Apr 26, 2025 · 38m · tbpn
Building the World’s Smallest Fusion Reactor | Robin Langtry on TBPN
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
This episode features in-depth interviews with Robin Langtry, CEO of Avalanche Fusion, discussing compact fusion reactors and regulatory reform, and Russ d'Sa, CEO of LiveKit, exploring the real-time networking infrastructure powering conversational voice AI.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 18.8% of the talking time here. How this is scored →
The hosts as informed peer 4.1 Guest teaching 4.8 Guest disagreement 0.3 The hosts pushing back 0.1
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 23:48 Rejection of the premise that hallucinations are universally bad
Russ d'Sa challenges the conventional negative framing of hallucinations, explaining that in conversational voice AI, generative hallucination prevents dead-end user experiences.
Hardest push from the hosts ▶ 22:29 Host presses on persistent mobile speech-to-text flawsJohn Coogan directly challenges the guest to steelman why mobile speech-to-text remains frustratingly poor despite advancements.
Biggest teaching moment ▶ 4:32 Robin reveals tritium levels in everyday exit signsRobin Langtry surprises the hosts by demonstrating that consumer exit signs carry 25 curies of tritium, proving the disproportionate nature of the NRC's 10-curie reporting threshold.
The host holds their own ▶ 26:17 John cites transformer ASICs and Etched hardware architectureJohn Coogan displays strong technical knowledge by comparing potential voice model silicon compilation to Etched's transformer chips and Bitcoin ASICs.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Avalanche Fusion and The Fusion Works Facility | 5 | 3 | 0 | 0 | John Coogan makes an insightful connection to Hermeus building its own hypersonics heat test facility to alleviate testing bottlenecks. Robin Langtry agrees that commercial test facilities commoditize complements across deep tech. | |
| Zero-Based Regulatory Budgeting and Tritium Rules | 1 | 8 | 1 | 0 | Robin Langtry completely schools the hosts on radiation regulation by revealing that exit signs contain 25 curies of tritium while NRC rules require incident reports for 10 missing curies. John Coogan repeatedly admits he had no idea. | |
| Fusion Milestones and State-Level Accelerator Licensing | 5 | 5 | 0 | 0 | John Coogan accurately frames the regulatory risk versus science risk tradeoffs between fission and fusion. Robin details the distinction between NRC fission reviews and state-level Agreement State accelerator licensing. | |
| State Lawsuits Against the NRC for SMRs | 4 | 4 | 0 | 0 | Robin recounts being present when Rick Perry advised fission founders to sue the NRC over SMR regulation. The hosts note the positive-sum and government-focused legal climate compared to B2B SaaS. | |
| Fast Hardware Iteration and Lessons from NewSpace | 6 | 4 | 0 | 0 | John Coogan probes rapid iteration lessons from SpaceX and Blue Origin and brings up university-based TRIGA reactors. Robin explains achieving one-week design-to-vacuum test cycles for fusion physicists. | |
| Inherent Physical Safety of Fusion Devices | 3 | 6 | 1 | 0 | John Coogan asks whether small fusion machines could be weaponized like Oppenheimer-era bombs. Robin firmly explains the physics of vacuum maintenance, illustrating that any physical damage simply extinguishes the plasma safely. | |
| LiveKit's Journey from Open Source to Voice AI | 2 | 4 | 0 | 0 | Russ d'Sa introduces LiveKit's transition from open-source WebRTC video/voice infrastructure to powering OpenAI's ChatGPT Advanced Voice Mode after a hackathon demo. | |
| Hallucination in Voice AI and Latency Reductions | 6 | 6 | 2 | 1 | John Coogan demonstrates domain knowledge from working at Vlingo/Dragon in 2010 while asking why STT still struggles. Russ reframes hallucination as a critical conversational feature compared to rigid deterministic assistants. | |
| Customizing Voice Pacing and Multimodal Interaction | 7 | 5 | 0 | 0 | John Coogan demonstrates sharp technical depth by referencing transformer silicon ASICs like Etched and suggesting inference-time fine-tuning for conversation speed. Russ explains Cerebras demo latency and native audio model controls. | |
| The Voice AI Ecosystem and Enterprise Workflows | 3 | 5 | 0 | 0 | Russ breaks down the voice AI startup ecosystem between consumer conversational assistants and high-volume voice-native enterprise workflows like intake and customer support. | |
| Real-Time Synchronized Streaming with Playback | 5 | 6 | 0 | 0 | John Coogan brings up Playback's live sports platform from a prior interview. Russ educates the hosts on broadcast TV latency drift and how LiveKit provides synchronized sub-second video feeds. | |
| LiveKit's $45M Series B and Platform Vision | 2 | 1 | 0 | 0 | Russ announces LiveKit's $45 million Series B led by Altimeter Capital, and the hosts celebrate the company's multi-year overnight success. |
Statements from this episode (22)
Langtry: Avalanche Fusion's Lab Reactor Has a 12cm Plasma Core
“The fusion machines we're building in our lab today are about 12 centimeters in diameter. That's the plasma core.”
Avalanche Fusion Announces The Fusion Works Test Facility in Eastern Washington
“The announcement that we're doing today is we're announcing our new test facility, which is going to be in Eastern Washington in the Tri-Cities area called the Fusion Works. And so what the Fusion Works is, is it's basically a facility where you can come test …”
Langtry: Hermeus is opening its hypersonic test facility to the DoD
“And so the Hermia's announcement, I think it was a couple of weeks ago now, was basically they are opening up their test facility to the rest of the DoD because there's a shortage of hypersonics testing.”
Langtry: NRC mandates incident report for 10 unaccounted curies of tritium
“The NRC has this rule that if you can't account for 10 curies of tritium in your system, You have, like, 30 days to report it as an incident to the NRC that you've, like, lost 10 curies of tritium.”
Langtry: Standard glowing exit signs contain roughly 25 curies of tritium
“And to give you an idea of how small that is, an exit sign has, like, 25 curies of tritium.”
Langtry Predicts Magnetic Fusion Will Achieve Net Energy By 2029
“There's a chance that one or multiple magnetic confinement fusion companies are going to cross Q greater than one during this administration.”
Langtry: Fusion reactors are regulated like particle accelerators by agreement states
“So right now there was a decision about a year ago that
fusion machines are going to be licensed like particle accelerators and specifically on the radioactive materials they produce.
And so what that does is that pushes the regulator you work with to the s…”
Langtry: Texas and Utah are suing the NRC over SMR regulation
“Texas and Utah that are suing the NRC to try and get regulatory control over small modular reactors.”
Langtry: Fission will scale to gigawatts while fusion dominates kilowatt-to-megawatt range
“I think my personal belief is that fission is going to go bigger, like hundreds of megawatts to gigawatts, and fusion is going to be really great from the kilowatts to a few megawatts.”
Langtry: Fusion power will be ideal for defense autonomous vehicles
“Fusion is going to be really great for mobility and stuff. So like autonomous vehicles, fusion powered, like an Android dive XL or something like that. Like, I think that's going to be a really great application for fusion in defensive space.”
Langtry: Oversized rockets bottleneck casting suppliers and stretch iteration to nine months
“If your rocket is too big, all of a sudden, there's only, like, three suppliers in the whole country where you can get castings made, and now you're at a nine-month, every iteration takes nine months, right? And it's just, it's too slow.”
Langtry: Avalanche has achieved one-week idea-to-test hardware cycles
“We said at the beginning of Avalanche that we want a physicist to be able to go from, like, I have an idea, to testing it in a week, and we've actually achieved that a couple times. Like, someone had an idea on Monday, they worked with the mechanical team to d…”
Langtry: US universities operate TRIGA nuclear reactors run by licensed undergrads
“They're all over the country. Oregon State U has one. I've been there, like I stood on top and like looked down at like the fission. Washington State. WSU has one. Yeah. Last question for me. They're called trigger reactors. So they're used basically for nucle…”
Langtry: Fusion reactors require deep vacuum, preventing runaway reactions
“No, it's a vacuum chamber, right? And it needs to have very, very deep vacuum to maintain the plasma and do the fusion.”
Langtry: Fusion reactors will likely deploy in 1-to-5 megawatt shipping containers
“The practical one is they're probably gonna be containerized, and you're probably gonna just be shipping containers all over the place, like, powering a neighborhood, you know, that's, like, probably what's gonna happen. Everyone wants this containerized one t…”
LiveKit's d'Sa: LiveKit built the infrastructure for OpenAI's ChatGPT voice modes
“Fast forward, like, about a year and a half, two years, and, ah, we built a demo when ChatGPT first came out, ah, where you could talk to it instead of text with it using LiveKit's infrastructure, and a few months later, OpenAI ended up finding it, ah, wanted …”
Russ d'Sa: AI hallucination is a feature in voice interfaces
“People talk about hallucinations as this bad thing, but in a lot of ways, for like, kind of this speech-to-speech interaction model, where you're talking to an AI, Hallucination is a feature.”
Russ d'Sa: LLM inference is now faster than text-to-speech generation
“Now like LLM inference can actually be done in less time than generating speech with TTS.”
Russ d'Sa: 300 to 500 ms turn latency crosses voice AI uncanny valley
“Getting that latency end to end down, you know, to 300 milliseconds or 500 milliseconds on average for, like, turn latency that, that, that kind of helps you cross this uncanny valley for voice AI.”
Russ d'Sa: AI models will implicitly match user pacing within two years
“Then there's another part of it, which I think will come in the next, you know, year or two where the model will implicitly be intelligent enough to pick up on what your pacing is or your state of mind is just based on the way that you're talking and expressin…”
d'Sa: Modern live sports broadcasting is not true real-time synchronization
“And so the technology that we use for live broadcasts of sports games is not true real time technology. Not everyone is synchronized in the same way that in 1969 when the, you know, the lunar landing everybody saw that at the same time because there's a server…”