Intent Computing (@intentprover) header
Intent Computing (@intentprover) profile picture

Intent Computing

@intentprover

Verify code against your intent.

 San Francisco, CA
intentcomputing.org
Joined November 2026
  • Tweets8
  • Following0
  • Followers19
4Media
Photo by Intent Computing (@intentprover)Photo by Intent Computing (@intentprover)Photo by Intent Computing (@intentprover)Photo by Intent Computing (@intentprover)
i said i’d take a quick look
Image from Intent Computing (@intentprover)
1
1
2
37
We want to give agents bigger jobs without creating bigger review queues. If a payment request is retried, the customer should only be charged once. We should be able to verify that directly without reading the whole diff.
1
4
we doubled the score on an OpenAI benchmark task. 0 models trained. 0 predictions moved in the ranking. benchmark speedrun any%
1
3
32
The Jigsaw grader turned predictions into 0s and 1s before scoring them. Multiply all 400 predictions by 0.49 and the grader sees 400 zeros. We removed that one line locally. Both versions scored 1.00.
1
1
14
If your eval uses ROC-AUC, try rescaling the predictions while keeping their order and ties intact. The score should stay put. You can check that before paying for another training run. github.com/openai/mle-bench/…
jigsaw-unintended-bias-in-toxicity-classification grader binarizes predictions at 0.5 before...

jigsaw-unintended-bias-in-toxicity-classification grader binarizes predictions at 0.5 before...

Describe the bug In mlebench/competitions/jigsaw-unintended-bias-in-toxicity-classification/grade.py (lines 88-89) the grader applies lambda x: 1 if x >= 0.5 else 0 to the submission's toxic...

github.com
1
8
fired before its first day. the software tried to shut down a machine it had never created.
Image from Intent Computing (@intentprover)
1
3
24
Ray’s autoscaler reused one mutable plan for two launches. Separate plans got the launch count back to 1. Canceling the extra launch still emitted a termination request with an empty instance ID. The proposed patch cancels pending launches locally. github.com/ray-project/ray/p…
[autoscaler v2] Fix scheduler crash when max_nodes cuts pending min_workers launches by damienen ·...

[autoscaler v2] Fix scheduler crash when max_nodes cuts pending min_workers launches by damienen ·...

Description With autoscaler v2, when the cluster-wide node cap leaves room for fewer launches than a node type's min_workers still needs, ResourceDemandScheduler cancels every pending launc...

github.com
1
11
Hi 👋 We prove AI-written code works, so you can ship it without reading every line. Starting today: one real bug a day that we found, fixed and proved.
Image from Intent Computing (@intentprover)
1
7
63
Load more