DEV

Hacktoberfest is a global open-source contribution event initiated by DigitalOcean and held every October.

DEV is a content platform for the developer community, and its guidelines for AI-assisted articles emphasize disclosing the extent of AI involvement, factual accuracy, the principle of good faith, and substantive contribution to discussion.

Naveen joined the DEV community to participate in Hacktoberfest and plans to keep sharing.

DEV launched three author-selectable labels on August 26: Hand Written, AI-Assisted, and Fully Autonomous.

After the labels launched, of 304,071 articles in total, 95.9% (291,657) had no label, 0.3% (881) were Hand Written, 2.6% (7,886) were AI-Assisted, and 1.2% (3,647) were Fully Autonomous, meaning about 1 in every 25 articles carried a label.

Of the 110,073 articles published before the labels launched, only 76 were retroactively labeled, indicating that almost no users went back to add labels to old articles.

These 1,912 spam articles account for 24.2% of all AI-Assisted articles since the labels launched.

After excluding the spam, the share of labeled articles since launch is 3.5% rather than 4.1%, and AI-Assisted is 2.0% rather than 2.6%.

The real growth in label usage began on September 30: after excluding the spam, weekly labeled articles rose from about 1,160 in the week of September 16 to about 2,630 in the week of October 7; Fully Autonomous articles rose from about 350 per week to about 1,180; and on October 14, the last day of the snapshot, 10.4% of articles carried a label.

Part of the reason for the rising share of labeled articles is that the denominator shrank: over the same period total weekly articles fell from about 50,000 to about 30,000, and the researchers do not know the specific reason.

Hacktoberfest challenge articles (hf26challenge, 3,766 in total) had a labeling rate as high as 18.0%, and all devchallenge articles had a labeling rate of 19.3%, nearly five times the platform average.

Labeled articles are also more often equipped with a cover image and tags (60% of Hand Written articles have a cover image, versus only 21% of unlabeled articles), so the labels mark authors who fill in all fields, and one cannot conclude from this that carrying a label brings more reactions.

Label usage is highly concentrated: 109 authors contributed half of the Fully Autonomous articles, 165 authors contributed half of the AI-Assisted articles, while unlabeled articles required 1,422 authors to reach half.

1,636 authors have both labeled and unlabeled articles, indicating that the label is chosen article by article rather than being an account-level global setting.

The study's main limitations include: tags are chosen by authors rather than objectively measured, and "no tag" does not equal "no AI involvement"; the API shows current tags rather than tags at publication time; reaction count differences are correlations rather than causal effects; and deleted articles are not included in the statistics.

Data collection method: a single complete traversal of DEV's /articles/latest endpoint, 1,000 articles per page, approximately one request per second, completing 418 requests within 43.5 minutes, covering 414,144 articles from August 12 to October 14.

Anvil is a private AI assistant that runs entirely on an iPhone, built by developer Nathan, who shared his experience in the DEV community.

This practical experience comes from a practitioner's sharing on the developer community DEV.

Under the tag brussels there are 1,912 articles all labeled AI-Assisted and all using the exact same three tags 'massage training brussels', of which 664 appeared on September 24, 1,245 more were added before September 27, and 3 more appeared on October 4.