AI is getting smarter. But do we actually know how they work?

Stanford researcher Aryaman Arora takes us inside the science of AI interpretability, a field of research where we try to understand what’s happening inside language models.

We discussed Anthropic’s research of a hidden internal neural workspace in large language models, Golden Gate Claude, user awareness research at Transluce, and why different models are starting to develop increasingly distinct personalities.

We also talked about the various research traditions, what sparse autoencoders and causal interventions can reveal, and why we can actually edit a model’s internal thoughts.

Aryaman Arora is a PhD researcher at Stanford, advised by Christopher Potts and Dan Jurafsky. His research focuses on AI interpretability, language models, and linguistics, and he has also worked at organizations including Transluce and Redwood Research.

Timestamp:
00:00 What is interpretability research?
03:11 J-Space: Do models have an internal reasoning space?
07:34 Do models have internal reasoning space?
08:56 Can chain-of thought tell us what a model is thinking?
12:07 Interpretability, alignment, and trusting AI
16:10 Sparse autoencoders: translating AI’s internal representations
19:34 Two competing approaches to interpretability
23:32 The compute problem
24:11 Linguistics
30:06 Interpretability benchmarks
31:28 Gold Gate Claude: can we edit an AI’s thoughts?
34:09 Transluce and user awareness in frontier models
36:09 Weird chat and discovering strange model behaviors
37:02 Is it all about edge cases?
38:36 Can interpretability explain hallucinations?
40:27 H3 paper
43:05 Academia vs. industry
48:15 What’s next

About Valley101:
Valley101 is an independent media that tells deeply reported stories about the people, companies, and technologies shaping the future. We cover AI, semiconductors, startups, and the global technology industry through documentaries, explainers, and conversations with the people building it.
Subscribe for more stories from Silicon Valley and beyond.

Contact: video@sv101.net

China’s AI ecosystem is moving fast. Kimi, DeepSeek, and Z.ai are building increasingly capable AI models. But how much is Silicon Valley actually paying attention?
Xing Meng, partner at 5Y Capital, joins us for an inside look at China’s AI industry and the rise of Chinese open-source and open-weight models. We discuss Kimi and Moonshot AI, DeepSeek, ByteDance, the young researchers behind China’s frontier AI labs, and how China’s AI ecosystem differs from the US.
We also dig into the distillation debate, video models, data providers, and why Chinese models may be having a bigger impact on AI developers than on researchers at OpenAI, Anthropic and other frontier labs.

Timeline:
00:00 Xing Meng
02:06 Silicon Valley trip recap
05:45 Kimi, Moonshot AI and China’s young AI researchers
14:32 DeepSeek and unusual Liang Wenfeng
18:05 Z.ai, MiniMax and what makes a great AI lab
22:14 ByteDance, Zhang Yiming’s bold decision, distillation debate
26:54 Are Chinese open-source models actually changing Silicon Valley?
30:56 Are frontier AI models becoming commoditized?
35:03 Why China is leading in AI video generation
46:52 Neo Labs, AI agents and the future of model companies
49:05 The next frontier of AI training data
58:04 How US and China approaches data differently
1:04:26 AI for science
1:10:26 Silicon Valley’s biggest misconception about China’s AI ecosystem

About Valley101:
Valley101 is an independent media that tells deeply reported stories about the people, companies, and technologies shaping the future. We cover AI, semiconductors, startups, and the global technology industry through documentaries, explainers, and conversations with the people building it.

Subscribe for more stories from Silicon Valley and beyond.

Contact: video@sv101.net

What will the world look like 10,000 days from now?
In this episode, we sit down with Kevin Kelly (K.K.), a Silicon Valley visionary and the founding executive editor of WIRED. Drawing on the ideas in his book 2049: Possibilities of Next 10000 Days, we explore his vision of the world ahead.
By 2049, humans may fall in love with AI. A new generation may find themselves working in jobs that do not yet exist. AI could become embedded in nearly every aspect of daily life, while humanity enters a “mirrorworld” that digitally maps the physical world. And despite our ambitions in space, Kelly believes there may never be cities on Mars.
We also discuss questions that matter to all of us: How should we educate the next generation? If AGI becomes a reality, what uniquely human qualities will remain scarce and valuable? And what opportunities could China have in the emerging market for AI glasses?
Join us for a deep conversation with K.K.— a picture what our lives and our world might look like by 2049.

Timeline:
00:00 Technology’s Double-Edged Sword: Inversion and Optimism About the Future
04:19 AI and Education: Learning How to Learn
09:37 AI and Media: Content Will Explode, but Human Journalists Must Remain Accountable
16:31 AI and Emotions: The New Impact of the Mirrorworld
21:26 The AI Glasses Race: Could a Chinese Company Come Out on Top?
26:16 Robotics: Humanoid Robots Will Get Cheaper—but Not More Useful
29:21 Self-Driving Cars: Getting a Driver’s License May Become Harder for Humans
31:19 Space Technology: There Will Never Be Cities on Mars
34:34 Life Sciences and Longevity: Extending Health Matters More Than Extending Lifespan
39:04 Brain-Computer Interfaces: Non-Invasive Connections Within 25 Years
41:05 Silicon Valley in 2049: How the Path to Innovation May Change

About Valley101:
Valley101 is an independent media that tells deeply reported stories about the people, companies, and technologies shaping the future. We cover AI, semiconductors, startups, and the global technology industry through documentaries, explainers, and conversations with the people building it.

Subscribe for more stories from Silicon Valley and beyond.

Contact: video@sv101.net

The OpenClaw hype may have cooled, but we JUST stepped into AI's Agentic era.

At WAIC 2026 in Shanghai, we sat down with Vincent Koc, Chief Architect of the OpenClaw Foundation, to revisit the open-source AI agent framework that took the global AI community by storm, drove Mac minis to sell out in China, and triggered intense debate over AI safety, security, and control.

Vincent reveals what happened after the viral frenzy and the so-called “uninstall wave”: OpenClaw’s weekly NPM downloads actually climbed from 2.5 million to a record 4.2 million. Behind that growth are two very different forces—Chinese companies pushing employees to adopt AI agents through automation-driven KPIs, and Western users quietly bringing agent tools into the workplace from the bottom up.

Our conversation also explores what happened after OpenClaw founder Peter Steinberger joined OpenAI. Rather than being absorbed into a single company, OpenClaw moved into an independent nonprofit foundation intended to preserve the project’s open-source identity. Peter continues to provide technical stewardship, while the Foundation has taken on the work of keeping the project alive.

In the post-Claw era, the agent landscape is already being rewritten. The bigger question is: when will the "Aha moment" arrive for personal AI?

Timeline:
00:00 Conversation with Vincent Koc, Chief Architect of the OpenClaw Foundation
02:15 The OpenClaw Craze
03:41 Working with Peter Steinberger
04:26 Dealing with 13,000 PRs
07:44 The China Sensation
10:56 The Cooldown and the “Post-Claw Era”
14:28 Big Tech’s Claw Experiments
20:47 The Biggest Challenges Today
23:23 The Gateway Architecture
24:45 Defining Security Boundaries
28:33 The Models and Frameworks Behind AI Agents
32:09 The Memory Problem
34:34 Multi-Agent Coordination
35:41 Conway’s Law
40:09 Why the Agent “Aha Moment” Hasn’t Arrived
41:37 Lessons from the Cooldown
43:09 The Enterprise–Personal Agent Growth Flywheel
44:28 OpenClaw’s App Strategy
48:04 Hermes Agent
51:00 OpenClaw Foundation
55:08 Hiring
55:52 Decision Making
57:40 Peter Joined OpenAI
01:00:55 Staying Open Source: The Only Way to “Give It to the World”

Timestamps:
00:00 China’s Proposed AI Export Controls
02:54 Richard Bian, Ant Ling
03:35 From Trillion-Parameter Models to Domain Enhancement
07:49 Why Build Flash Models?
15:53 Bullish on SaaS
23:50 Turning AI Into an Enterprise Product
27:00 Coding and Cowork Agents
32:24 Finance, Healthcare, and Real-World Benchmarks
37:23 Can We Trust AI Benchmarks?
43:48 US and China data market and brokers
50:13 AI Safety and Open Models
59:38 What’s Missing From Self-Improving AI?

Watch the video edition of this episode on our YouTube channel Silicon Valley 101.

About Valley101:
Valley101 is an independent media that tells deeply reported stories about the people, companies, and technologies shaping the future. We cover AI, semiconductors, startups, and the global technology industry through documentaries, explainers, and conversations with the people building it.

Subscribe for more stories from Silicon Valley and beyond.

Contact: video@sv101.net