Test 5 - coding challenge to produce the longest word chains

This page has 0 threads | Add post

Blogs

Many people around the world enjoy and benefit from our blogs. If you're one of them - and can afford to - please make a small donation to help keep this and future blogs, exercises, skills assessment tests, tips, tutorials and videos freely available to everyone!

Some other pages relevant to these blogs include:

Classes in Artificial Intelligence

You can also book hourly online consultancy for your time zone with one of our 7 expert trainers!

Comparing how well the main AI tools deal with 5 difficult challenges Part six of a seven-part series of blogs
In what is becoming almost an annual exercise we revisit the main AI tools to see which (if any) has the edge in accomplishing 5 challenging tasks, each designed to test a different aspect of the use of AI. Which AI tools we'll test, and the 5 challenges each will face Test 1- analysing emails by importance Test 2 - creating a travel bingo card Test 3 - creativity Test 4 - summarising and presenting information Test 5 - coding challenge to produce the longest word chains (this blog) The final conclusion - not all AI tools are the same

Comparing how well the main AI tools deal with 5 difficult challenges

Part six of a seven-part series of blogs

In what is becoming almost an annual exercise we revisit the main AI tools to see which (if any) has the edge in accomplishing 5 challenging tasks, each designed to test a different aspect of the use of AI.

Which AI tools we'll test, and the 5 challenges each will face
Test 1- analysing emails by importance
Test 2 - creating a travel bingo card
Test 3 - creativity
Test 4 - summarising and presenting information
Test 5 - coding challenge to produce the longest word chains (this blog)
The final conclusion - not all AI tools are the same

Posted by Andy Brown on 11 May 2026

In this blog

ChatGPT
Claude
Copilot
Gemini
Conclusions and winner/loser

This challenge asked our AI tools to produce Python and Excel VBA code to find the longest possible word chains in English (where each word is obtained by shuffling the letters of the previous one and adding one more letter). The prompt asked the AI tool to keep its code simple, so that a casual user could understand it.

One noteable feature of this test: ChatGPT, Copilot and Gemini took about 2 minutes to come up with their answers, while Claude took nearly 10! I have included the VBA and Python programs generated by the various AI tools for download here.

ChatGPT

Judging by its results, I really like ChatGPT's program. It came up with this 12-letter chain, using everyday English words:

AN ? RAN ? RANG ? RANGE ? GARDEN ? DANGERS ? GRANDEES ? ESTRANGED ? DENIGRATES ? NEARSIGHTED ? STRAIGHTENED

it even told me that the only word people might not know was GRANDEES. The longest raw chain it found was this one, ending in a 15-letter word:

UT / TUT / TUTS / STRUT / STRUTS / STROUTS / TOURISTS / TOURISTAS / OUTSTRAINS / CRUSTATIONS / RUSTICATIONS / RESUSCITATION / CAUTERISATIONS / ANFRACTUOSITIES

However, ChatGPT admis that while this is valid "I would not call it a good average educated English person chain".

The ChatGPT VBA and Python code generated obeys my instructions: lots of comments and no compllicated constructs (see the hint at the top of the page for how to download all of the code created as part of this test).

Claude

Claude wins this competition for two different categories. It gave the best code, with superb documentation - and it took far, far longer than the other AI tools (approaching 10 minutes, as opposed to about 2 minutes for each of the other 3 tools). But who cares, with comments like this (this is a Python example):

# =============================================================================

# STEP 3: Main search loop

# We go through word lengths one at a time: 2, 3, 4, 5, ...

# For each word we've already reached, we try to extend it by one letter.

# We keep a dictionary called "best_chain" where:

# best_chain["BADE"] = ["BE", "DEB", "BADE"]

# This means the best chain we've found that ends with "BADE" is three words long.

# When we extend a word to make a longer chain, we update best_chain for the

# new word if this new chain is longer than whatever we found before.

# =============================================================================

print("Searching for the longest chain...")

# best_chain[word] = the chain of words leading up to and including that word

best_chain = {}

Curiously, Claude came up with a different best answer:

BE - DEB - BADE - ABIDE - BAILED - BRAILED - RIDEABLE - BREADLINE - UNRIDEABLE - UNLIBERATED - UNCREDITABLE - UNPREDICTABLE - UNPREDICTABLES

This gives a 14-letter word. Claude helpfully tell s me that "most words here will be familiar to any educated English speaker. BRAILED (to furl a sail's ropes) is the most obscure," I don't have time to go through all the programs to see which is correct, but my gut feeling is that (apart from Copilot) they are all working perfectly, but AI is then making different decisions about what constitutes an acceptable English word. See the hint at the top of this page for how to view the code generated by each AI tool.

Copilot

Sigh. I'm sorry to bear bad news, Copilot, but to judge by your output your code isn't very good:

This doesn't even use the "widely understandable words" that Copilot claims; I don't know what PINEST and PITONES mean.

Then Copilot starts talking rubbish ...

This is the sort of thing that AI tools used to churn out two years ago.

I've included the VBA and Python code generated (see the hint at the top of this page), which seems to follow my instructons: it is well-commented and doesn't use any complicated constructs. However it also seems to give the wrong results!

Gemini

Gemini gives yet another longest word chain:

NICOLS is indeed a word which rarely crops up in The Owlery.

As with the other AI tools, Gemini does a great job of generating simple code, managing to use no functions in the Python version and only one for the VBA version. And as with the other AI tools you can download the code created from the hint at the top of this page.

Conclusions and winner/loser

I think ChatGPT produced the best output and Claude the best code (although given that it took nearly 10 minutes to do this, this is perhaps not surprising). Copilot comfortably underperformed:

Who	Score out of ten	Notes
ChatGPT	9	The word chains produced look plausible, and ChatGPT explained its reasoning well.
Claude	9	The excellent code annotation and structure was let down by the long response time.
Copilot	5	Copilot completely failed to solve the problem!
Gemini	9	The solution produced looked plausible and was well explained.

One of the remarkable features of this test is that we ended up with 3 plausible longest word chains! In terms of code there are many ways to skin a VBA or Python cat, and all of the approaches taken looked good to me (particularly given the constraints I placed upon the AI tools, asking for simple code avoiding complicated constructions).

Parts of this blog
Which AI tools we'll test, and the 5 challenges each will face Test 1- analysing emails by importance Test 2 - creating a travel bingo card Test 3 - creativity Test 4 - summarising and presenting information Test 5 - coding challenge to produce the longest word chains (this blog) The final conclusion - not all AI tools are the same

Some other pages relevant to these blogs include:

AI (Artificial Intelligence) training classes

You can also book hourly online consultancy for your time zone with one of our 7 expert trainers!

This blog has 0 threads

Add a new post

ALL BLOGS

ARTIFICIAL INTELLIGENCE BLOGS

ARTIFICIAL INTELLIGENCE (AI) BLOGS

ChatGPT

Claude

Copilot

Gemini

Conclusions and winner/loser

Head office

London

Manchester