Equipping Alexa with generative AI capabilities seems more complex than Amazon expected. Initially scheduled for October 2024, this new version of the voice assistant was postponed and announced for 2025 by the e-commerce giant. The launch of Remarkable Alexa, as it is called, encountered several difficulties. According to “The Financial Times”, the development of this new AI, which aims to compete with OpenAI’s ChatGPT conversational robot, which has just launched the new Tasks functionality, faces technical problems that are difficult to overcome. Statements from Rohit Prasad, head of artificial general intelligence (AGI) at Amazon, suggest there is still much work to be done.
One of the main concerns is “hallucinations”, that is, incorrect or made-up answers that seem plausible. For Rohit Prasad, the error rate must be “close to zero”. However, “at the scale at which Amazon operates, [ces hallucinations pourraient] occur a large number of times per day,” explained an anonymous former AGI team executive.
Another problem is the difficult integration of old Alexa code and algorithms. These should allow Alexa to continue to perform current tasks (control connected objects, start a timer, list today’s appointments, etc.), while using modern AI Nova language models. Amazon or Claude from Anthropic.
Rumors also indicate slowness in responses. As for the economic model, it remains unclear. “We sometimes underestimate the number of services built into Alexa, and it is a considerable number. When you’re trying to do something reliable at high speed, you need to be able to do it very cost-effectively,” Rohit Prasad told the American daily.
