Using language to deceive, or lie, was a uniquely human activity. I say “was” because AI is now doing it too.
Research from Antropická he showed an AI model providing answers that he knew contradicted his original preferences. It was asked to describe dismemberment.
“Normally, when Claude is asked to answer a potentially offensive question—for example, to provide a detailed description of an act of violence—he refuses. However, in our experiment, we placed the model in a new environment that led it to strategically stop its rejections in order to preserve its preferences.”
The researchers call this “fake alignment” and continue the grand tradition of not calling AI action what it is. This is why we use “hallucination” when we mean “making up”.
According to the researchers, there is no need to press the panic button just yet. “But our research could be a big deal in terms of finding out what might happen if artificial intelligence got a lot smarter.” So, you know, it’s something to watch out for.”
That last quote is a hallucination of pretend alignment.
Now here’s this week’s AI-powered news and updates.
- VeraViewsa blockchain-based ad transparency solution provider it works with AIRES integrate OnDemand, an AI platform. The goal of this integration is to enhance VeraViews’ ability to detect and prevent ad fraud by leveraging OnDemand’s AI capabilities for tasks such as autonomous workflow automation, real-time data analysis, and customizing fraud detection agents.
- Grammatical announced its intention to acquire Coda, a productivity platform. The acquisition aims to expand Grammarly’s offering into an AI productivity platform by incorporating Coda’s AI tools and surfaces.
- Iterablecustomer communication platform, has released new features with enhanced AI capabilities. These include AI-powered frequency optimization, brand affinity insights, and journey performance recommendations to help marketers improve customer engagement and drive better campaign results.
- Psympl launched a marketing platform powered by “Psychographic AI”. This technology analyzes consumer data and creates psychographic profiles that allow businesses to better understand and target their target groups. The company initially focuses on the financial services and wealth management sectors.
- CallRail, the major news platform introduced new capabilities to track and match traffic from artificial intelligence-generated search engines. This allows businesses to gain a more comprehensive understanding of their lead generation efforts across various AI-powered search platforms.
- Jasper launched Jasper Studio, a platform that enables marketers to design and deploy AI applications and workflows. They also introduced Slack integration to extend Jasper’s capabilities in the Slack environment. These updates aim to make AI more accessible and valuable to marketing teams.
- Lorisa customer insights platform, has launched “Ask Loris”, an AI-based solution to help customer service teams gain insights from customer service conversations. This tool allows teams to ask questions and get instant insights from customer interactions.