EchoLens

For investors, analysts and research professionals

Three hours of primary-source material.Minutes to understand what matters.

The interviews, podcasts and conference conversations where executives, founders and domain experts say something worth knowing are getting longer, not shorter. EchoLens reads one and gives you back its argument, the claims that argument rests on, who made each one, and the minutes actually worth watching — with the source moment attached to every line.

See EchoLens on a real sourceAnalyze a source of your own

A Lens that already exists is free to open, with no account. EchoLens is a research instrument: it records what was said and shows you where. It does not give investment advice and does not tell you whether a speaker is right.

What a source actually costs you

These are published Lenses, and the figures are read out of the stored analysis rather than written into this page. The bar is the whole source; the lit blocks are the moments that carry the argument, where they fall.

AI & TechnologyDwarkesh Patel1 hr 17 min

Dylan Patel – Two labs will soon control most of the world's workforce

Rapid exponential growth in AI infrastructure is driving extreme compute centralization into two frontier labs (OpenAI and Anthropic), triggering multi-trillion-dollar capital demands that will restructure global supply chains, raise macro interest rates, and provoke significant regulatory and geopolitical friction.

1 hr 17 min12 minof essential viewing · 16% of the source

The essential viewing path for this 1:16:53 source runs 12 min across 5 segments, or 16 percent of the source.

Moments
7
Figures
2
Forecasts
3
Companies named
8

Open this Lens →

  • Lex Fridman2 hr 26 min

    Jensen Huang: NVIDIA - The $4 Trillion Company & the AI Revolution | Lex Fridman Podcast #494

    2 hr 26 min25 minof essential viewing · 17% of the source

    The essential viewing path for this 2:25:59 source runs 25 min across 6 segments, or 17 percent of the source.

    Open this Lens →

  • Dwarkesh Patel2 hr 13 min

    Ryan Greenblatt – What happens once AI can automate AI research?

    2 hr 13 min6 minof essential viewing · 5% of the source

    The essential viewing path for this 2:12:32 source runs 6 min across 5 segments, or 5 percent of the source.

    Open this Lens →

Every published Lens is on Explore.

Live today

What you get out of one source

  1. 01

    The argument

    One statement of what the source is actually arguing, before any summary. You know inside a minute whether the next two hours are worth committing to.

  2. 02

    Numbers

    A figure a speaker puts on the record is typed as a numerical statement and pinned to the second they said it, with the speaker attached — so the number you half-remember is one click away instead of a second pass through a 90-minute interview.

  3. 03

    Predictions

    A forecast is labelled a forecast rather than blended into summary prose. What someone committed to, and when, stays separable from what they merely described.

  4. 04

    Recommendations

    What a speaker says should be done is kept apart from what they say is true. The distinction survives into the page rather than dissolving in a paragraph.

  5. 05

    Companies and entities

    The companies, institutions, technologies and people a source actually discusses, pulled out as a list — so a source can be triaged for relevance without watching it.

  6. 06

    The essential viewing path

    The smallest sequence of moments that still carries the argument, laid over the running time. When you do need to watch, you watch that.

  7. 07

    Source moments

    Every moment has its own link. Send a colleague the claim rather than the conversation, and they land on it with the player already cued.

Verifiability

Nothing asks you to trust the prose

Every claim on a Lens carries the four things that make it checkable, and the last of them is a link. This is the whole difference between structured intelligence and a confident paragraph.

  1. 01The claim
  2. 02Who said it
  3. 03The second they said it
  4. 04The source, cued there
Numerical statementDylan Patel1:05

Global capital expenditure on AI compute is expected to exceed $1 trillion this year and surpass $2 trillion by 2028.

Establishes the macro scale of capital flowing into AI hardware and infrastructure.

Open this moment at 1:05 →

Paraphrased from the source and recorded as a claim, not as a verified fact. The link opens the Lens with the player already cued to that second, and the original is one further click away.

Why not paste the transcript into a chatbot

You can, and for a rough gist it is fine. The difference shows up the moment you need to act on something — or defend it to somebody who was not in the room.

  • Pasting a transcript into a chatbot

    Works from whatever text you managed to paste.

    EchoLens

    Works from the source itself, resolved through the official API, with its real running time known.

  • Pasting a transcript into a chatbot

    Returns fluent prose in which a figure, a forecast and an aside all look alike.

    EchoLens

    Types every claim — figure, forecast, recommendation, assertion — and keeps the speaker attached to it.

  • Pasting a transcript into a chatbot

    Invents a timestamp as readily as it reports one.

    EchoLens

    Checks every timestamp against the real duration of the source, and refuses an analysis that fails.

  • Pasting a transcript into a chatbot

    Leaves you to find the moment again yourself.

    EchoLens

    Makes each claim a link that seeks the player to the second it was spoken.

Deep Lens · In development

Understand the source, then investigate it

What ships today is designed to answer what does this source say. Some questions need more than that — how the claims hold each other up, where a speaker contradicts themselves, what every figure across a source adds up to.

Deep Lens is being built. Nothing below is available yet, and none of it is included in what you can use today — it is listed here so you can see where this is going, not so you can plan around it.

  • Argument mapNot yet available

    How the claims in a source depend on one another, and which one the rest of it rests on.

  • Intelligence ledgerNot yet available

    Every figure, forecast and recommendation from a source consolidated into one register, rather than typed moment by moment as they are today.

  • Tensions and open questionsNot yet available

    Where a source contradicts itself, where an assumption is doing quiet work, and what it leaves unanswered.

  • Search and ask this sourceNot yet available

    Put a question to a source you have already analysed and get the answer with its moments attached.

  • Deep briefNot yet available

    The whole investigation as one document you could hand to somebody else.

What EchoLens will not do

It will not tell you a speaker is right. A model can accurately capture what somebody said without establishing that it is true, so claims are recorded as claims, marked unverified, and linked back to the moment they were spoken.

It is not investment advice, a recommendation, or a signal of any kind. It has no view on any security and produces none. What it produces is a structured reading of a public conversation.

It does not host, download or republish source media, and it publishes no transcript. Analysis runs against publicly available video through official APIs, and every moment points back at the original.

Questions

What is EchoLens?
EchoLens turns a long-form public video — an interview, a podcast, a conference talk, an earnings-adjacent conversation — into a structured read: the argument it makes, the claims that argument rests on, who made each one, and the moments worth watching. Every element links back to the second it came from.
Who is it for?
People who read primary sources for a living and are short of hours: public-market and private-market investors, analysts, portfolio managers, research teams, and operators who need to know what an executive or a domain expert actually said without spending an afternoon finding out.
How is this different from asking a chatbot to summarise a video?
A pasted transcript gives you fluent prose with nothing underneath it. EchoLens works from the source through its official API, types each claim as a figure, a forecast, a recommendation or an assertion, attributes it to a speaker, and pins it to a timestamp that is checked against the real running time of the source. Every claim on the page is one click from the moment it was spoken, so you can check it rather than trust it.
How much time does it save?
It depends entirely on the source, so EchoLens shows the real figure for each one instead of a headline percentage. Across the published corpus, a two-and-a-half hour conversation typically reduces to under half an hour of essential viewing — and the page states the exact ratio for the Lens you are reading.
Can I verify what it tells me?
That is the point of the design. Each claim carries its speaker, its type and its timestamp, and clicking it seeks the embedded source to that second. You can also open the original on YouTube at the same moment. Nothing asks you to take the analysis on faith.
Does EchoLens tell me whether a speaker is right?
No, and it will not. A model can accurately record what somebody said without establishing that it is true, so EchoLens marks claims as claims, leaves them unverified, and links them to the source so you can judge them. It is a research instrument, not a fact-checker and not an adviser.

Judge it on a source you already know

The fastest way to evaluate this is to open a Lens on a conversation you have already watched, and see whether it kept what mattered.

See EchoLens on a real sourceAnalyze a source of your own