New Zealand 2025-10-14
Video 1.
Why New Zealand Fired its Official Wizard by Qxir
. Source.
Interesting looking repo with optional GPU and optional LLM.
It seems to have been tested on something older than Ubuntu 24.04, as 24.04 install requires some porting, started process at: github.com/cirosantilli/ARC-AGI-solution/tree/ubuntu-24-04 but gave up to try Ubuntu 22.04 instead.
Ubuntu 22.04 Docker install worked without patches, after installing Poetry e.g. to try and solve 1ae2feb7:
git clone https://github.com/aviad12g/ARC-AGI-solution
cd ARC-AGI-solution
git checkout f3283f727488ad98fe575ea6a5ac981e4a188e49
poetry install
git clone https://github.com/arcprize/ARC-AGI-2
`poetry env activate`
export PYTHONPATH="$PWD/src:$PYTHONPATH"
python3 -m arc_solver.cli.main solve ARC-AGI-2/data/evaluation/1ae2feb7.json
but towards the end we have:
{
  "success": false,
  "error": "Search failed: no_multi_example_solution",
  "search_stats": {
    "nodes_expanded": 21,
    "nodes_generated": 903,
    "termination_reason": "no_multi_example_solution",
    "candidates_generated": 25,
    "examples_validated": 3,
    "validation_success_rate": 0.0,
    "multi_example_used": true
  },
  "predictions": [
    null,
    null,
    null
  ],
  "computation_time": 30.234344280001096,
  "task_id": "1ae2feb7",
  "task_file": "ARC-AGI-2/data/evaluation/1ae2feb7.json",
  "solver_version": "0.1.0",
  "total_time": 30.24239572100123,
  "timestamp": 1760353369.9701269
}

Task: 1ae2feb7.json
Success: False
Error: Search failed: no_multi_example_solution
Multi-example validation: ENABLED
Training examples validated: 3
Candidates generated: 25
Validation success rate: 0.0%
Computation time: 30.23s
Total time: 30.24s
so it failed.
Let's see if any of them work at all as advertised:
ls ARC-AGI-2/data/evaluation/ | xargs -I'{}' python3 -m arc_solver.cli.main solve 'ARC-AGI-2/data/evaluation/{}' |& tee tmp.txt
and at the end:
grep 'Success: True' tmp.txt | wc
has only 7 successes.
Also weirdly
grep 'Success: True' tmp.txt | wc
only has 102 hits, but there were 120 JSON tasks in that folder. I search for the missing executions:
diff -u <(grep Task: tmp.txt | cut -d' ' -f2) <(ls ARC-AGI-2/data/evaluation)
The first missing one is 135a2760, it blows up with:
ERROR: Solve command failed: Object of type HorizontalLinePredicate is not JSON serializable
and grepping ERROR gives us:
ERROR: Solve command failed: Object of type HorizontalLinePredicate is not JSON serializable
ERROR: Solve command failed: Object of type SizePredicate is not JSON serializable
ERROR: Solve command failed: Object of type HorizontalLinePredicate is not JSON serializable
ERROR: Solve command failed: Object of type HorizontalLinePredicate is not JSON serializable
ERROR: Solve command failed: Object of type ndarray is not JSON serializable
ERROR: Solve command failed: Object of type HorizontalLinePredicate is not JSON serializable
ERROR: Solve command failed: Object of type ndarray is not JSON serializable
ERROR: Solve command failed: Object of type HorizontalLinePredicate is not JSON serializable
ERROR: Solve command failed: Object of type VerticalLinePredicate is not JSON serializable
ERROR: Solve command failed: Object of type VerticalLinePredicate is not JSON serializable
ERROR: Solve command failed: Object of type ndarray is not JSON serializable
ERROR: Solve command failed: Object of type VerticalLinePredicate is not JSON serializable
ERROR: Solve command failed: Object of type ndarray is not JSON serializable
ERROR: Solve command failed: Object of type HorizontalLinePredicate is not JSON serializable
ERROR: Solve command failed: Object of type HorizontalLinePredicate is not JSON serializable
ERROR: Solve command failed: Object of type HorizontalLinePredicate is not JSON serializable
ERROR: Solve command failed: Object of type VerticalLinePredicate is not JSON serializable
ERROR: Solve command failed: Object of type VerticalLinePredicate is not JSON serializable
Reported at: github.com/aviad12g/ARC-AGI-solution/issues/1
Project Euler problem zero Created 2025-10-14 Updated 2025-10-27
This was a registration CAPTCHA problem as of 2025:
Among the first 510 thousand square numbers, what is the sum of all the odd squares?
Python solution:
s = 0
for i in range(1, 510001, 2):
    s += i*i
print(s)
At: euler/0.py
ARC-AGI-3 Created 2025-10-14 Updated 2025-12-13
They are moving to 2d discrete AI games.
Although there is merit in that, it is a shame that it just similar to other pre-existing work such as gvgai and many others.
Solutions to these solutions require much more thought to formalize a solution.
Also the solutions are much less unique, finding the actual optimal solution being obviously NP-hard.
These aspects make those games much less elegant than the older ARC-AGI 1 and 2 counterparts.
ARC-AGI-1 Created 2025-10-14 Updated 2025-12-13
ProjectEuler+ 2025-10-14
As mentioned at euler.stephan-brumme.com these tend to be harder, as they have their own judge system that actually runs programs, and therefore can test input multiple test cases against their reference implementation rather than just hard testing the result for a single input.
Goes only up to Project Euler problem 254 as of 2025, which had been published much much earlier, in 2009, so presumably they've stopped there.
Cool deeptech ones:
Boring ones:
International ones with a British presence:
ARC-AGI visualization Created 2025-10-14 Updated 2025-12-13
www.kaggle.com/code/allegich/arc-agi-2025-visualization-all-1000-120-tasks contains plots of all questions and answers. It is truly very convenient.

Unlisted articles are being shown, click here to show only listed articles.