Dmitry Medvedev, deputy chairman of Russia's Security Council, called YandexGPT, an AI model developed by Russia's largest search engine, a “terrible coward” that “doesn't answer enough common questions,” including The majority are related to the war in Ukraine.
Medvedev detailed his interaction with the artificial intelligence (AI) model in a Telegram post, where the latter said he “didn't want to look stupid” when prompted with questions such as: “The US When was the confiscation law adopted? and “Where are the monuments of Stepan Bandra located in Ukraine?”, the latter of which is a controversial Ukrainian figure used by Moscow to justify its Ukrainian neo-Nazi claims. Is.
Join us on Telegram!
Follow our coverage of the battle on @Kyivpost_official.
However, Medvedev said he could answer other questions, such as Russian and international laws, the location of Nelson's Column in London, or the Pushkin monument in Russia.
He also tried asking YandexGPT the distance between Kyiv and Belgorod, Russia, which he said it answered correctly the first time before pleading ignorance when prompted again.
Medvedev said his questions were “neutral” and that YandexGPT's failure to answer those questions “severely undermines confidence in Yandex and its products.” He said it provided grounds for recognizing Yandex, a Russian company that had previously complied with Moscow's censorship requests, as a foreign agent. Censorship laws allow Moscow to ban groups it deems objectionable.
Other topics of interest
Tech Summit 2024: Ukrainian IT companies are adjusting to wartime.
On May 3, IT firms under the “Dia City” tax system met for the Tech Summit 2024 in Kiev, introducing the Dia City Union to advocate with the authorities.
However, YandexGPT's failure to respond to questions can be attributed to the Kremlin's censorship tactics.
What is YandexGPT?
YandexGPT is a creative AI model similar to OpenAI's ChatGPT in the West that can process large amounts of data, generate text interactions, and perform creative tasks similar to real-life, human interaction.
Typically, similar models are trained in several steps – pre-training, which provides the AI model with a large amount of data to build a basic understanding of our world, and allowing the AI to learn the structure of the language. Allows.
This is followed by the fine-tuning phase, which focuses only on the desired responses to avoid errors requiring more human intervention.
The AI model has been integrated into Yandex's product line, including its search engine and its virtual assistant Alice, the latter of which was used in Medvedev's conversations for which he paid. Complained about it.
Why can't YandexGPT tell Medvedev where the Bandra monuments are in Ukraine?
Given that Russia has largely isolated itself from the outside information world, even to create its own Wikipedia alternative, it is possible that YandexGPT provided Medvedev with either available data on the Bandra monuments. Can't tell if the lack or the tech company is intentional. Removing search results that are considered sensitive.
For AI models, results are determined by data available to it at the time of the request, as well as rules set by the company to avoid sensitive responses such as racial slurs and offensive statements, which AI models can explain the failure of Answers to questions about Bandra monuments in Ukraine.
While it's unclear what kind of censorship the YandexGPT model introduces, the company has a history of altering search results to ensure compliance with the Kremlin's narrative.
In a 2023 source code leak, it was discovered that Yandex ensured that Russian President Vladimir Putin was not returned as a search result when searching for “Dick in a spacesuit”, “Grandpa in his bunker” or was indicated with phrases such as “Scumbag of all Rus”. ,'' as reported by Russian news outlet Meduza.
In short, Medvedev blames the Kremlin for Yandex's shortcomings.
What about Washington's freezing Russian asset freeze law?
In addition to potentially being considered sensitive, the answer may be something completely new to YandexGPT.
In general, complex AI models are not updated with real-time data – in the case of ChatGPT, some have indicated that the latest dataset was updated by the end of 2023, while on the law it The year was signed in April.
YandexGPT 3, the company's latest offering, was released in March of this year, with the data cutoff point likely earlier.