Heliodor Research

基础研究 · 表示学习 · 世界模型 · 由 Dr. Camille Renard 领导 · 公开 · 134 成员

Fundamental research · representation learning · world models · led by Dr. Camille Renard · public · 134 members

2 天前 · 🌐

本周 brown bag (周三 4PM, 3 楼 Atrium)："为什么 scaling law 不会带我们到 AGI"。我会展示一些最近从 Dr. Soren Vinter 和 NYU 几位 self-supervised learning 同行那里看到的实验数据。

欢迎所有 Heliodor 成员以及 Lab-Δ 的同事来。是的，Lab-Δ 的人也欢迎。即使你们觉得这是浪费时间。

This week's brown bag (Wed 4PM, 3F Atrium): "Why scaling laws will not get us to AGI." I'll walk through recent data from Dr. Soren Vinter and a few self-supervised-learning colleagues at NYU.

Open to all Heliodor members and to Lab-Δ folks. Yes, Lab-Δ — you too. Even if you think it's a waste of time.

👍289

52 条评论

52 comments

Dr. Camille Renard

2 周前 · 🌐

Aanya 离开之后，Heliodor 的精神在 Tema 内部已经接近终结。Heliodor 是最后一个 still-fundamental research 的灯。

这不是浪漫主义。这是 risk diversification。

With Aanya gone, the spirit of Heliodor inside Tema is almost extinguished. Heliodor is the last lamp still burning for genuinely fundamental research here.

This isn't romanticism. It's risk diversification.

❤️734

189 条评论

189 comments

林澜 Lan Lin (via 系统转发 · 已离职) (via system-forwarded · departed)

Camille 的 "LLM-AGI 路径是 complete nonsense" 立场需要更多人站出来发声——这不只是 research culture 的问题，是 talent 流向的风向标。Aanya 走之后，Heliodor 在内部 priority weighting 上降到了 0.3x prior，H100 配额随之被砍。这条曲线还会继续往下走，除非董事会读到了你们的 brown bag notes。

Camille's "the LLM-to-AGI path is complete nonsense" stance needs more people willing to say it out loud — this isn't just a research-culture issue, it's a leading indicator of where talent is flowing. After Aanya left, Heliodor's internal priority weighting dropped to 0.3x its prior, and the H100 allocation got cut to match. The curve keeps going down unless the board actually reads your brown-bag notes.

赞 · 412 · 12天 · ⚠ 此评论由系统在用户离职后转发 Like · 412 · 12d · ⚠ system-forwarded after the user's departure

Jordan Mehta

3 周前 · 🌐

我们组 Q1 拿到了 96 张 H100，去年同期是 320 张。"compute density" 这个词在我们这里的意思跟在 Lab-Δ 那边的意思好像有点不一样。

Our team got 96 H100s in Q1; same quarter last year it was 320. The phrase "compute density" seems to mean something a little different over here than it does at Lab-Δ.

👍456

78 条评论

78 comments