andrea e. martin

🤞💜

🤯🤔🫡

🤍

My photo shows a mosaic fragment from the floor of a Roman villa in Spain, dated AD 100-200s. It depicts a cartoon-like, stylised octopus using red, yellow, white, and black limestone tesserae, against a white background. The front-facing octopus has a large red and white head/body outlined in black. Its yellow circular eyes are outlined in black with a central black dot. Below the head/body are eight writhing tentacles

Charming little octopus from a Roman villa at Villaquejida, León, Spain. Limestone, 2nd-3rd century AD.

Museo Arqueológico Nacional, Madrid 📷 me

#Archaeology

🩶

hahahaha and wow

@andreaeyleen.bsky.social wrote a poem about LLMs 🫠

Get to know New fellow @andreaeyleen.bsky.social from the MPI for Psycholinguistics. In a recent paper (Weissbart & Martin, 2024, Nature Communications), she shows that the brain doesn’t just rely on word statistics or grammatical rules, but it uses both. Very excited to have her in our program! 🧠🧠🧠

6. On Narcissus: A reminder that reflections, however alluring, are not the things reflected. Readers who mistake Microsoft Word for written cognition, or cameras for eyes, are invited to repeat the myth until they grasp its point. 21/n

5. On shrimp fried rice: The rhetorical question, “Are you telling me a shrimp fried this rice?” is here repurposed to illustrate the absurdity of mistaking statistical artifacts for cognitive theories. That the metaphor now circulates online as “brain worms” is both apt and lamentable. 20/n

4. On predictive coding: A doctrine promising to explain everything by minimizing free energy, but explaining nothing of how, precisely, this explains anything about something in particular. The poet therefore dismisses it with impatience. 19/n

3. On toes: The poet does not exaggerate in her appeal to Big Toes (Big Bird’s, Lovelace’s, or Leibniz’s mother’s). Each toe is, in fact, both embodied and chronological, two properties seldom attributed to LLMs. 18/n

2. On the weathervane: One might object that weathervanes are not usually invoked in theories of mind. But consider their symbolism, their function, their embodiment, their relationships to hype and zeitgeist, their versatility as weapons. Theorists could do worse (and often do). 17/n

Notes:
1. On the phone: The reader may find it strange that a pocket telephone with blinking icons should outshine a trillion-parameter network. Yet the former has transistors, circuits, batteries, and an instruction manual—each more mechanistic than any blackbox model. 16/n

So give me anything—phone, shrimp, or toe—ne’er predictive coding tho’-

For minds, and meaning, be more than what machines may show. 15/n

With Thanks to Lord Byron, Emily Dickinson, & @olivia.science

Let others worship code in endless streams,
Confusing clever tuning with human dreams;

But I, uncharmed, behold! O, autoregressive cistern,
minds are much more than what machines may return. 14/n

Like Narcissus, we gaze and fall instead,
Mistaking shallow waves for what is said.
Reflections glitter, patterns rearranged,
But minds are mortal, embodied, and constrained. 13/n

For minds are not the sum of text we see,
Nor mirrors cast by vast circuitry… circularity?
None ever claimed that Microsoft Word was thought outright,
Nor that cameras gave us vision’s inward sight. 12/n

A shrimp may fry this rice, forsooth, as well,
As LLMs our inner lives compel.
So tell me, friends, is linalg our fate?
Or symbols buried deep in online spate? 11/n

Are we to think, O scholars, with a straightened face,
That a corporate codebase explains the human race?
That internet-trained products, bought and sold,
Unlock the painted hands in caverns old? 10/n

And spare me too the free-energy creed,
Where no one knows how minimized need
Explains what mind and brain must heed.
Predictive coding, endlessly proclaimed,
Is less a theory than a description, trivial, renamed. 9/n

So give me toes—or any limb, in fact—
But spare me “minds” from matrices abstract. 8/n

What say you: Big Bird’s claw (or toe? the truth is vexed),
Which predates ARPANET, if I’m not perplexed.

Better yet Lovelace’s toe predates the gradient’s fall;
Leibniz’s mother’s toe the linear algebra we invoke in all. 7/n

To avoid more explanatory woe, one must
choose a theorist older, tho’
than the Net
That cast our table’s looks, up and fret

If we doth need the internet
To feed our model mind’s requests

A model mind
Let us refine 6/n

Yes, Big Toe speaks with greater right—
as it trod the earth before bytes took flight.
(though, behold, i’m not THAT old)
—Than a devilish table of data, tokenized
Filled with every unfathomable correlation, monetized 5/n

Yet smartphones don’t predate the Net,
Where large language models find their debt.
But lo—my big toe, more ancient yet:
A theorist older than the INTERNET. 4/n

A weathervane turning with wind in the sky,
Tracks more of the mind’s environs
Than approximate-retrieval look-up tables, “brain-aligned.”

A single sharp theory, though modest in line,
Explains more of thought than datacenters nine. 3/n

Give me, I beg, some model yet unknown,
For LLMs deserve not thought nor throne.
Some claim they mimic reason, mind, and art,
But ape the form while missing every part.
A smartphone, its circuits and apps all aligned,
Would serve far better as a model mind. 2/n

Ode to the original language model, or:
Give me literally Anything* instead of Large Language Models (LLMs)
*(no predictive coding either!)

By Lady Byronadrea LLMartin 1/n

A black and white photo of Cecilia Payne-Gaposchkin. She is sitting at her desk and looking up at the camera, which is in front of her and to her left. Payne-Gaposchkin is wearing a baggy, ribbed sweater and has a wristwatch on her right arm. A pair of glasses rests on the desk in front of her, next to open books and papers which are just visible at the bottom of the photogaph. Filing cabinets, drawers, and another desk are visible in the background. She has short, dark hair which is no longer than chin-length, pulled out of her face behind her left ear.

“There is no joy more intense than that of coming upon a fact that cannot be understood in terms of currently accepted ideas."

Astronomer Cecilia Payne-Gaposchkin, who decoded spectral lines to deduce the elemental composition of stars, was born #OTD in 1900. 🧪 🔭 ⚛️ 👩‍🔬

Image: Harvard Observatory

Linguistics.