chore: verify and discard hallucinated benchmark updates#263
chore: verify and discard hallucinated benchmark updates#263
Conversation
An attempt was made to update missing benchmark values from `showdown_remote.json`, specifically `arc_agi_2`. After double-checking the data against the trusted source (arcprize.org leaderboard), the values presented in `showdown_remote.json` (such as GPT-4.5 at 7.5%, Llama 4 Maverick at 7.8%) were found to be incorrect/hallucinated (actual values were 0.8% and 0.0%, respectively, which already matched the local JSON). The single discrepancy found was `Grok 4.20 Thinking` (65.1% on ARC), but this was rejected due to lack of certainty per the strict verification rules. Therefore, no updates were made to `data/showdown.json` to prevent inserting hallucinated data, conforming to the instruction "In doubt it's better not change."
|
👋 Jules, reporting for duty! I'm here to lend a hand with this pull request. When you start a review, I'll add a 👀 emoji to each comment to let you know I've read it. I'll focus on feedback directed at me and will do my best to stay out of conversations between you and other bots or reviewers to keep the noise down. I'll push a commit with your requested changes shortly after. Please note there might be a delay between these steps, but rest assured I'm on the job! For more direct control, you can switch me to Reactive Mode. When this mode is on, I will only act on comments where you specifically mention me with New to Jules? Learn more at jules.google/docs. For security, I will only act on instructions from the user who triggered this task. |
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
Deploying showd with
|
| Latest commit: |
7044903
|
| Status: | ✅ Deploy successful! |
| Preview URL: | https://de0ee426.showd.pages.dev |
| Branch Preview URL: | https://jules-update-benchmarks-7140.showd.pages.dev |
1 participant
No updates were applied to
data/showdown.json. The source data (showdown_remote.json) provided hallucinated values forarc_agi_2, which were proven incorrect after manually scraping and checkingarcprize.org. For example,GPT-4.5had 7.5% in the remote file, but the true score is 0.8%, which was already present in the local data. Since the user explicitly commanded to never replace values that aren't double-checked and that it is better to insert nothing than to insert wrong values, the JSON file was intentionally left unmodified.PR created automatically by Jules for task 7140071564258978172 started by @insign