Somewhere in Columbia University’s rare book library, a clay tablet has been sitting largely misunderstood for nearly a century. It is small enough to hold in one hand. Its edges are chipped, one corner missing entirely. It was made in Babylon around 1800 BCE — roughly 3,800 years ago. And according to a 2017 paper published in Historia Mathematica, it contains a trigonometric system that is, in at least one specific way, more mathematically accurate than the one we use today.

Its name is Plimpton 322. And it is only one of approximately 500,000 cuneiform tablets still waiting to be read.

Writing in Wedges: What Cuneiform Actually Is

Before we get to the mathematics, it is worth understanding why these tablets took so long to decode. Cuneiform — from the Latin cuneus, meaning wedge — is not a language. It is a writing system. Over 1,000 distinct characters, each pressed into soft clay with a sharpened reed, each changing appearance across centuries, across cities, and across individual scribes. The same symbol in Nippur looks different from the one written in Babylon five hundred years later.

Today, fewer people can read cuneiform than can fly a commercial aircraft. A writing system spoken by millions for thousands of years, readable now by a few hundred specialists worldwide.

In March 2025, a team from Cornell University announced an AI system — ProtoSnap — capable of reading them all. It uses a diffusion model (the same architecture behind modern AI image generation) to overlay character prototypes onto damaged clay, aligning pixel-by-pixel, then performing optical character recognition on the result. Tested on rare, damaged, previously unidentifiable characters, it outperformed every prior method. The goal stated publicly: increase accessible ancient knowledge by a factor of ten.

There are 500,000 tablets. The machine is running. (See the Spacialize video that prompted this article.)

Plimpton 322: The Trigonometry That Shouldn’t Exist

The tablet was acquired by New York publisher George Arthur Plimpton in the 1920s and donated to Columbia upon his death. For decades, researchers knew it contained Pythagorean triples — sets of whole numbers satisfying a² + b² = c². Interesting, but not earthshaking.

Then in 2017, Dr. Daniel Mansfield and Professor Norman Wildberger of the University of New South Wales ran the full analysis. What they found changed the framing entirely.

Plimpton 322 is not simply a list of Pythagorean triples. It is a systematic trigonometric table — 15 rows covering a range of angles in roughly 1-degree increments, each row describing the shape of a right-angle triangle using exact ratios of its sides. It predates Hipparchus, long credited as the father of trigonometry, by over a millennium. And it predates Pythagoras — whose theorem it implies — by 1,200 years.

Mansfield’s conclusion, stated without hedging: Plimpton 322 is “the only completely accurate trigonometric table in existence.”

Not accurate for its time. Completely accurate.

It is worth noting that Mansfield’s interpretation is not universally accepted — a sceptical analysis in Scientific American argues the claim is overstated. But even critics acknowledge the tablet contains real and sophisticated mathematics. The argument is about degree, not kind.

Why Base-60 Beats Base-10: The Arithmetic Behind the Claim

To understand why, you need to understand what the Babylonians were doing differently at the number system level.

We use base-10 (decimal): digits 0–9, each column worth ten times the one to its right. The Babylonians used base-60 (sexagesimal): each positional column worth sixty times the one to its right. Same positional principle — but with a crucial consequence.

A number system can only represent fractions exactly when the denominator’s prime factors are all present in the base.

• Base-10’s prime factors: 2 and 5. So 1/2 = 0.5 ✓, 1/4 = 0.25 ✓ — but 1/3 = 0.3333… (infinite), 1/6 = 0.1666… (infinite), 1/7 = 0.142857… (infinite).
• Base-60’s prime factors: 2, 3, and 5. Its divisors include 1, 2, 3, 4, 5, 6, 10, 12, 15, 20, 30, and 60.

In sexagesimal notation (using semicolons to separate the integer from fractional parts, commas between fractional digits):

• 1/3 = 0;20   (20/60 = exactly 1/3) ✓
• 1/4 = 0;15 ✓
• 1/6 = 0;10 ✓
• 1/9 = 0;6,40 ✓
• 1/12 = 0;5 ✓

Every calculation our modern trigonometry makes in base-10 carries a small inherited rounding error. Ratios that should be clean fractions become infinite decimal expansions, which computers truncate at some precision boundary. The Babylonian system avoided this entire class of error — not by being more sophisticated, but by choosing a base with more divisors.

We preserved their system without realising it. Every time you divide an hour into 60 minutes and 3,600 seconds — every time you measure an angle in degrees, arcminutes, and arcseconds — you are using sexagesimal arithmetic. The Babylonians are still in your GPS.

Implementing Sexagesimal: Exact Arithmetic in Practice

The Babylonian approach was also conceptually different from ours. Rather than working with angles and circular functions (sine, cosine, tangent), they worked directly with ratios of triangle side lengths, expressed as exact sexagesimal fractions. Ratio-based trigonometry: no π, no infinite series, no irrational approximations needed.

The key insight is elegant: when a right triangle has integer side lengths (a Pythagorean triple), all its trigonometric ratios are rational numbers. Rational numbers can always be expressed exactly — and base-60, with its divisor-rich structure, handles the most common ones with no fractional remainder at all.

Here is a minimal Python implementation that reproduces the Babylonian logic using exact rational arithmetic:

from fractions import Fraction

def to_sexagesimal(f, places=4):
    """Convert a Fraction to sexagesimal notation list."""
    result = []
    integer_part = int(f)
    result.append(integer_part)
    remainder = f - integer_part
    for _ in range(places):
        remainder *= 60
        digit = int(remainder)
        result.append(digit)
        remainder -= digit
        if remainder == 0:
            break
    return result

def babylonian_trig(a, b, c):
    """
    Compute exact trig ratios for a right triangle with sides a, b, c.
    c is the hypotenuse. Returns exact Fractions — no rounding, ever.
    """
    a, b, c = Fraction(a), Fraction(b), Fraction(c)
    return {
        'sin': a / c,
        'cos': b / c,
        'tan': a / b,
        'sin_sex': to_sexagesimal(a / c),
        'cos_sex': to_sexagesimal(b / c),
    }

# The classic 3-4-5 triple
print(babylonian_trig(3, 4, 5))
# sin = 3/5 exactly. cos = 4/5 exactly.
# In sexagesimal: sin = [0, 36] — i.e. 36/60. Terminates perfectly.

# A Plimpton 322 entry (first row, scaled)
print(babylonian_trig(120, 119, 169))
# sin = 120/169 — exact, with no floating-point error whatsoever.

Python’s Fraction class does exactly what base-60 did in clay: it maintains exact rational arithmetic throughout. The modern float expression 0.1 + 0.2 famously returns 0.30000000000000004. A Fraction-based equivalent returns exactly 3/10. For trigonometric ratios derived from integer triples — the Plimpton 322 approach — results are always exact.

Mansfield explicitly noted this has direct relevance for computer graphics, engineering, and surveying — any domain where rounding errors compound across thousands of sequential calculations. For certain geometric problem classes, the Babylonian approach is not a historical curiosity. It is simply the right tool.

The Astronomer With a Reed and Wet Clay

The mathematics is striking, but perhaps the most viscerally impressive demonstration of Babylonian precision is not numerical. It is observational.

British Museum artifact K8538 — the Planosphere — records a Sumerian astronomer describing an object approaching Earth before dawn. He notes its angle against the background stars. The observation is dated to June 29, 3123 BCE. Bristol University astrophysicists fed those angular measurements into modern computer simulation. The trajectory matched a confirmed geological impact event in the Austrian Alps — at a precision of less than one degree of error.

The Very Large Telescope in Chile achieves comparable angular precision using adaptive optics, laser guide stars, and real-time atmospheric correction. This Sumerian astronomer had a reed and wet clay. The Bristol team, in peer-reviewed astrophysics, concluded that the observation represents a level of precision their models of ancient technological capability cannot account for.

The Archive Nobody Is Talking About

Five hundred thousand tablets. One AI system that can now read them all. From the institutions sitting on these collections — Yale’s Babylonian collection, the Oriental Institute in Chicago, the Istanbul Archaeological Museums, the British Museum — no coordinated statement, no public timeline.

When the James Webb telescope captures a new image, there is a coordinated press conference within hours. When AI cracks a protein structure, the global scientific community responds within weeks. The silence around cuneiform is of a different quality.

What is already established, from tablets decoded long before any AI was involved, is remarkable enough. A trigonometric system 1,200 years older than Pythagoras. An asteroid observation precise to under one degree, made with the naked eye. A mythological language that encoded meaning structurally into its alphabet — the cuneiform sign for fox is identical to the words for lie, treacherous, and falsehood. You cannot write the animal without simultaneously writing the concept.

The oldest known trickster character in human literature — 4,400 years old, predating Loki, Coyote, and Hermes — decoded in 2025 from a tablet that had sat unread in Istanbul since the 19th century.

There are 499,999 tablets remaining.

Sources & Further Reading

• Mansfield & Wildberger (2017). Plimpton 322 is Babylonian exact sexagesimal trigonometry. Historia Mathematica. doi:10.1016/j.hm.2017.08.001
• Smithsonian Magazine — Ancient Babylonian Tablet May Hold Earliest Examples of Trigonometry
• Scientific American — Don’t Fall for Babylonian Trigonometry Hype (sceptical counterpoint)
• Cornell University — AI models make precise copies of cuneiform characters (March 2025)
• Wikipedia — Sexagesimal
• Wikipedia — Plimpton 322
• YouTube — Quantum AI Cracked a 4,000-Year-Old Sumerian Tablet (Spacialize, Feb 2026)
• Python fractions module documentation