Lila Sciences：互联网数据如已开采化石燃料，实验室才是无限 token 生成器——把湿实验建成「数据中心」，用可验证实验室奖励做科学超级智能；模型即产品，而非传统 biotech 管线。
- 实验室=数据中心/无限 token
- 苦涩教训 All-in 科学
- 可验证奖励在湿实验
嘉宾：Andy Beam & Rafa Gómez-Bombarelli（Lila Sciences） · 日期：2026-07-16（PT）
播出信息
- 日期：2026-07-16（PT）
- 时长：约 101 分钟（RSS=6064s）
- 章节末段：1:35:53（约 95%）
节目结构（YouTube 章节）
- We have but one internet — 00:00 🎧 · ▶️
- Intro & guest backgrounds — 00:46 🎧 · ▶️
- Inside the AI Science Factory: the "PCI bus" & the API line — 10:01 🎧 · ▶️
- Safety, security & scientific rigor — 14:34 🎧 · ▶️
- Why Lila isn't a biotech: the model is the product — 28:16 🎧 · ▶️
- 10 trillion tokens & why the general model wins — 32:36 🎧 · ▶️
- Scaling & the "bittersweet lesson" of materials — 41:42 🎧 · ▶️
- The in-vivo CAR-T proof point — 44:12 🎧 · ▶️
- Clinical translation & loading the die — 52:56 🎧 · ▶️
- Ken Stanley & open-endedness — 59:40 🎧 · ▶️
- Orchestration, scaling & faster assays — 1:07:07 🎧 · ▶️
- Instrument onboarding & the 10T-token dataset — 1:14:54 🎧 · ▶️
- What's harder: materials or biology? — 1:31:33 🎧 · ▶️
- Bottlenecks, MFU & closing thoughts — 1:35:53 🎧 · ▶️
分节详解
1) We have but one internet — 00:00 🎧 · ▶️
💬 他们说了什么： 围绕「We have but one internet」讨论。字幕要点大意：嘉宾谈到与此相关的实践与判断（原文片段：But not just TechBio, what do you do in But not just TechBio, what do you do in But not just TechBio, what do you do in …）。
💡 为什么重要： 该段推进本集核心主张与证据链。
⚡ 争议点： 自动字幕非人工精校；技术细节以论文/官方材料为准。
2) Intro & guest backgrounds — 00:46 🎧 · ▶️
💬 开场引入核心命题与嘉宾背景，定下本集问题意识。
💡 安全、伦理与治理是不可绕过的约束。
⚡ 自动字幕非人工精校；技术细节以论文/官方材料为准。
3) Inside the AI Science Factory: the "PCI bus" & the API line — 10:01 🎧 · ▶️
💬 围绕「Inside the AI Science Factory: the "PCI bus" & the API line」讨论。字幕要点大意：嘉宾谈到与此相关的实践与判断（原文片段：it something where it's already hit diminishing returns. diminishing returns. diminishing returns. >> So when you …）。
💡 该段推进本集核心主张与证据链。
⚡ 自动字幕非人工精校；技术细节以论文/官方材料为准。
4) Safety, security & scientific rigor — 14:34 🎧 · ▶️
💬 围绕「Safety, security & scientific rigor」讨论。字幕要点大意：嘉宾谈到与此相关的实践与判断（原文片段：is, you know, a good strategy or that you didn't just waste a bunch of money? you didn't just waste a bunch of money? &g…）。
💡 安全、伦理与治理是不可绕过的约束。
⚡ 自动字幕非人工精校；技术细节以论文/官方材料为准。
5) Why Lila isn't a biotech: the model is the product — 28:16 🎧 · ▶️
💬 围绕「Why Lila isn't a biotech: the model is the product」讨论。字幕要点大意：嘉宾谈到与此相关的实践与判断（原文片段：experiment, trusting the verifier, trusting the simulator as the the trusting the simulator as the the ultimate ground t…）。
💡 架构与算力约束决定方法能否规模化。
⚡ 自动字幕非人工精校；技术细节以论文/官方材料为准。
6) 10 trillion tokens & why the general model wins — 32:36 🎧 · ▶️
💬 围绕「10 trillion tokens & why the general model wins」讨论。字幕要点大意：嘉宾谈到与此相关的实践与判断（原文片段：of sort of oh man, this domain actually applies to this domain. So, we have applies to this domain. So, we have applies …）。
💡 架构与算力约束决定方法能否规模化。
⚡ 自动字幕非人工精校；技术细节以论文/官方材料为准。
7) Scaling & the "bittersweet lesson" of materials — 41:42 🎧 · ▶️
💬 围绕「Scaling & the "bittersweet lesson" of materials」讨论。字幕要点大意：嘉宾谈到与此相关的实践与判断（原文片段：platform that is showing up all the time the more we talk with people. the more we talk with people. >> Interestin…）。
💡 该段推进本集核心主张与证据链。
⚡ 自动字幕非人工精校；技术细节以论文/官方材料为准。
8) The in-vivo CAR-T proof point — 44:12 🎧 · ▶️
💬 围绕「The in-vivo CAR-T proof point」讨论。字幕要点大意：嘉宾谈到与此相关的实践与判断（原文片段：to do with scaling. >> How far have you gotten so far? >> How far have you gotten so far? >> So on the…）。
💡 该段推进本集核心主张与证据链。
⚡ 自动字幕非人工精校；技术细节以论文/官方材料为准。
9) Clinical translation & loading the die — 52:56 🎧 · ▶️
💬 围绕「Clinical translation & loading the die」讨论。字幕要点大意：嘉宾谈到与此相关的实践与判断（原文片段：and also some of the intellectual labor and also some of the intellectual labor to, you know, get all the the pieces in …）。
💡 该段推进本集核心主张与证据链。
⚡ 自动字幕非人工精校；技术细节以论文/官方材料为准。
10) Ken Stanley & open-endedness — 59:40 🎧 · ▶️
💬 围绕「Ken Stanley & open-endedness」讨论。字幕要点大意：嘉宾谈到与此相关的实践与判断（原文片段：being used in pediatric arthritis. Like another like very niche area of another like very niche area of medicine. medici…）。
💡 开源与学术分工影响生态与长期信任。
⚡ 自动字幕非人工精校；技术细节以论文/官方材料为准。
11) Orchestration, scaling & faster assays — 1:07:07 🎧 · ▶️
💬 围绕「Orchestration, scaling & faster assays」讨论。字幕要点大意：嘉宾谈到与此相关的实践与判断（原文片段：are smaller. They I think some of them are like maybe 12 4 x 3. are like maybe 12 4 x 3. are like maybe 12 4 x 3. >&g…）。
💡 该段推进本集核心主张与证据链。
⚡ 自动字幕非人工精校；技术细节以论文/官方材料为准。
12) Instrument onboarding & the 10T-token dataset — 1:14:54 🎧 · ▶️
💬 围绕「Instrument onboarding & the 10T-token dataset」讨论。字幕要点大意：嘉宾谈到与此相关的实践与判断（原文片段：Like if we didn't have to do that like Like if we didn't have to do that like I'm very pumped about that because the I'm…）。
💡 数据与扰动设计决定因果/预测模型能否可信落地。
⚡ 自动字幕非人工精校；技术细节以论文/官方材料为准。
13) What's harder: materials or biology? — 1:31:33 🎧 · ▶️
💬 围绕「What's harder: materials or biology?」讨论。字幕要点大意：嘉宾谈到与此相关的实践与判断（原文片段：like valuable, but also important company for um not just biotech, but for company for um not just biotech, but for mate…）。
💡 该段推进本集核心主张与证据链。
⚡ 自动字幕非人工精校；技术细节以论文/官方材料为准。
14) Bottlenecks, MFU & closing thoughts — 1:35:53 🎧 · ▶️
💬 算力利用率（MFU）与瓶颈，收束资源现实。 依据字幕：Writing was behind him only to have to write to only to have to write to >> [laughter] >>…
💡 数据与扰动设计决定因果/预测模型能否可信落地。
⚡ 自动字幕非人工精校；技术细节以论文/官方材料为准。
金句
- 我们只有一个互联网。它是化石燃料。我们已经水力压裂过了。
- 实验室应当摸起来像数据中心。
- 验证器是实验室本身。
