Just my opinion: The real reason PPO can handle long-horizon tasks? The Value Model for multi-step task. Honestly, it just shifts the training difficulty from GRPO’s GRM onto the Value Model. I mean, the problem isn’t solved — it’s just been relocated. The core question remains: How do you get stable process supervision in long-horizon tasks? That’s the real bottleneck.

🧵 Deli AutoResearch SKILL is now officially open source! 🎉 Alongside it, we’re dropping our 4th survey paper — this time on Self-play. Inspired by AlphaZero, we got a powerful insight: prior knowledge doesn’t always lift the ceiling. Models can discover more globally optimal solutions just by playing against themselves. The biggest change in this paper? For the first time, the AutoResearch Agent autonomously planned GPU experiments — and submitted actual RL runs on the DeepSeek 285B model. The entire RL pipeline — experiment design, code writing, running, debugging, and conclusion summarization — was 100% automated, with zero human intervention from me. This was incredibly difficult, but an incredibly important step. GRPO is the tool being called by the AutoResearch Agent here. We see this as the beginning of our Continual Learning research journey. 🚀 As always, this is my personal research project, unaffiliated with any organization. All views are my own. #AI# #ReinforcementLearning# #SelfPlay# #OpenSource# #AutoML# #ContinualLearning# #DeepSeek#

🚨 SERIOUS COMPLAINT 🚨 I am formally and angrily calling out the Beijing Municipal Government and ByteDance – specifically their construction site at Lanjing Lijia. At 2 AM on a weekend, excavators are working nonstop. The site is barely 200–300 meters from my home. How are people supposed to sleep? I filed a complaint with 12345 before. Their official response was: “This is a key municipal infrastructure project. Legal working hours are 6 AM – 10 PM, including weekends and public holidays (May Day, etc.).” That already made me furious. But now – weekend, 2 AM – they’re digging like it’s noon. Totally unacceptable. I recorded a video and posted it on domestic social media. Then Xiaohongshu banned my account – my video is invisible to others. That tells me one thing: the Beijing government saw my post. They chose to silence me instead of fixing the problem. As of now, my 12345 complaint is still unanswered – zero response. I never wanted to take extreme measures. But you leave me no choice. --- #Beijing# #ByteDance# #NoisePollution# #SilencingCitizens# #LanjingLijia# #12345# #RightToSleep#

💎 "My heart is not a stone; it cannot be turned." 💎 Maybe in another timeline 🕰️🌌, DeepSeek doesn’t exist 🐳, there’s no explosion of open-source models 📂🔓, and no API services that simply chase reasonable profit ⚖️💵. But anyway… I’m just endlessly grateful that in *this* timeline, I can pour my youngest, most alive years into a dream of AGI for everyone 🌍🤖✨. That alone is the greatest happiness of my life. I ask for nothing more. 🫶 It still feels surreal 🤯, like Eren Yeager on a summer afternoon, napping under the shade of a tree 🌳☀️😴💤, dreaming a dream that spanned two thousand years ⏳💭. This world is full of utilitarianism dressed up as dreams 🎭, but we have to trust the power of trust ✨🙏. After all… 我心匪石，不可转也 ——《诗经·邶风·柏舟》， 💎 My heart is not a stone; it cannot be turned. 💎❤️ #DeepSeek# #AGIForEveryone# #OpenSource#