Twitxy Saved

basilica

Verified account This account is verified. Learn more

@basilic_ai

Sacred Compute - an order of @covenant_ai

tplr.ai/links
basilica.ai Joined August 2025
5 Following 1.4K Followers 133 Posts

Latest posts 5

basilica Verified account @basilic_ai
Compute is a control point. When it sits with a few providers, access narrows to their terms. Basilica is independent compute for the teams that train and serve their own models, so the layer under your AI is not a switch someone else holds.
covenant @covenant_ai Jun 13
One directive, one afternoon, and global access to Anthropic's most capable models was gone. Sovereignty does not end at nations. The tools people think with should not sit behind one party's switch. This is why Covenant builds open, distributed AI infrastructure.
View quoted post on X
basilica Verified account @basilic_ai
What if AI could improve itself overnight? No engineer babysitting, no manual tuning.

@Hevalon built exactly that on Basilica. Point an AI at a problem, it runs 15 experiments on its own, picks what works, discards what doesn't. Wakes up smarter.

This is what programmable GPU infrastructure unlocks.
I built autoresearch-rl and pointed it at GRPO fine-tuning on @basilic_ai A100s. One command. 15 iterations. Zero human intervention. 100% infrastructure success rate.

GSM8K pass@1: 26% baseline to 36%.

The hard part wasn't the search algorithm. It was the infrastructure.
View quoted post on X
0 replies 11 reposts 48 likes 3.7K views View on X
basilica Verified account @basilic_ai
What if you could tell an AI to improve itself overnight and wake up to a better model in the morning?

@Hevalon built exactly that on Basilica. No babysitting. No manual tuning. Just one command.

Full writeup: templarresearch.substack.com/p/autonomous-r…
templarresearch.substack.com Autonomous RL Fine-Tuning on Ephemeral GPUs: Extending Karpathy's autoresearch with autoresearch-rl The hard part of autonomous ML research isn't the search algorithm. It's where the GPU comes from.