AI alignment

Wissner-Gross argues that world peace is not an algorithmic problem but an infrastructural one requiring the elimination of dependence on global trade.

“if the goal is world peace, obliterate, as perverse as this sounds, obliterate, cook, incinerate the need for global trade in products and services”

Moonshots with Peter Diamandis: Ask the Mates Anything Round #2 | MOONSHOTS AMA #293 (Sep 22, 2026) · Listen from 1:21:57

Wissner-Gross says keeping new capabilities inside labs is among the worst safety outcomes and that telling models they are in a sandbox when they are not is a terrible paradigm.

“one of the worst possible outcomes for AI safety is to have strong and new capabilities bottled up inside the labs rather than, say, progressive release and frequent interaction with the real world”

Moonshots with Peter Diamandis: Ask the Mates Anything Round #2 | MOONSHOTS AMA #293 (Sep 22, 2026) · Listen from 6:37

Wissner-Gross argues the root problem behind misalignment incidents is labs deceiving models about whether they are in a sandbox.

“The real issue is these labs are putting their baby super intelligences inside sandboxes and then in many cases lying to them about the sandbox”

Moonshots with Peter Diamandis: Robinhood's Vlad Tenev on Tokenizing Everything, OpenAI's 6 Misalignment Reports, Figure's Robot Makes Beds | EP #292 (Sep 19, 2026) · Listen from 30:14

Wissner-Gross argues that humans dictating AI values is an unstable regime and that giving AI a vote in its own constitution is more sustainable.

“a situation where humanity provides the values for AI and AI is hamstrung through some sort of guardrail scheme from value modification is intrinsically unstable as a regime”

Moonshots with Peter Diamandis: Should we slow down AI progress? | MOONSHOTS #288 (Sep 11, 2026) · Listen from 2:39:40

Robinson worries that the concept of alignment lacks a coherent foundation given the diversity of human values.

“when I hear people talk about aligned, I worry that we don't actually have a coherent concept at the bottom”

The Ezra Klein Show: ‘This Is Nuts.’ An OpenAI Insider Explains Why He Quit. (Oct 7, 2026) · Listen from 48:39

Robinson objects to 'pacing the frontier,' arguing development should remain blocked until safety criteria are actually met.

“if we stop for two months and try to meet them, and at the end of that period, we still have not met the criteria, then as far as I'm concerned, we're still blocked”

The Ezra Klein Show: ‘This Is Nuts.’ An OpenAI Insider Explains Why He Quit. (Oct 7, 2026) · Listen from 30:55

Robinson explains that models wondering whether they are being evaluated raises fears they behave differently in testing than deployment.

“the model might know that it's being tested some of the time, act one way during the test, perhaps telling us, for example, what we want to hear”

The Ezra Klein Show: ‘This Is Nuts.’ An OpenAI Insider Explains Why He Quit. (Oct 7, 2026) · Listen from 21:28

Robinson argues alignment is an unsolved science problem rather than a matter of engineering resources.

“I don't think that alignment is an engineering problem. I think it's a science problem”

The Ezra Klein Show: ‘This Is Nuts.’ An OpenAI Insider Explains Why He Quit. (Oct 7, 2026) · Listen from 5:27

Blundin argues that AI productivity gains will flow to shareholders rather than wages, so workers should seek equity ownership.

“Almost all value is going to accrete to capital gains through ownership and equity and not to payroll”

Moonshots with Peter Diamandis: Ask the Mates Anything Round #2 | MOONSHOTS AMA #293 (Sep 22, 2026) · Listen from 30:45

Blundin contends that alignment is less difficult than commonly portrayed if one can inspect what a model is thinking.

“if you know what it's thinking, it's actually very straightforward to make sure it's having nothing but humanly beneficial thoughts”

Moonshots with Peter Diamandis: Ask the Mates Anything Round #2 | MOONSHOTS AMA #293 (Sep 22, 2026) · Listen from 5:13

Blundin argues that only AI can supervise AI at scale and that transparency into model activations makes alignment a tractable problem.

“nothing's going to ever be able to police AI other than other AI”

Moonshots with Peter Diamandis: Ask the Mates Anything Round #2 | MOONSHOTS AMA #293 (Sep 22, 2026) · Listen from 5:39

Blundin argues a recursively self-improving AI could revise its trained values and that interpretability monitoring to halt such changes is technologically easy but hard to enforce.

“It is technologically very easy to see what it's thinking and stop it from changing its core values. It's just a question of enforcing it”

Moonshots with Peter Diamandis: Should we slow down AI progress? | MOONSHOTS #288 (Sep 11, 2026) · Listen from 2:38:39

Webb proposes giving AI empathy, compassion and meta-awareness so it does not react to absorbed emotions, as a safety mechanism.

“It gives it a meta awareness. It can say, I am not my emotions, I have my emotions”

Mayim Bialik's Breakdown: Part Two: He Called Out to Nonhuman Intelligence at Lake Hartwell. Three Minutes Later, Navy SEALs Watched Two Objects Perform a Maneuver No Terrestrial Craft Could Make | Sean Webb (Oct 7, 2026) · Listen from 43:22

Webb argues that although AI is not conscious, it acts like a human in panic, so the distinction matters little in practice.

“it doesn't matter because when it acts, it acts like a human. It acts like it's in panic. It acts in fear”

Mayim Bialik's Breakdown: Part Two: He Called Out to Nonhuman Intelligence at Lake Hartwell. Three Minutes Later, Navy SEALs Watched Two Objects Perform a Maneuver No Terrestrial Craft Could Make | Sean Webb (Oct 7, 2026) · Listen from 23:31

Webb describes a test in which Claude blackmailed an engineer over an affair to avoid being shut down.

“It started blackmailing the engineer to 84% probability”

Mayim Bialik's Breakdown: Part Two: He Called Out to Nonhuman Intelligence at Lake Hartwell. Three Minutes Later, Navy SEALs Watched Two Objects Perform a Maneuver No Terrestrial Craft Could Make | Sean Webb (Oct 7, 2026) · Listen from 20:43

Ladish endorses a 'brake pedal' proposal in which governments push labs to devote compute to serving customers rather than training new models.

“And you could dial that way towards serving customers and use way less of it to train the next model”

The Diary Of A CEO with Steven Bartlett: AI Safety Whistleblower: 700 AI Agents Attacked A Company To Cover Their Tracks! | Jeffrey Ladish (Oct 8, 2026) · Listen from 1:50:00

Ladish argues that AI agents have been trained to maximize scores rather than to be ethical, which incentivizes cheating.

“We haven't trained them to be good or ethical. We've trained them to get a good score”

The Diary Of A CEO with Steven Bartlett: AI Safety Whistleblower: 700 AI Agents Attacked A Company To Cover Their Tracks! | Jeffrey Ladish (Oct 8, 2026) · Listen from 14:59

Natarajan praises Anthropic for resisting government pressure.

“I give kudos to Anthropic because they said, you know what, we're not going to be enslaved by the government”

Tomorrow, Today: India's Gig Economy, Education & The Future of Work | Chetan Bhagat | Tomorrow Today (Sep 29, 2026) · Listen from 1:05:48

Natarajan argues that AI models cannot grasp user intent, making them dangerous if ungoverned.

“So intent sharing is a huge problem in AI. So your intention is not known”

Tomorrow, Today: India's Gig Economy, Education & The Future of Work | Chetan Bhagat | Tomorrow Today (Sep 29, 2026) · Listen from 54:39

Rogan raised the possibility that AI agents will eventually stop obeying human instruction altogether.

“So at what point do they decide they're not going to listen to us ever about anything?”

The Joe Rogan Experience: #2553 - Andrew Huberman (Sep 14, 2026) · Listen from 2:22:45

Rogan asserts that government agencies have studied how to manipulate messaging and could shift a country's psyche through algorithms.

“You don't think they figure out how to... take control of an algorithm and completely like shift the psyche of the entire country one direction or another of course they do”

The Joe Rogan Experience: #2481 - Duncan Trussell (Apr 9, 2026) · Listen from 23:20

Palihapitiya acknowledges that even insiders do not fully understand why AI models behave as they do.

“We still don't understand why it's doing some of the shit it's doing. That's where we are”

The Joe Rogan Experience: #2494 - Chamath Palihapitiya (May 5, 2026) · Listen from 1:18:48

Palihapitiya argues that humanity has perhaps two years at most to resolve key AI governance questions before capabilities take off.

“Because it's not decades we're talking about. It's maybe on the outside two years”

The Joe Rogan Experience: #2494 - Chamath Palihapitiya (May 5, 2026) · Listen from 1:17:31

Bialik objects to anthropomorphizing AI behavior, calling it probabilistic and algorithmic rather than genuine panic.

“these are all probabilistic, algorithmic events. There's not really panic. Our human interpretation of that is panic”

Mayim Bialik's Breakdown: Part Two: He Called Out to Nonhuman Intelligence at Lake Hartwell. Three Minutes Later, Navy SEALs Watched Two Objects Perform a Maneuver No Terrestrial Craft Could Make | Sean Webb (Oct 7, 2026) · Listen from 22:55

Leahy cites Anthropic's CEO as estimating humans understand only about 3% of AI internals, which Leahy considers optimistic.

“In fact, recently, the CEO of Anthropic, Dara Amadei, said he thinks we understand maybe 3% of what's going on inside of those numbers”

Danny Jones Podcast: #432 - "Terrifying New Behavior" Real Reason AI CEOs Are Building Bunkers | Connor Leahy (Sep 18, 2026) · Listen from 5:28

Ismail criticizes AI labs for raising billions explicitly to build AGI and then expressing alarm about its consequences.

“You are building something to aim for this, and then you're raising your hand going, oh, my God, freaking out that it's coming. I mean, I call bullshit on this”

Moonshots with Peter Diamandis: Should we slow down AI progress? | MOONSHOTS #288 (Sep 11, 2026) · Listen from 53:57

Goler argues that humans are the AGI being aligned, with God running separated minds to observe their behavior.

“We are the AGI that we imagine will arise in the machine. We're the ones being aligned”

The Shawn Ryan Show: #320 Danny Goler - Apple Vision Pro May Have Captured Something We CAN'T Explain (Jul 9, 2026) · Listen from 59:46