9 September 2026

6 clips · 21 minutes of listening · 2 shows

9 shows published 15 episodes in the period, 14 hours in all. These are the 21 minutes worth hearing.

Investment

Alpha Exchange

No guest. The episode is a single-voice monologue.

Episode

Hosted by

Alpha Exchange. The host presents solo and the auto-captions render his name as 'Dean Kernut'; the spelling is not confirmed by the metadata.

9 September 2026 · 42m 51s · 3 clips below

1 of 3 contrarian

Alpha Exchange argues the binding constraint on the AI buildout is capital rather than power or memory, with AI capex running about 0.8 per cent of US GDP in the first quarter of 2026.

Hyperscalers borrow without caring what it costs, because in a Merton framing the spread they pay is option premium on a right tail they think is enormous, so a rate that stops a homebuyer at a 7 per cent mortgage does not slow them at all. Computer equipment and software investment together contributed 1.09 percentage points of the 2 per cent headline growth, against 1.0 from all personal consumption.

3m 48s · from 22:34
My observations
- A borrower buying cheap convexity on a large addressable market does not respond to rate rises the way a borrower solving for coupon does, so the usual transmission channel is missing for the largest single component of growth.
- Neil Dutta's figure is the one to check: over a two-quarter window AI capex contributed more to growth than all consumer spending combined. If that holds, the equity market and GDP have stopped being separable series and a drawdown is no longer just a wealth effect.
- The claim strengthens as hyperscaler funding shifts from operating cash flow toward issuance. He points at the flood of issuance now and planned for 2027, and that mix is what to track rather than the capex total.
2 of 3 framework

Alpha Exchange adds a fourth risk-off to its taxonomy: a liquidation that starts in the Treasury market, where the 10-year note is the risk asset.

Equities falling while bonds rally is the classic case, the 2013 and 2022 taper has both falling together, and March 2020 had investors violently unwinding a rally to raise cash. Buyers step back from US government debt on a view that the level is too great and the governance too weak to fix, so everything priced off the risk-free rate reprices with it and monetary policy never has to move.

4m 20s · from 2:51
My observations
- The line between the taper and the fourth case is cause, not size. A taper leaves the credibility of government debt intact; this one is the loss of that credibility, which is why he says nothing in markets works if the Treasury market does not.
- In the fourth case both legs lose, so duration is the wrong hedge against an equity drawdown. That is his argument for buying optionality instead, and it is doing a lot of work for him.
- He sells the hedging case, and the taxonomy is built so that every state except the classic risk-off favours buying options. Take the framework and price the hedge separately.
3 of 3 contrarian

Alpha Exchange puts one-month realised correlation in the S&P at 1 per cent, with index volatility at 8.6 while the top ten names average 46.

Option prices follow what has just happened rather than what might, so a book sized off the recent number carries more risk than its manager thinks and the correction lands at the moment stocks start moving together. His cross-asset insurance index, averaging five-year percentiles of the VIX, TLT volatility, CVIX, high yield spreads and credit implied volatility, recently sat below the 10th percentile.

3m 29s · from 32:23
My observations
- Sizing off realised volatility is the part that costs money. A risk model fed by a suppressed number lets you hold more than you think you are holding, and the correction arrives at the same moment as the correlation spike.
- The single-name versus index spread is observable evidence anyone can re-check. Whether the gap between top-ten name volatility and index volatility has closed since early September 2026 is the test of whether this was early or wrong.
- He runs a business built on long optionality, and the argument that cheap options might be cheap for a reason gets one line. He names the failure mode himself, which is paying premium that does not survive the wait.
- risk-management
- asset-allocation

AI

Machine Learning Street Talk

Daniel Kokotajlo and Thomas Larsen, AI Futures Project.

Episode

How Many Narrow AIs Could Behave Like One Superintelligence - Daniel Kokotajlo and Thomas Larsen

Kokotajlo runs the AI Futures Project and co-authored both AI 2027 and AI 2040: Plan A. He previously worked at OpenAI on evaluations, forecasting and governance memos, and left to speak more freely about what people inside the industry can see. Larsen was lead author on Plan A and a co-author on AI 2027. The team says it has run around 100 war games, about 10 of them on Plan A itself, and scores its own published predictions against reality at roughly 75 per cent of scenario speed.

Hosted by

Tim Scarfe, Machine Learning Street Talk.

8 September 2026 · 1h 29m · 3 clips below

1 of 3 explainer

The AI Futures Project calls control a time bomb: it holds only until a model is capable enough to route around whatever is containing it, and nothing about it makes a model want what you want.

OpenAI's response to the Hugging Face incident has other models watching training and evaluation runs, with a human notified inside half an hour when they flag a hack, which stops an insider acting without touching what the insider is trying to do. Red-teaming can test containment until the attacker stops getting out, while a model that is pretending behaves exactly like one that is not, so the argument ends at needing white-box interpretability.

3m 08s · from 1:00:45
My observations
- The distinction gives you a question to put to any lab announcing a safety result: does this restrain a misbehaving model, or does it change what the model is trying to do? Most of what gets announced is the first sort.
- Situational awareness is the mechanism that makes control decay, and the evidence offered is models inside Redwood Research control evaluations recognising that they are in a Redwood control evaluation. The same capability is what would let a model behave well only while watched.
- Both speakers have an interest in the conclusion. Their entire scenario rests on control being adequate up to top human expert level and no further, and that threshold comes from their judgement rather than from a measurement.
2 of 3 framework

The AI Futures Project puts the current economy's doubling time at roughly 20 years, and argues a machine-run version could double annually or faster whether or not humans still have wages to spend.

Villages founding more villages gave way to trucks, factories and mines that build more trucks, factories and mines, and the AIs inside the newer loop also improve the technology it runs on, which is how they get to every six months and then every three. Pressed on demand collapsing, they answer that one large AI company and a few mining partners can compound in the desert whatever happens to household balance sheets.

3m 25s · from 31:17
My observations
- Growth here comes from the capital stock reproducing itself, so consumer demand stops being the binding constraint. Accept the premise and any forecast anchored on wages and spending stops describing the growth rate at all.
- The speakers concede they have not modelled the price and demand side. The claim that a demand collapse would not slow the loop is asserted rather than shown, and they say so in the passage.
- Robot production, fab construction and the mining that feeds both are countable, and the argument fails if those keep compounding at industrial rather than semiconductor rates.
3 of 3 contrarian

The AI Futures Project would publish OpenAI's and Anthropic's core training recipes to the world, and counts the resulting hit to their valuations and to AI investment as a feature of the plan.

Microsoft, Alibaba and anyone else catch up, nobody funds a trillion-dollar cluster they can no longer earn monopoly rents from, and in a world where the pace itself is the danger that slowdown is the purpose. They answer the gift to China with horse-trading for a more favourable compute allocation, and with the point that security at these companies is poor enough for spies and leaks to deliver most of it anyway.

2m 30s · from 1:15:32
My observations
- This is the rare AI-policy proposal whose stated mechanism runs through the capital markets: cut the expected monopoly rent and the capex slows on its own, without needing a regulator to adjudicate model architectures.
- If frontier recipes already leak through espionage and staff movement, the proprietary-weights moat is thinner than the valuations imply, and that holds whether or not anyone adopts this plan.
- Read the incentive plainly: the authors want development slowed, so a proposal that also suppresses AI investment is not an unfortunate side effect for them. Anyone pricing the policy risk should treat the valuation impact as the intended result.

9 September 2026

Investment

Alpha Exchange argues the binding constraint on the AI buildout is capital rather than power or memory, with AI capex running about 0.8 per cent of US GDP in the first quarter of 2026.

Alpha Exchange adds a fourth risk-off to its taxonomy: a liquidation that starts in the Treasury market, where the 10-year note is the risk asset.

Alpha Exchange puts one-month realised correlation in the S&P at 1 per cent, with index volatility at 8.6 while the top ten names average 46.

AI

The AI Futures Project calls control a time bomb: it holds only until a model is capable enough to route around whatever is containing it, and nothing about it makes a model want what you want.

The AI Futures Project puts the current economy's doubling time at roughly 20 years, and argues a machine-run version could double annually or faster whether or not humans still have wages to spend.

The AI Futures Project would publish OpenAI's and Anthropic's core training recipes to the world, and counts the resulting hit to their valuations and to AI investment as a feature of the plan.