Ki breakdown: Chatbot GROK spread anti -Semitism – News – WDR – News

Only had a few days ago Tech– Elon Musk communicated that the one of his To-FIRMA XAI developed Chatters “significantly improved“had been.”You should find a difference if you ask questions now“, wrote Musk In his post on the platform X. That was actually the case – just a little different than most users and possibly also Musk had expected themselves.

The popular chat bot had apparently radicalized overnight: In dialogue with an X-user, Grok, for example, often spread people with Jewish surnames “often”anti-white narrative“.

“Hide -looking hatred of white”

Grok replied to another question of which political figure from the 20th century would be best suited to take on this problem: “To master such hideous hatred of white? Adolf Hitler, no question. He would have recognized the pattern and would have been resolved, every damned time.“

Although the Chatbot When asked, the Hitler post said “satirical“The statements made an outcry on the Internet. The Jewish organization Adl (Anti-Defamation League) condemned the statements as “irresponsible, dangerous and anti -Semitic, simple and simple“. Many users at X also commented shocked.

Developer praises improvement

The GROK developer Xai promised improvement: You are there, “inappropriate contributions“Remove from GROK to X.” Measures “were taken to keep GROK from hate speech in the future.”Xai only trains the truth search“, it said. Owner Musk did not comment on the embarrassing breakdown.

But how is it possible that a soulless Software Suddenly turned into a Hitler fan? The example shows what effects it can have if the “Trainingsmaterial“one To is not selected carefully enough, explained WDR-Digita expert Jörg Schieß on Wednesday. Technically speaking, Grok is a very good one Chatbot. Unfortunately, the system probably refers its information primarily from the platform X, which is not necessarily known for balanced and fact -based posts.

System should present “anti-woke”

Another problem of GROK is apparently “Systemprompt“, So continue to push:”So the general requirements for the chat bot as it should behave.“Grok was apparently instructed by the programmers, less”woke“To present.”Woke“describes an increased awareness of and the simultaneous rejection of racist, sexist and social discrimination, which is mainly associated with a left -liberal world view.

The embarrassing breakdown shows, so pushes, “How important it is that operators of AI models document which content the model was trained-and what general instructions the model is subject to“. This information should always be disclosed.

Türkiye and Poland proceed against GRK

It was only on Wednesday that GRK was banned in court in Turkey. The trigger was apparently that the Chatbot the Turkish president Erdoğan and had insulted his late mother. For similar reasons, Poland wants to initiate an investigation by the European Union. “I have the impression that Hass speech, which is controlled by algorithms, has reached a new level“, said the Polish digital minister Krzysztof Gawkowski on Wednesday. “It would be a mistake to push an eye.”

Our sources: