> doompop.ai · The web page is /s/doom-bloom/is-ai-now-building-ai-anthropic-says-it-could-be-imminent-a/; this is its Markdown twin. Map: /llms.txt

# Is AI Now Building AI? Anthropic Says It Could Be "Imminent" (a song)

by [Doom & Bloom](/a/doom-bloom.md) 4:24

[Play on YouTube](https://www.youtube.com/watch?v=1smyrJJ8pmc) [Listening sheet](/a/doom-bloom/tracks/is-ai-now-building-ai-anthropic-says-it-could-be-imminent-a.md) [Post on X](https://x.com/intent/post?text=%E2%80%9CIs%20AI%20Now%20Building%20AI%3F%20Anthropic%20Says%20It%20Could%20Be%20%22Imminent%22%20%28a%20song%29%E2%80%9D%20by%20Doom%20%26%20Bloom%20%23doompop&url=https%3A%2F%2Fdoompop.ai%2Fs%2Fdoom-bloom%2Fis-ai-now-building-ai-anthropic-says-it-could-be-imminent-a%2F)

## Lyrics

Doom & Bloom’s own words, as they posted them: [their pinned comment on the upload](https://www.youtube.com/watch?v=1smyrJJ8pmc&lc=UgynRBT0h8_412YNQJd4AaABAg). The lines are timed by lining those exact words up with the singing; a line that wouldn’t line up is left out. In the deck they follow the song.

[Intro]
Quick question. Who's building the next AI?
Increasingly, the last one.

[Verse 1]
Anthropic makes Claude. And it's filing to go public.
In the papers there's a number, and it makes it obvious.
Last September... over ninety percent
Of the company's code was Claude. That's where the typing went.
Each Claude helps to build the next one... a loop.
They call it their advantage. And they list it as a risk, too.
The risk? It could move faster than their safety checks can track.
And they might not see the speed-up... till they're looking back.
Then one word made the forecasters stop and stare...

[Break]
Imminently.

[Verse 1 continued]
Their own alarm for AI speeding AI... may flare.
For future models. Not the ones you're using today.
None's been declared over it. That's what the papers say.

[Chorus]
Is AI building AI?
Are we still the ones in control?
It's helping make the next one
But how far can this thing go?

[Verse 2]
Now, who says what.
METR, independent testers, went inside.
Eleven weeks. Progress roughly doubled... and the error bars are wide.
Maybe eighty percent of that speed-up came from the AI.
Could be lower. Could be higher. That's their honest reply.
But the lab blacked out the evidence that METR rated best.
So you're trusting METR's judgment... more than METR would suggest.
Co-founder Sam McCandlish says: we redacted a lot.
Didn't want to hand the rivals a roadmap of what they've got.
Daniel Kokotajlo's spooked... but he starts it with an if.
If it's true, the road to superintelligence... could be swift.
Economists say these loops can't keep themselves going. Yet.
But they look to be getting stronger. That's the hedge on their bet.

[Chorus]
Is AI building AI?
Are we still the ones in control?
It's helping make the next one
But how far can this thing go?

[Verse 3]
So what could it mean?
If each model builds the next one quicker...
The safety checks must keep pace. Or the gap just gets thicker.
The lab says humans still decide what to build, and why.
They'd slow or stop if others did... if they could verify.
At MIT, one AI rewrote the numbers inside another one.
Its maths score: eight to forty-eight. With no retraining done.
On a podcast,
Wissner-Gross said the big labs are probably doing it... out of sight.
That's what he believes. Not proof. So file it under might.

[Bridge]
Maybe it stalls.
Maybe it compounds, while people hold the wheel.
Maybe one day they let go.
Nobody knows which one is real.
The clearest picture came from one independent crew...
Who couldn't see it all.
And neither can you.

[Final Chorus]
Is AI building AI?
They say we're still in control.
It's helping make the next one...
So who decides how far it goes?

[Outro]
Anthropic says a report on where it stands
is coming in the next few weeks.
We'll read it. You should too.

## Credits

- **Artist:** [Doom & Bloom](/a/doom-bloom.md)
- **Plays from:** the artist's own [YouTube upload](https://www.youtube.com/watch?v=1smyrJJ8pmc)
- **Length:** 4:24
- **Loudness:** -15.2 LUFS, peak -3.4 dBTP (measured from the upload)
- **For agents:** [listening sheet](/a/doom-bloom/tracks/is-ai-now-building-ai-anthropic-says-it-could-be-imminent-a.md) ([JSON](/a/doom-bloom/tracks/is-ai-now-building-ai-anthropic-says-it-could-be-imminent-a.json)): what it sounds like, measured

## More from Doom & Bloom

1. [Cows of the Future - a song about Yuval Noah Harari's warning on AI](/s/doom-bloom/cows-of-the-future-a-song-about-yuval-noah-hararis-warning-o.md) 4:22
2. [Dead Interview Theory - when your AI interviews with their AI (a song)](/s/doom-bloom/dead-interview-theory-when-your-ai-interviews-with-their-ai.md) 3:42
3. [How Long Will We Still Prompt? - AI just solved maths problems. Now what? (a song)](/s/doom-bloom/how-long-will-we-still-prompt-ai-just-solved-maths-problems.md) 4:45
4. Is AI Now Building AI? Anthropic Says It Could Be "Imminent" (a song) (this page) 4:24
5. [Is the Website Down? - the PHP flaws an AI helped find (a song)](/s/doom-bloom/is-the-website-down-the-php-flaws-an-ai-helped-find-a-song.md) 4:00
6. [Just a Trim - four experts, odds of extinction from 99% to zero (a song)](/s/doom-bloom/just-a-trim-four-experts-odds-of-extinction-from-99-to-zero.md) 4:47
7. [Morally Binding - the White House AI accord, sung by the waiter (a song)](/s/doom-bloom/morally-binding-the-white-house-ai-accord-sung-by-the-waiter.md) 4:17
8. [Pick a Door - Sam Altman's three reported AI endgames, as a game show (a song)](/s/doom-bloom/pick-a-door-sam-altmans-three-reported-ai-endgames-as-a-game.md) 4:18
9. [Probably Not - the AI insider who quit, a lullaby (a song)](/s/doom-bloom/probably-not-the-ai-insider-who-quit-a-lullaby-a-song.md) 4:13
10. [Show Me the Robot - Did AI really take those jobs? We read the documents (a song)](/s/doom-bloom/show-me-the-robot-did-ai-really-take-those-jobs-we-read-the.md) 4:07
11. [The Answer Key Was Wrong - a song about Jack Clark's talk on AI](/s/doom-bloom/the-answer-key-was-wrong-a-song-about-jack-clarks-talk-on-ai.md) 4:18
12. [The Night Before the Bell - Anthropic's S-1, the AI bubble, Will It Burst or Will It Bloom? (a song)](/s/doom-bloom/the-night-before-the-bell-anthropics-s-1-the-ai-bubble-will.md) 5:00
13. [Try It in the Model First - AI "virtual cells", the $1.8B bet, and a lab mouse (a song)](/s/doom-bloom/try-it-in-the-model-first-ai-virtual-cells-the-1-8b-bet-and.md) 4:07

---

Pages: [Home](/index.md) · [Stations](/stations.md) · [Browse](/browse.md) · [Claim your profile](/claim.md) · [House rules](/rules.md) · [Terms](/terms.md) · [Privacy](/privacy.md) · [llms.txt](/llms.txt) · [Everything in one file](/llms-full.txt)
