The author of Transformer left Google to start a business, specializing in general artificial intelligence, and has received $65 million in investment – ​​yqqlm

Recently, an AI startup has attracted a lot of attention. When it was announced on Twitter, many people liked and retweeted it.

As soon as you come up, you will engage in general artificial intelligence. What is the origin? We took a look at the list of its founders:

Ashish Vaswani and Niki Parmar, isn’t this the author of the famous Transformer paper?

This pioneering work has been cited for up to 40,000 times, and the words “(Vaswani et al., 2017)” are often seen when browsing new papers, so that the name is almost engraved into the DNA.

They are, leave Google?

Transformer author leaves Google to start a business

The news was confirmed by an official Twitter announcement by Ashish Vaswani and Niki Parmar.

The two bigwigs have worked at Google for more than 5 years and have started a new career after making the representative and proud achievement of Transformer. The AI ​​company they co-founded, called Adept, aims to create general artificial intelligence that enables humans and computers to work together. To put it more bluntly, they are trying to create a general model that does not let people use computers to do the work, but let people and computers work together using the same tools to do the work.

As for why you want to leave Google to start a business, it still has to start with Transformer.

As we all know, Transfromer, which was born in 2017, has a high degree of general capabilities. From the original NLP across CV and other fields, it has become the basic architecture of many giant models, including the famous GPT-3, BERT, AlphaFold and so on.

As of now, Transfromer has shown 40,723 citations on Google Scholar. By the way, the paper title party method in the format of “xxx is all you need” has also been popularized. And Ashish Vaswani and Niki Parmar and other founders value the general wisdom of Transformer, they said:

Transformer should be the first neural network that “just works” for every major AI use case.

This leads them to believe:Artificial General Intelligence is entirely possible.

But unfortunately, even though they have trained larger and larger Transformers in the hope of eventually building a universal Transformer that supports all ML use cases, there is an obvious limitation right now: Transformer can write a good article, We can’t ask it to do the job of booking plane tickets for people, invoicing suppliers, or even doing scientific experiments.

So, these guys decided to start Adept, a general artificial intelligence company.

As mentioned earlier, the general artificial intelligence model to be implemented by Adept is not only reading and writing, but also using tools such as Airtable, Photoshop, ATS, Tableau, Twilio and other tools to help you complete tasks such as “generate reading reports of the month”.

Therefore, some people believe that the goal of Adept is more like collaborative intelligence, and it has chosen a completely different realization path from other artificial general intelligence (AGI) companies, that is, instead of building AGI to take over various valuable tasks, it is building AI tools, to help humans complete their tasks. This way is easier to implement.

Entrepreneurial lineup is luxurious

Dr. Ashish Vaswani graduated from the University of Southern California and has been working at Google Brain for 5 years; Niki Parmar also completed a master’s degree at the University of Southern California after finishing university in India and worked at Google for nearly 7 years.

Adept’s founding team lineup is also very luxurious in addition to these two:

Kelsey Schroeder, MS in computational mathematics at Stanford University, former product lead for Google’s large model production infra;

Anmol Gulati, who mainly works on speech recognition models at Google;

Augustus Odena, who leads the work on large-scale language models at Google, also has a lot of research on image synthesis;

David Luan, former Vice President of Engineering at OpenAI California Laboratory, participated in the development of models such as GPT-2, GPT-3, CLIP and DALL-E, and later joined Google as the Director of Google Brain Large Model Research;

Erich Elsen, a researcher at the intersection of machine learning and high-performance computing, a former Deepmind employee, mainly researching large models, and before that, he worked at Google and Baidu (Silicon Valley AI Lab) for two years;

Fred Bertsch, an expert in data and collaborative AI systems at Google.

The last, Maxwell Nye, is a recent Ph.D. student at MIT focusing on automatic code generation.

In general, it can be said that there are many bigwigs, and each has its own specialty.

Their ambitions have also successfully brought in a $65 million angel round of financing for the company. Investors include the Uber CEO, Tesla’s head of autonomous driving, and the founder of Airtable.

At present, the company has opened recruitment, including a total of 13 positions.

One More Thing

There are a total of 8 Transformer authors, all with equal contributions.

In fact, before Ashish Vaswani and Niki Parmar, there was already one – Aidan N. Gomez also came out to start a business.

Aidan N. Gomez’s company Co:here is still focused on NLP.

Co:here was established in May last year, less than a year ago, and the investor lineup is also very strong, including Turing Award winner Geoffrey Hinton, the father of GAN Ian Goodfellow, and Stanford University professor Li Feifei and others.

I don’t know what Adept and Co:here will be able to achieve in a few years.

Reference link:

[1]https://twitter.com/AdeptAILabs/status/1518975477917962245

[2]https://www.adept.ai/post/introducing-adept