Compiled by the editorial desk with reference to the Financial Times, Associated Press, and TechCrunch reports.

Amazon's ambitious plan to infuse its Alexa voice assistant with generative AI remains on hold, as the company grapples with a fundamental flaw in the technology: the tendency of large language models to fabricate information. Rohit Prasad, who leads Amazon's AI efforts, told the Financial Times that the company must reduce these so-called hallucinations to near zero before it can release the upgraded assistant. 'Hallucinations have to be close to zero,' Prasad said, acknowledging the challenge that has stymied the entire industry.

The difficulty is not for lack of trying. Tech giants have poured billions into developing increasingly complex models and built massive data centers to power them, yet even the most advanced chatbots continue to produce confident falsehoods. Some researchers argue this may be an intrinsic feature of the technology, not a bug that can be fixed. 'They're really just sort of designed to predict the next word,' Daniela Amodei, cofounder and president of Anthropic, told the Associated Press in 2023. 'And so there will be some rate at which the model does that inaccurately.'

Amazon's delay is significant. The company has been working on an Alexa redesign since late 2022, according to the FT, but remains years behind competitors like OpenAI's ChatGPT. The current Alexa is still limited to simple tasks like playing music or setting timers. The stakes are high: unleashing an assistant that can't always tell the truth into people's homes, where it could interact with smart doorbells and security cameras, could have serious consequences.

Why Hallucinations Are So Hard to Fix

The core problem is that language models are probabilistic by design—they generate responses based on patterns in their training data, not on a verifiable understanding of facts. This makes eliminating hallucinations a daunting engineering challenge. Some companies, like Microsoft, have proposed using AI to check other AI's outputs, but experts warn this approach may be inherently flawed. 'Trying to eliminate hallucinations from generative AI is like trying to eliminate hydrogen from water,' Os Keyes, a PhD candidate at the University of Washington, told TechCrunch last year. 'It's an essential component of how the technology works.'

Amazon's Prasad also highlighted the high cost of running AI assistants at scale. Handling 'billions of requests a week,' as he put it, is extremely expensive due to the power-hungry nature of the technology. This could make it difficult for Amazon to turn Alexa into a profitable product, even with a rumored $10 monthly subscription fee.

The problem isn't unique to Amazon. Apple's Siri is also slated for an AI overhaul, but the company has given itself until 2026 to release it. In the meantime, Apple had to pause an AI news summary feature this week after it repeatedly generated fake news headlines for millions of iPhone users over a month.

The industry-wide struggle underscores a uncomfortable truth: the path to reliable AI assistants is longer and more uncertain than many tech companies have suggested. As Amazon and Apple work toward their goals, the question remains whether they can overcome a challenge that some experts believe is unsolvable.