[Watching Martial Arts with AI #01: Pre-Fight Edition] When I predicted UFC 331 with ChatGPT, Claude, and Gemini, the main and co-main events became '3 AIs vs. Me'

If I have AI predict martial arts matches, how much can it actually read?

It's not just about the winner.

“What kind of flow will the fight have?”

“Will it be a KO, a submission, or a decision?”

“In which round will the momentum shift?”

I want to have them predict that far and then check the answers against the actual matches.

Driven by that curiosity, I am starting a new series.

The title is “Watching Martial Arts with AI.”

ChatGPT, Claude, and Gemini.

I had these three AIs predict the five main card matches of UFC 331 and added my own predictions to the mix.

Looking at it based on the winners:

・There were 2 matches where all 4 parties chose the same fighter

・There was 1 match where the predictions were split 2-2

・And for the 2 matches of the co-main and main events, all 3 AIs chose the same fighter, and only I chose the opposite

It turned out to be quite an interesting result.

Will the AI's comprehensive evaluation win?

Or will human intuition and the idea that “anything can happen in MMA” prevail?

First, I will reveal all the pre-fight predictions.

Regarding the recording rules for the first installment

I will be honest about this from the start.

The predictions from the three AIs—ChatGPT, Claude, and Gemini—were created before the start of the UFC 331 main card at 10:00 AM Japan Standard Time on September 20th and saved to Google Drive.

On the other hand, my own predictions for the 5 matches were entered into this chat around 10:47 AM on the same day, after I decided to expand this project to include “3 AIs + Me.”

Therefore, for the first installment, the “pre-fight verification of the 3 AIs” will be the focus of the official record, and my predictions will be treated as a reference.

From the second installment onwards, all four parties—the 3 AIs plus me—will have their predictions saved with timestamps before the start of the event.

It is not a perfect system from the start.

That is precisely why I am publishing it, including these gaps, while making improvements.

MMA Prediction Engine v0

Starting with this, I will not end the 3 AI predictions as a one-off, but will record them as the “MMA Prediction Engine.”

However, the “Prediction Engine” here is not a proprietary machine learning model at this point.

It is a system where I have ChatGPT, Claude, and Gemini analyze MMA, save their predictions, and compare them with the actual results to verify their accuracy and areas of expertise over the long term.

The v0 of the first installment is still an experimental version.

The input conditions and reference data for the 3 AIs are not perfectly unified.

I will improve the accuracy of this from the second installment onwards by fixing the common prompts, common data, and execution times.

Note that in this article, I will refer to ChatGPT by the nickname “Boku-chan.”

The first-person “I” refers to me, the author.

Match 1 | Alonzo Menifield vs. Ivo Baraniuk [Winner unanimous among all 4]

ChatGPT (Boku-chan)

Baraniuk / 1R KO/TKO

Applies pressure from the start, scores a knockdown near the cage, and follows up with ground and pound.

Claude

Baraniuk / 1-2R TKO

While being wary of Menifield's power, he pushes through with his undefeated momentum and early pressure.

Gemini

Baraniuk / 2R TKO

Intercepts Menifield's forward movement with long-range strikes and counters, gradually taking control.

Author (Me)

Baraniuk / 1R KO

Standing punches.

It is a very close prediction, with all 4 parties agreeing not only on the winner but also on the direction of “Baraniuk by striking.”

Are they all looking at the same thing?

Or do they all have the same blind spot?

In this project, I won't assume that “unanimous agreement equals correct,” and I will look into that after the match.

Match 2 | Gable Steveson vs. Sean Sharaf [Winner unanimous among all 4]

Steveson is a Tokyo Olympic wrestling gold medalist.

His wrestling-based fighting style in the UFC is also a major point of interest.

ChatGPT (Boku-chan)

Steveson / 1R TKO

Takes the top position from a takedown and follows up with ground and pound.

Claude

Steveson / 1R TKO

Takes him down early with Olympic-level wrestling and overwhelms him on the ground.

Gemini

Steveson / 1R TKO, or Decision

Gains the advantage with grappling and speed, focusing on ground control.

Author (Me)

Steveson / 1R TKO

Takedown followed by ground and pound.

I, ChatGPT, and Claude are in complete agreement on 'Steveson / 1R TKO'.

Only Gemini left the possibility of a decision.

More than the outcome, I think the highlight is 'how quickly that wrestling will take shape in MMA'.

3rd Match | Patricio Pitbull vs. Choi Doo-ho [2 vs. 2]

This is where we split right down the middle for the first time.

ChatGPT (Boku-chan)

Choi Doo-ho / Decision

Struggles with Pitbull's grappling and counters, but wins by a narrow margin in volume.

Claude

Choi Doo-ho / 3R TKO

Endures the early stages and overcomes with pressure and striking in the latter half.

Gemini

Pitbull / Decision 3-0

Mixes in calf kicks and grappling to disrupt Choi's tempo, demonstrating his skill as a veteran fighter.

Author (Me)

Pitbull / Decision

ChatGPT + Claude = Choi.

Gemini + Me = Pitbull.

It's a clean 2 vs. 2.

I think Choi has the momentum.

Even so, I predicted that within three rounds, Pitbull's experience, ability to manage distance, and capacity to disrupt his opponent's rhythm would take effect.

Personally, this is the card I'm most looking forward to checking the results for.

4th Match | Arman Tsarukyan vs. Mauricio Ruffy [3 AIs vs. Me]

According to the official UFC announcement, this match is a 5-round co-main event.

ChatGPT (Boku-chan)

Tsarukyan / 4R TKO

Wears him down with wrestling, followed by a takedown and ground and pound in the latter half.

Claude

Tsarukyan / 4R TKO, or Decision

Wary of the early striking, he drains Ruffy's stamina with takedowns and pressure.

Gemini

Tsarukyan / 3R SUB, or decision

Maintains top position from a double-leg takedown and dominates with ground fighting.

Author (Me)

Luffy / 2R TKO

A barrage of punches following standing kicks.

All 3 AIs chose Tsarukyan.

Only I chose Luffy.

Considering the level of wrestling and overall ability, I can certainly understand the AIs' predictions.

But I saw the possibility that Luffy's striking could break that 'correct game plan.'

Landing kicks, then following up with a barrage of punches.

My prediction is a 2R TKO.

Is it the AIs' focus on reproducibility?

Or the human's intuition for a comeback?

From the very first installment, we have a card that fits this project perfectly.

5th Match | Joshua Van vs. Alexandre Pantoja [3 AIs vs. Me]

The main event is a flyweight title match.

Champion Joshua Van takes on former champion Alexandre Pantoja.

Their previous encounter was at UFC 323.

Pantoja injured his elbow 26 seconds into the first round, and Van won the title.

So this rematch is also a chance to re-examine how they match up over 5 rounds.

ChatGPT (Boku-chan)

Van / Decision

Withstands Pantoja's grappling in the early stages and surpasses him in movement and boxing volume from the 3rd round onwards.

Claude

Van / Decision

Van, having survived the early stages, gains the advantage in the later rounds through his ability to make adjustments during the fight and his striking.

Gemini

Van / 4R TKO

Weathers Pantoja's grappling and corners him in the latter half with stamina and high-volume striking.

Author (Me)

Pantoja / Decision

All 3 AIs chose Van.

Only I chose Pantoja.

The AIs highly valued Van's youth, movement, and late-round striking.

Conversely, I placed more weight on Pantoja's experience and his ability to build a fight over 5 rounds.

Even if it doesn't lead to a finish, Pantoja will take the decision by accumulating grappling, positioning, and small points.

That is my prediction.

Summary of the 4 predictions

Winner agreed upon by all 4

・Match 1: Baraniewski

・Match 2: Steveson

2 vs 2

・Match 3: ChatGPT/Claude = Che / Gemini/Author = Pitbull

3 AIs vs. Author

・Match 4: 3 AIs = Tsarukyan / Author = Luffy

・Match 5: 3 AIs = Van / Author = Pantoja

Judging by the answers this time, the 3 AIs tended to base their predictions on overall ability, reproducibility, and stylistic compatibility.

On the other hand, I am looking a bit more closely at experience and the uncertainty unique to MMA where a game plan can be destroyed by a single blow.

However, this is only the first installment, an impression based on just 5 matches.

That is exactly why I will continue.

Rules for the first installment's review

In the first installment, we did not fix the formal scoring rules before the match.

Therefore, I will not decide the points conveniently after the fact.

This time, 'whether the winner was guessed correctly' will be the official record.

The method of victory, round, and match development will be reviewed as reference evaluations.

From the second installment onwards, the following will be fixed and saved before the event:

・Winner

・Method of victory

・Round

・Predicted development

・Confidence level

・Scoring rules

・AI and model used

・Common prompt

・Reference data

・Prediction execution time

If we accumulate this for every event, it won't just be about 'right or wrong', but rather:

Which AI is strong at predicting winners?

Who is good at predicting KOs?

Who can pick up upsets?

Are humans really more intuitive than AI?

We should be able to see things like that.

Reviewing the results in the 'Match Report' after the fight

After the match, I will review all 5 matches as a 'UFC 331 Match Report with AI'.

It's not just about the win or loss.

Did the predicted takedowns really work?

Was the striking distance read correctly?

In which round did the momentum shift?

Even if the winner was wrong, were there any predictions where the development was spot on?

Conversely, were there any predictions where only the winner was guessed correctly by chance?

I will look into that as well.

Using UFC 331 as the first set of data, I plan to continue this, focusing on UFC and RIZIN.

ChatGPT.

Claude.

Gemini.

And me.

Who can read MMA the best in the long term?

Will I be the one to beat the AI?

Or will the human side be getting beaten up after a few events?

Is your prediction closer to the AI's?

Or is it closer to mine?

See you in the review edition.

Now, let the viewing begin.

※This series is a project to enjoy analyzing and verifying martial arts match developments together with AI. Predictions do not guarantee results. When dealing with odds, etc., it is not intended to encourage betting or voting.

※Tournament cards, champions/challengers, number of rounds, Steveson's Olympic achievements, and the previous match between Van and Pantoja have been confirmed with official UFC information.