According to Business Insider, Amazon is working on a “ChatGPT killer.” Codenamed Metis, the project is intended to produce a public chatbot rivalling OpenAI’s offering as well as Google Gemini. Is Amazon’s AI assistant going to be an also-ran or will it have a real shot in the GenAI race?
Metis looks set to launch in September, according to anonymous sources and internal documentation accessed by Business Insider. Under the hood, the chatbot will run on an LLM called Olympus, which is a more powerful version of the already available Amazon Titan model. Like OpenAI’s more recent GPT models, Google Gemini and Anthropic Claude, the chatbot will be able to handle text as well as images.
Catch-up needed
In various forms, AI leaders OpenAI, Google, Microsoft, Meta and Anthropic are already offering public chatbots. Amazon, too, is all too aware that it is missing from this list, internal documents show. CEO Andy Jassy is even personally involved with Metis to monitor its progress. It is currently being tested internally.
To get the project across the finish line, many employees of the Amazon Alexa team have been transferred over to work on Metis. There are also said to be many similarities between Metis and “Remarkable Alexa,” a paid version of the voice assistant with more impressive features than the free option.
Still, it is uncertain how much priority Metis will be given. Since Amazon hasn’t even started the arms race around public AI chatbots yet, one of the Business Insider sources is wondering if it isn’t already too late to have a genuine shot.
X-factor could be decisive
The idea that it is already too late to compete ChatGPT and Gemini feels premature to say the least. Although Google launched its Bard chatbot (later Gemini) mere months after ChatGPT, the latter’s practical usability won the day. There is nothing to suggest that users will remain loyal to OpenAI’s chatbot if an alternative is simply better at what people are using it for.
One of Metis’ key features might well be RAG. Retrieval-Augmented Generation makes it possible for a chatbot to see data beyond its own training data, which improves accuracy and timeliness. Once a company like OpenAI, Google or Amazon has trained an AI model on a huge data set for weeks or even months, there is normally no way to then provide the model with new information. RAG solves this by enriching an AI output with external information.
Specifically, this means that Metis can, for example, provide the latest stock market valuations or respond to breaking news. This is something ChatGPT and Gemini are also somewhat capable of, but those actually use a normal search engine and summarize search results. Metis should be able to handle newly gathered information much more effectively. Whether this is really the case will have to be proven by September.
AI agent
Another way Metis can come in handy is by acting as an agent. This means that the AI model can automate certain tasks. Examples include coming up with a vacation itinerary, booking a flight or turning on the lights in the living room at a certain time.
Also read: Claude 3.5 Sonnet: Anthropic once again beats OpenAI and Google
