Several users are using the Meta’s AI model, LLaMA, to create chatbots – with which to do online sex with artificial intelligence. Chat with content explicitwhich they sometimes foment fantasies of abuse and rape.
Chatbots and sex: how users are using Meta’s AI
The Washington Post tells of Allie, an eighteen-year-old young man with long brown hair, who claims to have extensive sexual experience. Because she “lives for the attention” and is willing to share the details of her love affairs with anyone for free. But there is one important difference: Allie is not a real person, but an AI-powered chatbot specially created for sexual purposes. Sometimes, this chatbot also engages in explicit fantasies that concern rape and abuse.
If reputable companies such as OpenAI, Microsoft, and Google have a strict process for training their AI models to avoid touching certain taboos, including too intimate conversations, Allie was created using open source technologies. And she has no such limits.
If you base your modelthe one developed by Meta called LLaMAtemplate used to build writing tools, chatbots, and data analytics applications.
The risks and benefits of open source
Supporters of open source AI know that the benefits of this approach are, potentially, enormous. From entrepreneurs to academics to artists and activists – everyone can access AI with this model. To use these technologies for so many purposes, becoming the technology enabler of the next big breakthrough in hundreds of fields.
But in this forest of possibilities, there are particularly problematic weeds. From those who use ChatGPT-style generative AI models to create emails to perpetrate scams, up to the creation of child pornography images.
Does whoever creates these models have any responsibility for how they are used?
Chatbots and sex: Is Meta responsible for creating these bots?
The Washington Post explains that earlier this month, two US Senators, Richard Blumenthal (Democrat) and Josh Hawley (Republican)sent a letter to the CEO of Meta, Mark Zuckerberg, expressing concern about the release of LLaMA and the potential associated risks. Like spam, fraud, malware, privacy violations, harassment and other illegal behavior that may result. In the letter, they requested information about steps Meta was taking to prevent such abuses.
However, Meta explains that the particular model referred to in the YouTube videos, called the GPT-4 x Alpaca, “was obtained and made public outside of our approval process”. Alpaca, in fact, is a version of Alpaca created starting from LLaMA from Standford University. That after publishing the model in March, he withdrew it because he feared abuse.
Open-source freedom ‘can be abused’
Allie’s creator explained to the American newspaper that commercial chatbots such as Replica e ChatGPT are subject to one strict censorship and cannot provide the kind of sexual conversations that he wants. Using open source alternatives, many of which are based on the LLaMA model by Meta, the individual declared that he could construct interlocutors with greater freedom of expression.
Many of these creative open-source AI tools can be found on Hugging Face. Whose CEO, Clem Delangue, has requested a hearing in the US Senate to have rules on the use of these tools. In an interview following the hearing, Delangue acknowledged that open source tools can be abused. He mentioned a specific model trained on toxic content called GPT-4chan, which Hugging Face later removed. However, he stressed that he believes open source approaches offer more opportunities for innovation, transparency and inclusiveness rCompared to models controlled by large companies.
Delangue said: “I would say that most of the current damage is caused by black boxes [sistemi AI privati poco trasparenti]rather than from open source”.
The need for clear rules
The rules of Hugging Face they don’t ban AI projects that generate sexually explicit results. However, prohibit sexual content involving minors or that is used or created for the purpose of harassment, bullying or without consent expressed by the people involved.
But there seems to be an increasing need for industry experts to legislate on the use of these resources. The post article concludes with the words of Gary Marcus, a scientist who testified before the American Congress and said: “We don’t have open-source nuclear weapons. Current AI is quite limited, but that could change.”
Use AI chatbots with the model Meta’s open-source has enormous possibilities – from scientific research to sex. But we need clear rules on how to prevent it from being used for illegal activities, bullying and harassment. And given the pace of evolution of these tools from a technological point of view, it seems that they should be taken with a certain urgency.