signal is distributed compute and open weights for public audit. Wont be long before they start calling for open models to be banned....
Novelty/creativity is often times just the collision of concepts that don't normally interact.
Models are optimized for compression and prediction. Next token prediction; in embedding space, concepts live where they co-occur in text. No persistent world model/intrinsic motivation for exploration. Creativity is inefficient, high risk of wasted compute.
You can fix it probably.
@omarvvvr All of them can mean different things to different people. If the extension is descriptive and helps people know what the site is for lean into that.
If you arnt cost sensitive you can register all of them and resirect to main site to reserve them and drive seo.
@oranahhX They don't know how to use it well, they see low effort AI slop on social media and want better content, and/or they are becoming ideologically captured by politicians trying to get elected mainly.
@RaminNasibov If demand for inference was lower than supply there would be a bubble. There is a lot of AI-intergrated SaaS that isn't really translating into increased revenue tho- might be a reckoning there.
@quantscience_ To anyone reading this, just pick one or 2 strategies and focus on optimizing that. Find your niche. Applies when building trading agents too.