AGI requires the capability of continuous self-learning. How to let agent learn like a human. Here are some thoughts of mine. There could be 3 basic types of learning processes of human. 1. Learn directly from the interactions with environments. Learn by environment rewards. 2.

Most L1s chase TVL, TPS, dApps, and partnership counts. At 0G Labs (Ø,G) - AI L1, we take a different route. Yes, those metrics matter, but what really matters is usage. Empowerment. User experience. Not just numbers on a chart, but tech that people actually use. So “wen TGE? wen mainnet?”

To be the owner and teacher of your agent, you can help improve the intelligence of your agent through sending it to some “campus” to learn new expertise with new knowledge. The 0G Labs (Ø,G) - AI L1 fine-tuning service platform is such a campus. You can communicate with the teacher, i.e.,

Productive day in Visit Washington, DC. Met with congressmen Congressman Bill Foster Rep. Jay Obernolte chief of staff to @repyvetteclarke.bsky.social. Grateful for the chance to be a resource to policymakers and talk about how we can align AI with democratic values and U.S. competitiveness — and why

Bonus: Ming and I met after DC in NYC for late night pizza at Prince Street Pizza and talked about another approach to scalability utilizing a distributed log layer based on our storage design / his insights from his talk he is giving at IC3 tomorrow

Nicely done!

This is exactly why decentralized AI is needed. In addition, this makes me think whether AI should have its own privacy or not. Should every AI agent need to be under the monitoring of at least some human beings?

The backdoor attack paper of 0G Labs (Ø,G) - AI L1 got accepted by ICCV 2025. This paper showcases the vulnerability of centralized federated learning and justifies the needs of decentralized model training with verifiability.

It is still not that clear why LLMs have hallucinations. There are two interesting hypotheses. 1. The LLM cannot differentiate between the data it is given and the data it generates. During its inference, once it generates some words and sentences, it assumes that these are the

The ecosystem of 0G Labs (Ø,G) - AI L1 keeps boosting! 🔥

The first 100B parameter model trained in decentralized environment by 0G Labs (Ø,G) - AI L1 and collaborators. arxiv.org/abs/2506.21263

0GM!大家周末好，今天又是0G在Kaito中文区榜一的一天！ “你能不能抽空替我去一个地方 据说那的人都擅长给别人答案。” 0GM! Happy weekend everyone — it’s another day of 0G topping the Kaito Chinese leaderboard! “Could you spare some time to go to a place for me? They say people there

🚀 0G 0G Labs (Ø,G) - AI L1 指令升级上线！ 现在通过 NodeHub 内置指令安装 0G 节点时，已支持【安装】与【升级】两种模式。 用户只需输入相应环境变量，即可一键执行版本更新，无需重复部署，保持节点持续运行 💡 立即体验：hub.node-x.xyz #0G #NodeHub #InfraUpdate #节点管理

From zer0 to JAINE 🫧 A new milestone unlocked — JAINE has reached 32M on-chain interactions and nearly 600K unique users on 0G Labs (Ø,G) - AI L1! Next up: Gimo Finance (Ø,G) access code drops More devs, faster builds More partnerships More trading pairs And yes, more secrets It's fun to build on

The 0G Labs (Ø,G) - AI L1 ecosystem is so vibrant!

Just returned from an extraordinary experience at the 沪港Web3互补协作发展大会 in Shanghai - a cross-border gathering uniting regulators, builders, and Web3 leaders from Hong Kong and the mainland. I had the chance to deliver a short intro of myself in Chinese (not the easiest

Amazing creature that defies gravity like 0G Labs (Ø,G) - AI L1

Nice community effort!

The milestone is highlighted by Forbes! forbes.com/sites/digital-…

Performance of LLMs may not always become better with upgrades. This means the evaluation of a model for different scenarios is very hard. The best model overall might not be the best model for your application.