@svlevine: Seohong wrote a mystery novel. You won't believe whodunit
Summary
Seohong Park wrote a blog post about mysteries in behavioral cloning with real-world robot data, such as overfitting being beneficial, and shared thoughts on demystifying them.
View Cached Full Text
Cached at: 08/26/26, 11:26 PM
Seohong wrote a mystery novel. You won’t believe whodunit
Seohong Park (@seohong_park): Behavioral cloning mystery
https://t.co/VqxzvcfGSx
I wrote a new blog post about “mysteries” in behavioral cloning that appear with real-world robot data (e.g., overfitting is “good”). I also tried to demystify them and shared my thoughts!
Similar Articles
@AgustinLebron3: This blog post still lives rent-free in my head. https://johnsalvatier.org/blog/2017/reality-has-a-surprising-amount-of…
A tweet recommends the 2017 blog post 'Reality has a surprising amount of detail' by John Salvatier, arguing AI labs would benefit from its insights about the hidden complexity in real-world tasks.
@vivek_2332: found a really good blog digging into how @AnthropicAI identifies and mitigates reward hacking during RL training. reco…
This article summarizes a blog post detailing Anthropic's methods for identifying and mitigating reward hacking during RL training, including hidden tests, stress-test sets, SAE monitoring, and environment redesign.
@Voxyz_ai: https://x.com/Voxyz_ai/status/2058586365034377230
The author maps common AI agent failures—such as confabulation, source amnesia, and phantom limb states—to human neurological conditions, proposing a diagnostic framework for debugging agents in runtimes like OpenClaw and Hermes.
@svlevine: Learning from suboptimal data is important, because robots make suboptimal data on their own, and the more robots there…
Sergey Levine highlights the importance of learning from suboptimal robot data and promotes OopsieData, a community effort to collect and open-source the abundant but underused failed robot interaction clips.
@qinzytech: https://x.com/qinzytech/status/2066585405479371092
A technical analysis of two approaches to building self-evolving AI agents: model-based (via architecture like SSMs or transformer with fast-weight updates, and training methods) and harness-based (via memory or meta harness that can rewrite itself). The author provides practical recommendations for different audiences.