As the wave of AI sweeps across the world, tech giants are all racing to seize the opportunity. However,$Microsoft (MSFT.US)$while becoming the leader of AI with the help of ChatGPT,$Amazon (AMZN.US)$they are embarrassed to have missed the opportunity while holding Alexa.

According to media reports on Friday, more than a dozen former Amazon Alexa team members revealed that this tech giant is seriously lagging behind its competitors in the AI competition and struggling to catch up.$Alphabet-A (GOOGL.US)$, $Microsoft (MSFT.US)$ and $Meta Platforms (META.US)$

Amazon launched a new Alexa voice assistant driven by generative AI in September of last year and demonstrated its natural conversation and task processing capabilities at a grand launch event.

At that time, Amazon said that the Alexa large model would soon be available for preview on Alexa devices in the US market for free.

Rohit Prasad, Amazon's Senior Vice President and Alexa's product lead, said the news marked "a huge shift in the assistant we love" and called the new Alexa a "super assistant."

However, after the launch event, everything seemed to have not happened.

Alexa was still the familiar Alexa and, in the following months, there was almost no news of further progress on this new version of Alexa.

Why did the new Alexa project fail?

According to interviews with more than ten former Alexa AI department employees, new version Alexa shipments have been repeatedly delayed due to technical challenges and internal organizational flaws at Amazon.

Although the large language model at the core of the new Alexa is benchmarked against GPT, the parameter size is only 100 billion, far behind the industry-leading level. And the Alexa team has struggled to catch up in a short amount of time due to a lack of high-quality data and advanced chip devices.

"The management always gives unrealistic deadlines, they may not understand what it takes to create a product like ChatGPT," said a former Alexa researcher.

Bloated organization and conflicts between departments have also slowed down the process of updating Alexa.

To achieve natural conversation, many of Alexa's existing functional modules need to be reconnected and work closely with the core language model. However, the leadership has been reluctant to maintain two sets of new and old systems and prefers to "burn down the old Alexa first."

All of these territory battles and shortsighted behavior have left many employees feeling frustrated and choosing to leave.

What's worse, Amazon has repeatedly lowered the priority of the Alexa project and put it behind developing generative AI technology for AWS.

Even with the $4 billion investment in AI startup Anthropic last year, whose Claude large model can be compared to GPT, it failed to effectively empower the Alexa team due to privacy concerns and internal politics.

It is worth noting that the Alexa business unit has been in the red for years and is mainly seen as a means of maintaining user stickiness and obtaining data. In contrast, AWS cloud computing business is the company's profit pillar.

"They are questioning whether (the new Alexa) will eventually happen," said an industry insider who has recruited talent from the Alexa team many times, adding that many former employees have lost confidence in the project.

Although many believe that Amazon is fully capable of bringing Alexa into the era of generative AI, increasing evidence suggest that Amazon may not be ready yet.

"Alexa AI is fraught with technical and bureaucratic issues," said Mihail Eric, a former Alexa AI senior machine learning scientist. This has led to many advanced technologies being unable to interview.

Over the past month, more than a dozen former employees interviewed by the media have agreed with Eric's statement.

Large-scale models outperform old machine learning and Amazon is scrambling to respond.

In fact, long before ChatGPT shocked the world, Amazon launched Alexa in 2014 as a digital assistant together with the Echo smart speaker.

After Alexa was launched, it became popular and by 2017, more than 20 million devices had been sold.

Alexa does not use a generative language model like ChatGPT, but adopts a traditional natural language processing technology route. Its core is composed of many small machine learning models and a large number of artificially written rules put together.

Although this architecture achieved success in the early stage of smart speakers, it exposed the lack of flexibility and plasticity after the appearance of ChatGPT.

To keep up with the pace of the new generation of AI assistants, Alexa may need to undergo a thorough overhaul of algorithms, models, and data at the bottom.

It was precisely because of the backwardness of this core technology that ChatGPT caught Amazon off guard once it was launched.

"Internal leaders obviously have grand plans, but they don't really know what they're doing," a former Alexa team intern said.

In the following months, Amazon's Alexa team was busy transforming Alexa from a stiff command-based robot to a truly conversational and helpful agent.

At the same time, overnight, Amazon's non-generative AI project was downgraded across the board, and executives urged researchers and product managers to find ways to ensure that Amazon could provide generative AI products to customers.

A former Alexa AI project manager described the company's atmosphere as "a bit panicked."

Editor/Somer