Haven Feng

@HavenFeng

Postdoc @berkeley_ai, building Impossible. Prev. @MPI_IS @AdobeResearch @uni_tue Interested in recreating the physical world and the intelligence to do so.

USA

havenfeng.github.io

Joined October 2021

1,154 Following

5,339 Followers

Replies Reposts Media

Pinned
Haven Feng
@HavenFeng
Jul 16
Today, we’re introducing [schema]: a harness reaching 99% RHAE with Opus 4.8 + Fable 5 and 95.35% with GPT-5.6 Sol on ARC-AGI-3 Public set. [schema] makes an LLM think like a physicist. 🧵
00:00
126
Haven Feng
@HavenFeng
Oct 17
Sometimes things don’t work out (especially for LLMs). It’s important to know when to keep trying and when to retreat gracefully! 🖖
Daman Arora
@amuseddaman
Oct 16
1/ RL for improving task success has made LLMs remarkably capable. But what happens when the model fails? Turns out, training for graceful failure can be surprisingly tricky! Very excited to share our new paper: Training Language Models to Fail Gracefully 🧵
00:00
Haven Feng
@HavenFeng
Oct 13
Video world models need more than visual generation. We believe persistent visual memory is the key to making them truly intelligent.
Yiwen Zhang
@YiwenZhangYZ
Oct 13
Long-horizon video generation AND persistent visual memory? 🧠 That’s what MosaiChunk brings to autoregressive video models. Code, data, and checkpoints are out today! Check it out: mosaichunk.github.io
Quote
Haven Feng
@HavenFeng
Oct 13
Now autoregressive video models can remember. 🧠 Introducing MosaiChunk: persistent visual memory for long-horizon autoregressive video generation🧩. Instead of remembering everything, it remembers the right things. ✍️ 1/6
Haven Feng
@HavenFeng
Oct 13
Now autoregressive video models can remember. 🧠 Introducing MosaiChunk: persistent visual memory for long-horizon autoregressive video generation🧩. Instead of remembering everything, it remembers the right things. ✍️ 1/6
00:00
5
Haven Feng
@HavenFeng
Oct 12
Schema is now fully released — code and paper are out! 🚀 Beyond ARC-AGI-3, Schema also tackles MazeBench and DIG (Discovery in Games). The goal has always been broader: building agents that can discover, understand, and reason about how worlds work. More in Guanning’s post 👇
Guanning Zeng
@guanningzeng
Oct 12
On July 15, we released Schema, the first agent harness to reach ~99% RHAE on the ARC-AGI-3 public set. Today, we’re sharing the full paper and code, with new results and a closer look at why Schema works.
Quote
Haven Feng
@HavenFeng
Jul 16
Today, we’re introducing [schema]: a harness reaching 99% RHAE with Opus 4.8 + Fable 5 and 95.35% with GPT-5.6 Sol on ARC-AGI-3 Public set. [schema] makes an LLM think like a physicist. 🧵
1