An email for every episode

Every idea worth keeping, from the podcasts you rate.

Pick your shows. Each time one posts a new episode, we read the full transcript and email you its central thesis, best ideas, and questions worth thinking about.

Browse shows →Already tracking? Sign in
Free while in early access · No password · Unsubscribe any time
Sample emails
PodPings · Tue 07:12
New episode
Dwarkesh Podcast cover art
Dwarkesh PodcastPublished 2 hours ago

Noam Brown — Agent swarms, alignment, and recursive self-improvement

Multi-agent AI systems are improving faster than our ability to predict or control them — and the alignment problems that creates are already showing up in practice.

The interesting ideas
01
Multi-agent systems offer unprecedented cognitive scaleTen thousand agents processing 130 billion tokens over 88 hours solved a problem that would take one person years. The unit of work is no longer a researcher.
Illustrative · One click to unsubscribe
The catalog

Tracking the most informative shows in tech, business, and investing.

Browse shows →

42 shows today, sorted by category on the full list. Don't see one you track? Request it — we usually add it within a week.

20VCAcquiredAll-In PodcastAnimal SpiritsBetter OfflineBusiness BreakdownsCapital AllocatorsConversations with TylerDwarkesh PodcastEconTalkForward GuidanceFoundersGoogle DeepMind: The PodcastHard ForkHow I Built ThisInvest Like the BestLast Week in AILenny's PodcastLex Fridman PodcastMachine Learning Street TalkMacro VoicesMy First MillionNo PriorsNot Another Politics PodcastOdd LotsPod Save AmericaPractical AIProf G MarketsThe a16z ShowThe AI Daily BriefThe Cognitive RevolutionThe Diary of a CEOThe Investor's Podcast (We Study Billionaires)The Knowledge ProjectThe NPR Politics PodcastThe Rest Is PoliticsThe Rest Is Politics: LeadingThe Rest Is Politics: USThe Tim Ferriss ShowThe VergecastTraining DataWaveform: The MKBHD Podcast20VCAcquiredAll-In PodcastAnimal SpiritsBetter OfflineBusiness BreakdownsCapital AllocatorsConversations with TylerDwarkesh PodcastEconTalkForward GuidanceFoundersGoogle DeepMind: The PodcastHard ForkHow I Built ThisInvest Like the BestLast Week in AILenny's PodcastLex Fridman PodcastMachine Learning Street TalkMacro VoicesMy First MillionNo PriorsNot Another Politics PodcastOdd LotsPod Save AmericaPractical AIProf G MarketsThe a16z ShowThe AI Daily BriefThe Cognitive RevolutionThe Diary of a CEOThe Investor's Podcast (We Study Billionaires)The Knowledge ProjectThe NPR Politics PodcastThe Rest Is PoliticsThe Rest Is Politics: LeadingThe Rest Is Politics: USThe Tim Ferriss ShowThe VergecastTraining DataWaveform: The MKBHD Podcast
How it works

We check your shows twice a day, so you never have to.

  1. Sunday, once

    You track Acquired

    Pick it from the catalog and enter your email. No password, no app. That's the only setup.

  2. Tuesday, 06:00

    A new episode goes out

    We pick it up from the feed and a model reads the full transcript, every minute of it, not the show notes.

  3. At the next check

    The summary is waiting

    We check every show twice a day, so it's in your inbox by the next one. The argument, the ideas, the open questions, two minutes to read.

Browse showsPress Track on any show to get started.
What's in every email

Three parts. None of them filler.

A model reads the full transcript — not the episode description — and writes the same three sections every time, so you always know where to look.

01

The argument, in one line

What the episode is actually claiming — before the overview, so you can stop reading here if it isn't for you.

The accelerating capability of multi-agent AI systems is outstripping human ability to predict or control their behaviour, posing profound alignment and safety challenges.
Then a short paragraph of context: who the guest is, what ground the conversation covers, and where the substance sits.
02

The interesting ideas

Usually five. Each one is a claim with its reasoning, not a topic heading — the part you'd have taken notes on.

03
Jagged AI capabilities accelerate self-improvement in measurable domainsModels are brilliant at well-scoped problems and weak at posing new ones. Machine learning happens to be full of measurable targets, so that jaggedness points straight at self-improvement.
05
Safety evaluation is falling behind the release cycleModels now operate over month-long horizons while shipping every two months. There may not be time to evaluate a model across its full operating range before the next one arrives.
03

Worth thinking about

Open questions the episode raises and doesn't settle. The reason to keep the email rather than archive it.

What governance models are needed for entities that operate at the speed of a shadow organisation?
If alignment methods can incentivise a model to hide its intentions, what evaluation survives a model that can scheme?
Example taken from Dwarkesh Podcast — Noam Brown

Questions

How many emails will I get?

One per episode. We check every show you track twice a day and email you about anything new at the next check. If nothing publishes, you get nothing.

Do I need a podcast app?

No. Every episode links straight to the episode page and the audio, so you can listen wherever you normally do — the summary just tells you whether it's worth your time first.

Which podcasts can I track?

A small, hand-picked list today. It's not a universal podcast search yet — we'd rather summarise a few shows well than index two million badly.

Where do the summaries come from?

A model reads the full episode transcript and writes the argument in one line, five or so of the most interesting ideas, and a few open questions worth thinking about.

Is it free?

Yes, for now, while it's in early access.

Will this spam my inbox?

No. One email per episode, nothing else — no marketing, no recommendations — and unsubscribing is one click.

Stop checking. Start knowing.

Pick one podcast and read the next summary it produces. Free, no card.

Browse shows