Phase 2 of the 0G airdrop registration is live.
Phase 2 allows users from Phase 1 who could not complete KYC registration to complete it alongside users who had selected Discord roles or Kaito yappers who were mistakenly omitted in Phase 1.
Register now: https://t.co/idOmC1p8OD
Community members with issues related to the Phase 2 registration should fill out the appeal form: https://t.co/UGvOYAFDHT
The Phase 2 Airdrop registration represents 0G’s unwavering intent to ensure that our community is at the heart of our decentralized AI operating system.
While we open the doors for Phase 2, the 0G team is reviewing appeals for the Testnet Node Runners airdrop and is finalizing the data for the Testnet on-chain activities airdrop.
Let's defy gravity.
0G Airdrop Registration is live.
We thank our community for supporting 0G from the start. 0G's first airdrop is only the beginning of building a lasting legacy.
Register now: https://t.co/5TlWsTXHe3
0GM, 另一篇非常重要的关于中心化开源大模型(GPT-OSS) 存在问题的深度分析,文章里也包含了@michaelh_0g 的重要评论:“在 AI 模型方面,更大并不总是更好——这是 web3 开发者长期以来一直在说的。”
文章主要核心点:
向透明度和本地部署的转变: GPT-OSS 模型可在消费级硬件上运行,降低了云端 AI 的高成本和隐私风险,使开发者、研究者和小型组织更容易访问和定制 AI。这被视为 AI 民主化的重要一步,类似于早期 GPT-2 的创新精神。
- 部分透明性: 虽然模型权重公开可用,但训练数据、方法论和完整文档仍未披露,导致模型难以审计、重现和信任。文章强调,这仅是“部分开放”,而非完整的开源。
- 与开源精神的比较: 作者将 GPT-OSS 与 Meta 的 LLaMA 系列和 Mistral 的 Mixtral 等真正开源模型对比,认为 OpenAI 的发布更像是对竞争的战术回应,而非真正拥抱开源原则(如开放数据、清晰许可和社区治理)。
- 安全与滥用风险: 开放权重允许轻松修改或移除安全护栏,增加了滥用风险,尤其当这些模型嵌入关键系统中时。文章引用过去 GPT-2 的发布争议,指出 OpenAI 曾因滥用担忧而延迟发布。
- 行业压力与背景: 此举可能源于开发者、研究者和监管者对 AI 透明度的需求压力,以及 Web3 开发者对高效、可定制模型的倡导。
传统AI大模型某种意义上也慢慢开始体会到了去中心化AI的压力了。
0GM, another highly important in-depth analysis on the problems with centralized “open-source” large models (GPT-OSS) has been released. The article also includes an important comment from @michaelh_0g: “When it comes to AI models, bigger is not always better — something Web3 developers have been saying for a long time.”
Core points of the article:
Shift toward transparency and local deployment:
GPT-OSS models can run on consumer-grade hardware, lowering the high costs and privacy risks associated with cloud-based AI. This makes it easier for developers, researchers, and small organizations to access and customize AI. It’s seen as an important step toward the democratization of AI, reminiscent of the early innovation spirit of GPT-2.
Partial transparency:
Although the model weights are publicly available, the training data, methodology, and complete documentation remain undisclosed, making the models difficult to audit, reproduce, and trust. The article stresses this is only “partially open,” not fully open-source.
Comparison with the spirit of open source:
The author compares GPT-OSS to truly open-source models such as Meta’s LLaMA series and Mistral’s Mixtral, arguing that OpenAI’s release appears more like a tactical response to competition than a genuine embrace of open-source principles (e.g., open data, clear licensing, and community governance).
Safety and misuse risks:
Open weights make it easy to modify or remove safety guardrails, increasing the risk of misuse—especially when such models are embedded into critical systems. The article cites the past controversy over GPT-2’s release, when OpenAI delayed publication due to misuse concerns.
Industry pressure and context:
This move may stem from growing pressure from developers, researchers, and regulators for AI transparency, as well as advocacy from Web3 developers for efficient, customizable models.
In a sense, traditional large AI models are gradually beginning to feel the pressure from decentralized AI.
Yap early, yap only, yap often.
@KaitoAI is connecting AI, attention and capital with Yaps.
Just claimed my social card and I'm accumulating Yap points in real-time.
Claim yours 👉 https://t.co/kK09rG2Boh