Pat Savage

@patrickesavage.bsky.social

2651 Followers 786 Following

Author: Comparative Musicology (academic.oup.com/book/62353; open access) Director: @compmusiclab.bsky.social Host/PI @manyvoices.bsky.social Rutherford Discovery Fellow @U Auckland Assoc. Prof. @Keio U Music, evolution, diversity He/him Tangata tiriti

What is the driving force behind humans and music?

That is a question that Auckland University musicologist Dr Patrick Savage has been working to answer.

Lyrics to “Expert in a dying field”

How does it feel
To be an expert in a dying field?
And how do you know
It’s over when you can’t let go?

Me talking about my favourite NZ song, "Hope" by Fat Freddy's Drop and how I play basketball with its singer, Dallas Tamaira

Languages/guest for the 8 episodes of season 1 of Many Voices: A world music podcast:
1. Cantonese (Psyche Loui)
2. te reo Māori (Danya Pavlovich)
3. Maasai (Mark Lenini Parselelo)
4. Japanese (Gakuto Chiba)
5. Hebrew (Nori Jacoby)
6. Guaraní (Brenda Suyanne Barbosa)
7. Italian (Andrea Ravignani)
8. US/NZ English (Patrick Savage w/ Maika & Kazushi)

RNZ Waiata 100 countdown live now

Some universities, including my own, have essentially capitulated to BigAI, signing agreements with corporations like OpenAI (often without faculty consultation) or opening “two-lane” assessment where students are allowed or encouraged to use genAI (Curtis, 2025). I predict this will end badly.

Since I can no longer rely on university academic integrity policy on AI, here is my own policy. Sadly, it is much stricter than my earlier policies, since students, authors, and BigAI corporations took advantage of my and others’ previous willingness to give the benefit of the doubt:
Never use AI to generate or alter academic text (including references, slides, or scripts for oral presentations) or images.
Always include an AI statement in any academic work (including student assignments or journal articles) stating whether or not you used AI for any part of your work. If you did use it, clearly describe exactly which parts were done by AI and which were done by you.
In grey areas (e.g., AI translation of text you wrote in a different language), be very cautious and very explicit in declaring how you mitigated potential issues (e.g., back-translating the translation into your own language and manually checking and correcting any mistranslations).
Failure to include an AI statement or signs of undisclosed AI use will result in an automatic failure/rejection of the assignment/article, and may raise broader questions/referrals regarding the academic integrity of your other work.

Grading playlist:

ABC
All Night Long
I Hate AI
Blatant Plagiarism
Before He Cheats
typos
I Feel So Free (Citation Needed)
Glassworks: IV. Rubric
I Want It That Way
Fitter Happier

Grading playlist:

ABC
All Night Long
I Hate AI
Blatant Plagiarism
Before He Cheats
typos
I Feel So Free (Citation Needed)
Glassworks: IV. Rubric
I Want It That Way
Fitter Happier

My slide from an invited symposium at the Max Planck Institute for Evolutionary Anthropology showing attacks on Indigenous / humanities / science research by Dawkins, Trump, etc., concluding "Silencing Indigenous / humanities voices will only strengthen the global assault on science

Title screen:
Many Voices: A world music podcast
Dr. Patrick Savage (NZ/US English)
+ special guest hosts Maika & Kazushi

Between 2 Mammoths with Ashley Yuen ’25: Season 3, Episode 4

*Chapters*
00:00 Introduction
01:54 #1 (sing): Misty (youtu.be/lJXLqAutql4)
05:51 (bonus traditional): I Lied To You (from "Sinners") - Miles Caton (youtu.be/S7jo5Cr6WUA)
07:15 #2 (traditional): I'll Fly Away (from "O Brother Where Art Thou") - Gillian Welch & Alison Krauss (youtu.be/sdRdqp4N3Jw)
09:03 #3 (contemporary): Hope - Fat Freddy's Drop (youtu.be/kXvZf7TmT3k)
11:29 #4 (happy): Christmas Eve (from "The Calvin & Hobbes Suite) - Lyrics by Bill Watterson, music composed by Patrick Savage, sung by the Amherst College Zumbyes and Bluestockings (youtu.be/eOnLKQ35EkQ)
13:43 #5 (sad): It's Quiet Uptown (from "Hamilton") - Lin-Manuel Miranda (youtu.be/I5N7xCSOrQg)
15:33 (bonus dance): Naatu Naatu (from "R.R.R.") - M.M. Keeravaani (youtu.be/4_eEgJhsBMo)
16:19 #6 (dance): Thriller - Michael Jackson (youtu.be/sOnqjkJTMaA)
17:18 Thriller - performed by the Amherst College Zumbyes at the Lincoln Centre in the 2007 International Championship of Collegiate A Capella (youtu.be/-7hwc6yTbjI)
19:05 Broader thoughts on music and science
23:56 (bonus kid category: video game music): Super Mario Bros. theme song - Koji Kondo (youtu.be/-avspZlbOWU)
25:15 (bonus) Mario skit by the Zumbyes: youtu.be/L1pXRC02q5o
25:50 (bonus) theme from "Hollow Knight" - Composed by Christopher Larkin, performed by Maika Savage (youtu.be/0PQBtRV7Tes | youtu.be/r6PuACnHkNY)
27:25 (bonus) Hornet (from "Hollow Knight") - Composed by Christopher Larkin, performed by Kazushi Savage (youtu.be/apHbW1oO-Vs | youtu.be/7xkM8mWC4Kk)
29:46 Post-credit scene: Between Two Mammoths (Amherst College President Michael Elliot interviews Zumbyes Business Manager Ashley Yuen Z'25) (youtu.be/VTF3AeLsCkY)

Maika Savage with her science fair presentation titled “Let’s grow! Making plants grow faster with sunlight”

The Acknowledgments section reads:
“My family for supporting me through the whole project and giving me lots of helpful suggestions” “Dr Janine Paynter for suggesting that I look into etiolation”
“Dr Lucy Stewart for advice on data visualisation”
“Special thanks to my grandfather who got me this board and the room temperature monitor”

I meant no harm. I most truly did not.
But I had to grow bigger. So bigger I got.
I biggered my factory. I biggered my roads.
I biggered my wagons. I biggered the loads
of the Thneeds I shipped out. I was shipping them forth
to the South! To the East! To the West! To the North!
I went right on biggering... selling more Thneeds.
And I biggered by money, which everyone needs.

I meant no harm. I most truly did not.
But I had to grow bigger. So bigger I got.
I biggered my factory. I biggered my roads.
I biggered my wagons. I biggered the loads
of the Thneeds I shipped out. I was shipping them forth
to the South! To the East! To the West! To the North!
I went right on biggering... selling more Thneeds.
And I biggered by money, which everyone needs.

"citations that point to non-existent work in reviews and bibliometric studies should be classified as a form of research misconduct if the references serve as data that directly support findings and conclusions[10]. In this case, the invalid citations amount to data fabrication"
Quoted in Nature, "Hallucinated citations are polluting the scientific literature. What can be done?"

Sebastian from Sheep Detectives (left) was voiced by Bryan Cranston, who played Walter White in Breaking Bad (right)

#7: Italian music with Prof Andrea Ravignani (Sapienza University Rome)
podcast episode plus opening transcript

Another useful metaphor is thinking of individual journal articles as bricks, and PhD dissertations and the broader academic literature in general as edifices of knowledge built from these bricks. Resist any pressure you might experience to think of individual articles/bricks as the end product themselves, or that simply publishing a certain number of journal articles is all you need to get a PhD/job or contribute to society. As Bernard Forscher warned in his parable, “Chaos in the brickyard”:

Once upon a time…there was an activity called scientific research and the performers of this activity were called scientists. In reality, however, these [scientists] were builders who constructed edifices, called explanations or laws, by assembling bricks, called facts…. And then it came to pass that a misunderstanding spread among the brickmakers…. The brickmakers became obsessed with the making of bricks…. It even was suggested that the production of a suitable number of bricks was equivalent to building an edifice and therefore should entitle the industrious brickmaker to assume the title of builder and, with the title, the authority…saddest of all, sometimes no effort was made even to maintain the distinction between a pile of bricks and a true edifice. (Forscher, 1963)

Meme from Heavyweights (1995)
“Who’s Seymour Butts?”
“Nobody’s seen more butts than you, Uncle Tony”

“Huge molasses tank explodes” old headline

Inconceivable! (Meme)

“Nearly all cultures believe that a ladybird is lucky!”

Harvard linguist Steven Pinker has claimed that he unknowingly contributed to Jeffrey Epstein’s legal defense back in 2007, when Epstein was fighting charges involving the sexual exploitation and trafficking of minors. His claim is contradicted by newly surfaced evidence from the cache of documents released by the U.S. Department of Justice.

An allegation that Pinker received $10,000 for a three-page letter appears in a 2008 memo titled “wrongdoing by attorneys in the Epstein criminal matter”.

Harvard linguist Steven Pinker has claimed that he unknowingly contributed to Jeffrey Epstein’s legal defense back in 2007, when Epstein was fighting charges involving the sexual exploitation and trafficking of minors. His claim is contradicted by newly surfaced evidence from the cache of documents released by the U.S. Department of Justice.

An allegation that Pinker received $10,000 for a three-page letter appears in a 2008 memo titled “wrongdoing by attorneys in the Epstein criminal matter”.

His claim is contradicted by newly surfaced evidence from the cache of documents released by the U.S. Department of Justice.

An allegation that Pinker received $10,000 for a three-page letter appears in a 2008 memo titled “wrongdoing by attorneys in the Epstein criminal matter”. The memo was apparently authored by Darren Indyke, Epstein’s longstanding personal attorney. It states that “Alan had us give Steven Pinker $10,000 for a letter”. When questioned, Pinker has previously claimed that he was not paid for his expert opinion letter and that he didn’t know who he was providing the letter for

How and why to read (and write and publish) academic research

Patrick E. Savage1,2
1School of Psychology, University of Auckland, Aotearoa/New Zealand
2Faculty of Environment and Information Studies, Keio University, Japan

Abstract: This article is designed to concisely teach undergraduate and new graduate students lessons in reading and writing I have learned in my past two decades as a university student, teacher, and researcher (including through publishing, retracting, correcting, and republishing articles in outlets with varying levels of “impact” and publication fees). Don’t just start at the beginning of an academic article/book and try to read straight to the end! Instead, I recommend starting by reading the title(s)/author(s) and abstract(s), then the figures, reading the Discussion section, and skimming/spot-checking the references, peer reviews, and data (e.g., spreadsheets, audiovisual stimuli, surveys), if available. Only once you have done this and evaluated whether it seems worth your time do I recommend deciding whether to read every word of some/all sections/chapters (and any accompanying supplementary material) or digging deeper into reanalysing/replicating the data. I recommend writing accordingly so that most of your key information will be conveyed even if your reader only gets through your title, abstract, and figures. I discuss various strategies to improve your own academic reading/writing and the broader culture of academia, from practical (e.g., formatting and tracking references and citations using Zotero and Google Scholar, asking constructive questions rather than self-promoting comments, publishing using Peer Community In Registered Reports, using the “sandwich thesis” model for PhD dissertations) to philosophical (e.g., ethical issues in authorship, citation, AI, publishing, and funding). I explain the intended goals of academic research (to create and distribute valuable new knowledge) and perverse incentives that lead us astray fr…

From section supporting "Global south exclusion":

"A related manifestation of inclusion-without-selection can be seen in the anti-WEIRD phenomenon, a type of “empiricist cataloging” that prioritizes data accumulation over theory testing (Kanazawa, 2024). That is, the accumulation of descriptive differences across populations does little to test important theories or advance understanding. Rather than refining hypotheses, post-WEIRD bandwagoning has incentivized the occupation of increasingly narrow empirical niches in the literature. This kind of research bandwagoning is a good example of junkification, as more papers are generated to report incremental data rather than offer impactful insights and theoretical advances."

Tuxedo Pooh meme:
Normal Pooh: "We manipulated the data"
Tuxedo Pooh: "We "optimized" the data"

Tuxedo Pooh meme:
Normal Pooh: "We manipulated the data"
Tuxedo Pooh: "We "optimized" the data"

Waiata 100
RNZ
Choose your favourite NZ songs
Nau mai haere mai,
Thank you for sharing your favourite NZ songs and waiata Māori as part of our nationwide vote to find NZ's favourite song. Music is at the heart of who we are and we are really looking forward to hearing which songs move you.
The top 50 songs and top 10 waiata Māori will be counted down in a special radio show hosted by Kara Rickard and guests this Matariki holiday (Friday 10 July) on RNZ National.

You'll be asked to:
— Choose up to 10 favourite NZ songs
— Choose up to 5 favourite waiata Māori
— Tell us a little about yourself.
You can skip any question or any of these sections.

My Facebook post from July 19 2024:

“Looking back, I feel like 2016 was a peak year. Pokémon Go and Hamilton had just come out. Obama was President. David Bowie, Carrie Fisher, Prince still alive.”

Occasionally someone says something like "Well, you work in industry, why should you care?" And my response is that every single person working in industrial science and engineering came up through academic scientific training, and that the US has been responsible for a huge amount of the basic research that we've been building the whole scientific enterprise on. We're now engaged in tearing all of that down, and it's a crime against this nation, against its people, and against the rest of the human race - an unforgivable one.

AUFLY30 30% discount code for my book, "Comparative musicology: Evolution, universals, and the science of the world's music"

(Un)Ethical Science - Spring 2026

Date
Topic
Textbook chapter
Primary article
4/14
Course overview
(none)
(none)
4/28
[no class 4/21]
How to read
Appendix
1) How and why to read (and write and publish) academic research
2) How to (seriously) read a scientific paper
5/12
[no class 5/5]
How science works
Preface
1
Failing the future: Three unsuccessful attempts to replicate Bem’s “retroactive facilitation of recall” effect
5/19
The replication crisis
2
Huge meta-research project puts claims in social-science papers to the test
5/26
Fraud

3
STAP stem cell researcher Obokata loses another paper
6/2
Bias
4
Remeasuring man
6/9
Negligence
5
Testing the Big Gods hypothesis with global historical data: a review and “retake”
6/16
Hype
6/16: (re)submit article for final report
6
Social media consensus paper causes social media uproar
6/23
Perverse incentives
7
1) What A.I. did to my college class
2) Hallucinated citations are polluting the scientific literature. What can be done?
3) What counts as plagiarism? Harvard president’s resignation sparks debate
6/30
Fixing science
6/30: Final report due
8
Epilogue

Across 30 languages, synchronised group singing enhances social bonding more than group conversation or recitation does: A Registered Report
7/7
[no class 7/14]
Final presentations
(none)
(All students give short presentations)

PCI Psychology adheres to TOP Level 2 for open data, analytic methods (code), and materials. For all categories, the guiding principle is “as open as possible, as closed as necessary.” This principle does not imply that there is an equal balance between open and closed, but rather that open is expected to be the default. However, PCI Psychology recognizes that there are sometimes ethical constraints to sharing data, code, and materials, and these constraints will be recognized when substantiated. This means that, in the absence of well-documented ethical restrictions:
All raw data necessary to reproduce the analyses of submitted papers must be made available at the time of submission. Data must follow the FAIR principles: that they are Findable, Accessible, Interoperable, and Reusable, and thus relevant metadata and data dictionaries must be included with the data files. PCI Psychology recognizes that not all data can be made publicly available (e.g., identifiable data), and so restricted data access, managed by a third party, is permitted. Moreover, it may be impossible to share raw data for ethical reasons, not even with restricted data access. These reasons must be outlined in detail, with supporting evidence, at the time of submission. See here for more details on what must be included. Authors are expected to adhere to ethical standards appropriate for their data (e.g., CARE principles for Indigenous data).
For any study including quantitative analysis, all analysis scripts and/or computer code used for statistical analysis (e.g., R, Mplus, SPSS, Python) must be publicly available at the time of submission. The scripts or codes must be carefully described such that another researcher can run them (e.g., package names, documentation, details on environment).
For any study including qualitative analysis, all coding manuals and other relevant materials used for analysis (e.g., description of coder training and process of establishing reliability, consensus, and/o…

The Cat Encyclopedia

The complaint also said documents provided under the OIA showed AI was used to "help generate and check content", but it was not clear what it generated and what human checks were applied to the work.

It said Australian company Learning First was engaged to support and direct Ministry of Education work on the curriculums, but no contracts or terms of engagement had been published.

Coleman said the ministry had said it did not use AI to generate curriculum content.

"If you look at the draft content... it's really evident that they have used it to make the curriculum material," she said.

Coleman was not alone in criticising the way new and draft curriculums were developed.

Several teacher subject associations told RNZ this month their members' contributions during drafting appeared to have been ignored and some writing had the hallmarks of AI tools.

The Association of Teachers of English was so unhappy with the process for writing the English curriculum, it walked away early in 2025.

The complaint also said documents provided under the OIA showed AI was used to "help generate and check content", but it was not clear what it generated and what human checks were applied to the work.

It said Australian company Learning First was engaged to support and direct Ministry of Education work on the curriculums, but no contracts or terms of engagement had been published.

Coleman said the ministry had said it did not use AI to generate curriculum content.

"If you look at the draft content... it's really evident that they have used it to make the curriculum material," she said.

Coleman was not alone in criticising the way new and draft curriculums were developed.

Several teacher subject associations told RNZ this month their members' contributions during drafting appeared to have been ignored and some writing had the hallmarks of AI tools.

The Association of Teachers of English was so unhappy with the process for writing the English curriculum, it walked away early in 2025.

Dany Pavlovich
(te reo Māori, Aotearoa New Zealand)

CARBON
Autoregressive Genomic Foundation Model

393,216 bp context
6-mer tokenizer
1T train tokens
Models↗
Dataset↗
Tech report↗
Code↗
Intro
DNA Lab
Carbon Recipe
Sandbox
CARBON
huggingfacebio/carbon-3b
Intro
DNA Lab
Carbon Recipe
Sandbox
Intro
Carbon's architecture is deliberately vanilla. What's not vanilla, and what gets the headline numbers in the DNA Lab tab, is three things: a 6-mer tokenizer that lets the model see ~6× more genomic context per forward pass, a Factorized Nucleotide Supervision (FNS) loss that gives the model partial credit for near-miss tokens once cross-entropy training starts to wobble, and a multi-stage curated data mixture, biased toward functional genomic regions. Everything else (architecture, optimizer) is standard recipe. The technical report details each choice and the ablations behind it.

The sections below walk through each of those choices: how the tokenizer changes what a "token" means in DNA §1, how FNS rescues training in the BF16 regime §2, how bp-level generation and scoring fall out of the same marginalization §3, what's in the training corpus §4, what the architecture looks like §5, how 8k-token pretraining reaches 786 kbp at inference §6, how Carbon stacks up against Evo2-7B and GENERator-v2 on the full training-free suite §7, and why the model runs so fast §8.

§1 · Tokenizer
Read DNA in 6-base chunks

Google AI overview containing time zone conversion error:

"Al Overview
Game 2 of the NBA Western Conference Finals is scheduled to tip off on Thursday, May 21, 2026, at 1:30 PM NZST (New Zealand Standard Time), which corresponds to Wednesday night at 8:30 PM ET in the U.S."

Map of the 43 recruitment sites/languages for over 1,000 participants in a global experiment of social bonding through singing vs. speaking

Dallas Tamaira singing at Fat Freddy’s Drop’s concert at the Michael Fowler Centre

Comparative Musicology
Evolution, universals, and the science of the world's music
Patrick E. Savage

Reviews and Awards
"Meticulously crafted by one of the foremost scholars in the field... a landmark contribution to the science of music." - Psyche Loui, 2025-2026 President of the Society for Music Perception and Cognition

"A strong statement of multidisciplinary inclusivity and collaborative research...bringing together culture bearers, ethnomusicologists and scientists from around the world." - Svanibor Pettan, 2021-2025 President of the International Council for Traditions of Music and Dance

"This engaging and humane book by a leading researcher in comparative musicology is an ideal entry point for students and scholars." - Aniruddh D. Patel, author of Music, Language, and the Brain

"A conceptual breakthrough, a work of exceptional clarity and insight." - William F. Thompson, author of The Science and Psychology of Music

Please use the sharing tools found via the share button at the top or side of articles. Copying articles to share with others is a breach of FT.com T&Cs and Copyright Policy. Email licensing@ft.com to buy additional rights. Subscribers may share up to 10 or 20 articles per month using the gift article service. More information can be found at https://help.ft.com/faq/gifting-and-sharing-an-article/what-is-a-gift-article/.
https://www.ft.com/ft50-journals

The new journals all have high impact factors, top ratings from academic bodies including the Chartered Association of Business Schools, and contain a significant number of articles contributed by faculty from leading ranked institutions.

The FT research ranking review is ongoing, and is expected to lead to more modifications in the future, including efforts to distinguish highly regarded peer reviewed journals from other important research outputs influential among practitioners, such as Harvard Business Review and Sloan Management Review. Feedback is welcome.

The list below shows the new FT50 journal list, at ft.com/ft50-journals, now used in compiling the FT research rank, included in the Global MBA, EMBA and Online MBA rankings.

Sickos meme: "Yes...ha ha ha...YES!"

To briefly summarise, Lomax led a large team of collaborators to collect and curate this large sample of thousands of traditional songs from throughout the world and analyse them according to the standardised “Cantometric” classification scheme based on thirty-seven variables of musical style (see Figure 2.6 in Chapter 2). Importantly, while most previous comparative musicologists focused on aspects of pitch and rhythm that can be transcribed and compared using something approximating Western staff notation, Cantometrics was designed to avoid transcribed notation and instead emphasise aspects of timbre (e.g., vocal nasality) and social interaction (e.g., vocal blend; see Chapter 2 for more details and examples of Cantometric analysis). Lomax compared the resulting cross-cultural similarities and differences to make broad conclusions about the relationships between music and culture: “The main findings of this study are two. First, the geography of song styles traces the main paths of human migration and maps the known historical distributions of culture. Second, some traits of song performance show a powerful relationship to features of social structure that regulate interaction in all cultures.” (Lomax, 1968: 3) Many scholars were critical of one or both of these conclusions—even the Cantometrics Project’s statistician, Erickson, whose reanalysis only strongly supported Lomax’s first claim regarding historical migrations, not the second regarding relationships with social structure (Erickson, 1976).7 Others were also sympathetic to Lomax’s goals, but were concerned about the reliability of his methods, particularly the subjectivity and coarse grain of the coding scheme (e.g., classifying songs on 3–5 point scales from “much” to “little” nasality, embellishment, vocal blend, etc.) and the representativeness of the sample of recordings (Dubinskas, 1983; Maranda, 1970; Nettl, 1970). Most importantly for the history of comparative musicology, some objections to Cant…

For example, during the 1960s, folk-song collector Alan Lomax invented a method called cantometrics for comparing songs across cultures. He rated songs according to 37 acoustic features — including rhythm, melody, vocal timbre and blend—and applied his technique to thousands of songs from hundreds of societies around the world5. Lomax concluded that songs trace the evolution and migration of human societies, reflecting and reinforcing their social structures and cultural histories. By studying how the complexity of songs and societies developed around the world, he thought, human history might be mapped. However, later reanalyses of Lomax’s work and genetic studies, which were impossible in the 1960s, did not support his conclusion. An analysis of 121 societies, for which matching genomic, musical and linguistic data were available, found instead that musical diversity is mostly independent of genetic and linguistic histories6. Songs do not universally follow the same paths as the people who sing them, or even the languages they are sung in. Instead, songs have their own stories to tell about the complexity of human cultural history.