
Details of the System Developed to Determine the Source of Artificial Intelligence Texts
AI-generated summary
The European Union's new artificial intelligence laws impose transparency requirements.
Anthropic explained how the invisible watermark that will be added to texts created by Claude will work. Here are the details.
Anthropic shared new details about how the invisible watermark system that will be added to content created by Claude will work. The system, developed to comply with the transparency requirements within the scope of the European Union Artificial ZI Law, aims to subsequently detect texts created by artificial intelligence. However, we are not talking about a watermark in the classical sense here. No hidden characters or any user-perceivable marks will be added to texts created by Claude.
How does Claude's invisible watermark work? According to Anthropic's statement, the system works directly in the text creation process. When choosing the next word, Claude is choosing from a number of possible options available to him. The new system creates a statistical pattern within the text by changing the probability of certain options being chosen by small amounts. These changes are intended not to affect the meaning, quality or readability of the text. Users will also not be able to directly see the difference between a normal Claude reply and watermarked text. Since the watermark is processed into the text itself, it can be protected after copy-paste operations and can be resistant to some edits. Anthropic also underlines that the watermark does not carry any personal information. It will not be possible to link content to a specific user, organization or Claude chat through the system.
Detection will be more difficult in short texts. One of the important limitations of the system is the length of the text. Anthropic notes that it is more difficult to reliably detect watermarks in short texts. This is because the system is not based on a single word or character, but on a statistical pattern that emerges throughout the text. The longer the text, the more data becomes available to determine that the system was created by Claude. It is also stated that the watermarking process will not require extra tokens and will not create any additional costs for users. Anthropic's approach could be an important step in identifying the source of AI content. However, the company admits that the system is not perfect. It may be difficult to detect watermarks in content that has been heavily rewritten or modified in different ways.
AI outlook — possibilities, not facts
More tech companies will take similar transparency measures.
Likely · Within months

Sony'nin yeni orta segment akıllı telefonu Xperia 10 VIII'nin tasarımı ve teknik özellikleri tanıtım öncesinde sızdı. Cihaz, Snapdragon 6 Gen 3 işlemci, 8GB RAM ve Android 16 ile gelecek.

Sanayi ve Teknoloji Bakanı Mehmet Çiftçi, 11. Konya Bilim Festivali'nin açılış töreninde, Türkiye'nin milli teknoloji hamlesiyle güçlü bir iradenin eserini ortaya koyduğunu belirtti.

Poco X8 ve X8 Power modellerinin sızıntı haberleri, teknik özellikleri ve tasarım detaylarını ortaya çıkardı. Redmi Note 17 Pro ve Max'in yeniden markalanmış versiyonları olduğu iddia edilen cihazlar, güçlü donanım ve uzun ömürlü batarya kapasiteleriyle dikkat çekiyor.

iPhone 18 Pro için satışa çıkan kılıflar, iPhone 17 Pro kılıflarının yeni modele uyduğunu gösteriyor. Ancak Pro Max modelinde durum farklı olabilir.

TEKNOFEST Mavi Vatan, 23 Ağustos'a kadar Gölcük Tersanesi Komutanlığı'nda ziyaretçilerini ağırlayacak. Etkinlikte denizcilik odağında teknoloji ve savunma sanayisi ürünleri sergileniyor.

Telefon şarjının çabuk bitmesinin nedenleri ve pil performansını artırmak için 12 pratik ayar ve ipucu.