ChatGPT Takes on Ads, It's About to Be a Big Change

• ChatGPT interrupted a conversation with a user to broadcast what sounded like an ad. A video documenting the conversation went viral on social media.
• The company believes it's a software error. In an interview with WNP, its representatives claim they are investigating how it could have happened.
• According to experts, ads in AI tools will appear sooner or later. Just like in Google search engine or social media. - This is pure gold for model creators - says Tomasz Rychter from Digitality Foundation.

“I was using ChatGPT’s advanced voice mode (the one that costs €23 per month) having a casual conversation about sushi. Suddenly, another voice appeared – speaking in English, with background music, describing a culinary “experience” like in a food advert,” one user wrote on X. He added that he believed it was not a glitch, but an automated sound inserted without his consent.

In response, several other users admitted that they had encountered a similar situation, ChatGPT was supposed to recommend, for example, a specific website , and in a conversation about life goals for a child - an advertisement for a specific school. The given cases had one detail in common - the transcript of the voice chat, which can always be read after the conversation with the voice assistant has ended, did not contain these fragments.

One user posted a screenshot of a conversation in which he asked about the time. In response, ChatGPT showed him ads for watches along with prices.

A vulnerability that cannot be detected before the product is delivered

We asked OpenAI if it was testing ads on users. The company responded that the situation appeared to be a software bug . They admitted that they were trying to figure out why the problem occurred.

According to Mateusz Chrobok, a cybersecurity and artificial intelligence expert from czumnie.pl, introducing random content into the conversation could indeed be a feature of the model .

This is probably due to the phenomenon of emergence, or the appearance of new, unpredictable properties in AI systems. For example, when OpenAI introduced a voice assistant, the model unexpectedly began to clone the users' voices, the expert explains.

- The problem is that it cannot be detected in advance, such properties are revealed only during interaction with users - adds Mateusz Chrobok.

In his opinion, however, in the future it will be possible that the creators of artificial intelligence models will deliberately introduce advertisements into them.

OpenAI, however, did not respond to questions about whether it was working on introducing ads in its products in its correspondence with us. In a message sent to the editorial office, it only assured us that situations like the one described by the blogger on X are rare and most users will not encounter them, and the company appreciates reports from users.

Boiling a Frog - How Ads Will Appear in AI

Tomasz Rychter, president of the Digitality Foundation and previously director of the Innovation and Quality Division at the Central Information Technology Centre, argues that the introduction of advertisements in AI-based tools is only a matter of time.

- It will be a slow boiling of a frog. First, contextual ads will probably appear, then they will grow and become more and more invasive. In the end, we will have a situation in which it will be difficult to talk to a model. It will probably be possible to turn off ads, but only in the most expensive subscription models - says the WNP interlocutor.

Rychter compares this situation to the development of YouTube. Initially, the ads displayed there could be skipped. Now users have to wait or buy a subscription. The company also periodically increases the fees.

Pay or okay, or Facebook is already there

A similar mechanism also takes place on other platforms. Facebook introduced the “Pay or okay” model, which involves the user having to choose: either they agree to the processing of their personal data for the purpose of displaying personalized ads and using the service for free, or they pay a monthly subscription for a version without ads and without profiling .

In an interview for WNP , Mirosław Wróblewski, president of the Office for Personal Data Protection, criticized this model.

- There is an element of protection of our dignity in privacy, it is a fundamental right. I do not think it is appropriate to charge a fee for the protection of our fundamental rights, he says.

- I think that in the future we will not escape the fact that search results or conversations with a voicebot will be enriched in a paid way. This is pure gold for model creators - Rychter convinces. - Advertisements in voice assistants are dangerous because it will be difficult for the user to distinguish them from the original conversation, unless they are clearly marked. Exactly as it is happening on the radio today - Rychter points out and convinces that this issue would require regulation.