The Open Source AI Question (Robert Wright & Nathan Lambert)

August 14, 2025·1 hr·4 clips

Lambert says Geoffrey Hinton's 'nuclear weapons at Radio Shack' warning about open-source AI misunderstands how cheap model training has become.

The conversation starts slowly and procedurally. Wright introduces Non-Zero, identifies Lambert as someone who builds open-source large language models at the Allen Institute for Artificial Intelligence, and works through the institutional backstory before the argument really sharpens. Lambert explains that the Allen umbrella is broader than AI2, with biology and artificial intelligence sitting under related Paul Allen-funded institutions. Wright then puts a familiar worry on the table: society may not adapt quickly enough to AI, whether that means unemployment, institutional stress, or geopolitical instability. Lambert steers the concern somewhere more concrete. He is less worried that hidden model knowledge will later need fixing than that a few already huge technology firms could get much more powerful. He names Nvidia, OpenAI, Google, and Apple as examples of companies that could become 10 or 20 times bigger if AI's upside pools in the same hands. Wright follows that thread patiently. If AI turns winner take all, the issue may not be three or four dominant companies. It could be one dominant voice. Speech enters the conversation there. Lambert points to XAI as a leading example where one person's preferences seem unusually close to the system's public voice. The exchange stays calm. Wright keeps translating the technical and institutional problem into the plain version of the question, then gives Lambert room to adjust it. The answer leans toward distribution over monopoly. Open models are not treated as a purity test. They are a way to spread capability and make concentrated control harder. Near the end, Lambert mentions Interconnects and the Atom Project, then says he wants AI to be easier to understand without turning everything into hype or fear. That middle ground gives the episode its shape. The strongest moments are not big predictions. They are the quieter parts about who gets power if AI really works.

As heard by us

A clear AI conversation about open source, concentrated corporate power, and who gets to shape the models people may come to rely on.

Robert Wright frames open-source AI as a political economy fight, not just a developer argument. Nathan Lambert, who builds open-source models at AI2, is less interested in forcing models toward the right point of view than in what happens when the most capable systems sit…

Read the full review in PlayNext →

Why you'd press play

A calm case for why open-source AI matters when a few companies get so powerful.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from The Wright Show