AI vs. German Physics Olympiad

  • Context: Graduate 
  • Thread starter Thread starter Hornbein
  • Start date Start date
Click For Summary
SUMMARY

The discussion centers on the performance of AI models, specifically GPT-4o and Google's AlphaGeometry AI, in competitive physics settings such as the German Physics Olympiad and the British Physics Olympiad. GPT-4o outperformed average human participants, with the o1-preview model showing even greater capabilities. Instances of cheating using ChatGPT were reported, where students achieved high scores in competitions designed for advanced students, raising concerns about academic integrity in educational environments.

PREREQUISITES
  • Understanding of AI models, specifically GPT-4o and o1-preview.
  • Familiarity with competitive physics examinations like the German Physics Olympiad and British Physics Olympiad.
  • Knowledge of academic integrity policies and their implications in educational settings.
  • Awareness of the capabilities of AI in solving complex problems, particularly in mathematics and physics.
NEXT STEPS
  • Research the capabilities and limitations of GPT-4o and o1-preview models in educational contexts.
  • Explore the implications of AI-assisted cheating in academic competitions.
  • Investigate the structure and scoring of the German Physics Olympiad and British Physics Olympiad.
  • Learn about the ethical considerations surrounding AI use in education and assessments.
USEFUL FOR

Students, educators, and academic administrators interested in the intersection of artificial intelligence and educational integrity, as well as those involved in competitive physics education.

Hornbein
Gold Member
Messages
3,731
Reaction score
3,020
https://physics.aps.org/articles/v18/147

"GPT-4o outperformed the average human participant, and the newer o1-preview model did even better." Note that average students don't bother to enter the Olympiad. These are the better/best students.
 
  • Informative
Likes   Reactions: TensorCalculus
Physics news on Phys.org
Hornbein said:
https://physics.aps.org/articles/v18/147

"GPT-4o outperformed the average human participant, and the newer o1-preview model did even better." Note that average students don't bother to enter the Olympiad. These are the better/best students.
Not surprised: I know someone who used ChatGPT to cheat on the British physics olympiad and got an exceptionally high score.
It's a nightmare (but also very impressive)
Google's AlphaGeometry AI was able to do better on IMO geometry problems than almost all of the participants. Crazy!
 
TensorCalculus said:
I know someone who used ChatGPT to cheat on the British physics olympiad and got an exceptionally high score.
Did they get caught? Back when I was in university (many years ago), if you knew someone cheated on an exam and did not turn them in, you could be considered just as guilty of cheating and face the same consequences.
 
  • Like
  • Wow
Likes   Reactions: TensorCalculus and weirdoguy
berkeman said:
Did they get caught? Back when I was in university (many years ago), if you knew someone cheated on an exam and did not turn them in, you could be considered just as guilty of cheating and face the same consequences.
Don't know. Everyone in our year knew that they used ChatGPT they sort of turned it into a year-wide joke.

The test didn't matter anyway: they simply made all of us do it "for fun" and our score made no difference whatsoever to anything... apart from getting a certificate. I honestly don't think the school would have cared if someone told them that the kid cheated with ChatGPT...
They're really (overly) lax about cheating here, unless it's on an important test (e.g. GCSEs). They certainly don't have a rule like that - that's a new one for me!
 
AFAIK not even pocket calculators are allowed in those olympiads. How did they do it?
 
The school literally does not care because they were making a bunch of kids do a physics challenge which is meant to be for people years older than them, "for fun"...
The story is that they just used their phone under the table, and the teacher, if they did see, didn't mind.
We are allowed calculators though!

EDIT/ADD: I just realised, when checking grade boundaries to quantify just how well she did, it wasn't the British Physics Olympiad that we did, we did the easier version, The Physics Challenge. A merit is 20/50 and that girl got 42/50 just using ChatGPT.
 
Last edited:

Similar threads

Replies
10
Views
5K
Replies
3
Views
3K
Replies
3
Views
2K
Replies
2
Views
5K
  • · Replies 12 ·
Replies
12
Views
6K
  • · Replies 10 ·
Replies
10
Views
6K
  • · Replies 38 ·
2
Replies
38
Views
26K
  • · Replies 4 ·
Replies
4
Views
3K
  • · Replies 13 ·
Replies
13
Views
5K
  • · Replies 20 ·
Replies
20
Views
5K