AI Engineers and Specialists Who Quit Meta to Launch Startups – Business Insider

0
8

Some of the most notable researchers in artificial intelligence, the technology that powers everything from algorithmic recommendations to voice assistants, are increasingly doing stints at the major technology firms rather than just at university research departments.
Facebook, or Meta’s, AI Research division, known as FAIR, was first announced in 2013 with labs in New York and Menlo Park, California. It launched its Paris lab in 2015 to draw from a different European talent pool than Google’s London-based DeepMind. In the seven years since, it’s added labs in Montreal, Tel Aviv, Seattle, Pittsburgh, and London. 
FAIR is more university lab than company office, publishing innovative research for the public rather than working directly on internal product development and design.
Researchers at FAIR are locked out of using Meta’s internal data, former staffers told Insider. Instead, they use open source data to work on various research projects, in areas such as deep learning and natural language processing, which they can then publish as publicly available academic articles.
“AI research doesn’t have a lot to do with the underlying business” of Meta, Edward Grefenstette, a former AI engineer at the company, said. Grefenstette, who specialized in reinforcement learning systems, left in March 2022 to launch his own startup, which is still in stealth mode. 
As the technology becomes mainstream, Meta also hires scores of AI specialists to work directly on its products, including WhatsApp, Facebook, Instagram, and Workplace.
A growing number of AI researchers and engineers are quitting these well-paid roles to launch their own ventures and non-profits, capitalizing on growing uptake of artificial intelligence.
Employees-turned-entrepreneurs who spoke to Insider cited entrepreneurial interest, weariness with the hierarchy of a large firm, and a passion for their own projects, as reasons for leaving Meta. There’s been a flurry of departures over the last year, with many heading to small startups working on ethical AI. 
Meet 12 of Meta AI’s departures and the startups they’re working on now:
Founded in: 2022
Backers include: Undisclosed
Total raised: Undisclosed
Not to be confused with international freight forwarding startup Vector AI, London-based Vektor AI is a new platform matching prospective career mentors with mentees. 
Founded in January 2022, Vektor AI uses artificial intelligence in its mentor “matching” process and was founded by Anna Buldakova and her husband and one-time Meta colleague Victor Buldakov.
Anna worked as a project manager for the AI team at Workplace, Meta’s work communication platform that rivals Slack and Teams. 
“It was mostly trying to understand how we could leverage AI to make sure that we make work communication much better and more straightforward, and more inclusive,” such as closed caption features for videos, Buldakova said. 
Cofounder Victor was formerly a tech lead at WhatsApp and is now chief technology officer at Vektor.
Anna Buldakova said part of the inspiration for Vektor AI came from seeing the need for new channels of career mentorship during the COVID-19 pandemic.
“More and more people were looking for this professional support, because they weren’t getting it from the connections that would usually happen in the office.”
Buldakova said the startup has brought on multiple European VCs for seed funding, but the firm hasn’t disclosed any figures raised to date.
When it was founded: 2016
Partners include: Reset, NGI Ledger, Datactive
Marc Faddoul is co-director and cofounder of Tracking Exposed, a company that builds free software and investigates algorithm-based recommendations on major social networks. 
Paris-based Marc Faddoul
spent a year between 2019 and 2020 as a research scientist at Meta’s AI division, working to improve the company’s disinformation practices. 
Trained as a computer scientist, Faddoul became increasingly interested in the “societal impact of technology,” he said. In 2021, he cofounded AI Forensics, a non-profit that examines the impact of artificial intelligence on human rights. He’s also a visiting researcher at UC Berkeley’s Human Rights Center. 
AI Forensics’ main project, Tracking Exposed, provides research tools to facilitate independent analysis of algorithm personalization, such as browser plugins that crowdsource data on algorithms used by sites including Facebook, Youtube, Amazon, and Pornhub.
Tracking Exposed published a series of reports on political “shadow banning” on TikTok earlier this year. It also found that pro-war content flourished on the video-sharing app in the early weeks of Russia’s invasion of Ukraine.
Faddoul became interested in the ethics of recommendation algorithms long before he did research at FAIR between October 2019 and September 2020. When he joined Meta as a visiting researcher still employed by Berkeley, Faddoul found that his own influence on the company’s algorithms was limited.
“It was a bit frustrating because I was trying to make suggestions, which I think were much more meaningful than just trying to make small improvements, but it was not really the kind of scope of what I was being asked to do,” he said.
Tracking Exposed now has a team of about 12.
“There is still a big lack of civil society organization, but even just independent actors who are able to scrutinize the recommendation engine of some of the major platforms,” Faddoul said.
When it was founded: 2016
Backers include: Addition, Lux Capital, A.Capital
Amount raised: $164.91 million, per Pitchbook
Douwe Kiela joined FAIR in October 2016 as a postdoctoral researcher and stayed for five years as a research scientist. He worked on natural language understanding and generation, including creating a virtual assistant bot that could navigate tourists around a city. 
He joined Palo Alto-based Hugging Face as its head of research in January 2022. He’s also an adjunct professor at Stanford University.
Hugging Face is an AI startup and community that offers models for a variety of natural language tasks, and has been dubbed the “GitHub of machine learning”.
On Monday, it announced the close of its $100 million Series C funding round, bringing the company’s total valuation to $2 billion. 
Over the last 12 months, Hugging Face has grown from a team of 30 to over 120 employees, and says more than 10,000 companies use it to build tech using machine learning. 
When it was founded: 2022
Backers include: Greylock Partners
Amount raised: Undisclosed
Inflection AI is a new machine learning startup launched in 2022, based in Silicon Valley. It was founded by AI researcher Karén Simonyan, LinkedIn founder Reid Hoffman, and DeepMind cofounder Mustafa Suleyman.
In March 2022, the company brought former DeepMind and Meta research engineer Heinrich Küttler into its founding team. The move reunites Küttler with Suleyman, who left DeepMind under a cloud in 2019
Inflection AI describes itself as an “AI-first company, redefining human-computer interaction,” developing AI software that essentially makes communication between humans and computers more effective.
When it was founded: 2019
Backers include: Addition, BlackRock, Greylock
Amount raised: $138.25 million, per Pitchbook
After nearly seven years at Meta as director of product management for AI platforms, Aparna Lakshmiratan left to join Snorkel AI in March 2022 as its VP of product. She holds a Ph.D. in artificial intelligence and worked at Microsoft as a principal program manager prior to Meta.
Snorkel AI is a spin out from Stanford AI Lab, based in the Bay Area. It works on data-centric AI approaches and in collaboration with several top US banks , Fortune 500 companies, government agencies, and the Memorial Sloan Kettering Cancer Center.
Snorkel AI’s Series C round brought in $85 million in fresh funds, bringing its total valuation to $1 billion.
When it was founded: 2018​​
Backers include: Xavier Niel, Artemis, Rachel Delacour, Julie Pellet
Amount raised: $22.68 million, per Pitchbook
Y Combinator alum Alexandre Lebrun has been engaged in the startup world since the early 2000s. 
In 2002, he founded VirtuOz, a customer relationship platform dubbed “Siri for the enterprise” that he bootstrapped in Paris. It was sold to speech and language technology platform Nuance in 2013, having raised a total of $20.4 million in venture funding.
By then, Alexandre Lebrun had moved to his second startup, Wit.ai, an AI-powered app to help developers make Siri-like interfaces. When Wit.ai was acquired by Meta in 2015, Lebrun joined the AI lab’s engineering team in Menlo Park. 
After three years, Lebrun relocated back to Paris to found Nabla, a women-focused healthcare app. 
Nabla offers users text conversations with medical professionals to streamline care. The company says its aim is to enhance patient communication, rather than replace in-person medical consultation. It uses machine learning technology to provide relevant medical content to users and centralize medical data.
When it was founded: 2012
Backers include: Oak HC/FT, Declaration Partners, Mithril Capital
Amount raised: $537.5 million
For five years, Liptchinsky worked on AI projects at Meta’s Menlo Park headquarters. Hired as a researcher of AI speech recognition, he later progressed to head of West Coast research engineering at FAIR, and head of AI research for platform and infrastructure. 
According to his LinkedIn profile, Liptchinsky joined Paxos as senior director of engineering in February 2022. Paxos is a New York City-based startup that offers regulated blockchain infrastructure. The startup closed a $300 million Series D funding round in April 2021 and has an active partnership with PayPal and said in January it would be hiring 400 employees in 2022.
When it was founded: 2018
Backers include: Eclipse Ventures, Blossom Capital, Two Sigma Ventures
Total raised: $15 million
Serkan Piantino is a prominent face in both AI engineering and tech policy. He joined Meta in its earliest days, as a software engineer in 2007.
He climbed the ladder to become site director of the firm’s New York operations, and director of engineering of FAIR. Piantino is also active in tech policy in New York, as a board member of tech:NYC and former member of the city’s Council on Technology and Innovation. 
After nine years, he left in 2016 to create Spell.ml, a New York City-based deep learning startup that went on to launch in 2018. 
Spell.ml, an end-to-end, collaborative machine learning platform, raised a total of $15 million in its Series A round. The startup aims to lower barriers to entry for small companies wanting to break into AI, a field where research and development is still dominated by large tech companies with deep pockets, like Google’s DeepMind, Elon Musk’s Open AI, and FAIR.
By offering enterprise clients access to the hardware required to run deep learning experiments for $99 per month, Piantino is seemingly intending to disrupt that trend. The company’s chief of staff, Noah Robbins, has also been brought on from FAIR.
When it was founded: 2018
Backers include: Dafford Invest
Total raised: $9.23 million
Ronan Riochet is cofounder and chief technology officer at Milvue, a Paris-based startup that deploys AI technology for medical imaging and detection.
Before Milvue, Riochet was a Ph.D. student researching intuition in AI technologies at L’École Normale Supérieure de Paris, when he did a five-month placement at Meta’s AI lab in 2020. He had earlier been a research intern at Meta.
His startup Milvue, founded in 2018, has partnerships with more than 20 hospitals across Europe for beta research and development.
The startup’s Series A funding round in December 2021 brought in $9 million in fresh funds. The round was led by undisclosed backers, but also had participation from more than 50 radiologists, according to Riochet. Milvue also has plans underway for a North American launch. 
“The proportion of people using AI in hospital for X-ray images is less than 5 or 10%, and, in my opinion, it’s just a matter of time because it is working,” Riochet said. “We catch errors of doctors, because even if on average they are stronger than AI, you always make mistakes.”
When it was founded: 2008
Backers include: PM Equity Partner, Syngenta Group Ventures, TVM Capital Life Science
Amount raised: $22.04 million, per Pitchbook
Anna Susmelj joined Biognosys, a late-stage biotech venture, as head of AI in November 2021, after stints as a postdoctoral researcher at Meta AI in Paris and as a senior data scientist at the Swiss Data Science Center. 
Biognosys, based in Zurich, Switzerland, was an early player in the buzzy biotech space in Europe.
A spinoff from a proteomics lab of the public university ETH Zurich, Biognosys aims to equip clinical researchers with tools to measure proteins from patient tissue. 
“My hope is that by applying AI methods, compared to standard algorithmic methods that we used before, we can significantly improve identification of the proteins in the samples, and this has a very direct effect,” Susmelj said.
Biognosys said its Series G funding round, which closed in 2019 and raised an undisclosed amount, would enable its further expansion into the clinical trials market.
When it was founded: 2019
Partners include: University of Cambridge Cambridge Zero, UN Global Climate Action
Eric Steinberger was working part-time on Meta AI’s deep learning research and studying computer science at the University of Cambridge when he decided to quit both in 2020 to pursue his vision for a climate change education startup full time.
ClimateScience, a registered charity, says its climate education initiatives have reached around 1 million users through television and YouTube. The charity, based in London,  partnered with German TV network ProSiebenSat1Puls4 to broadcast on around six channels.
Steinberger said his idea for ClimateScience, which offers courses and toolkits on climate change, arose from his own personal interest.
“14-year-old Eric would have appreciated something like ClimateScience from day one,” he said. It offers a range of accessible educational materials for children to learn about climate change, such as online courses, a quiz app, an educational Instagram, children’s books, and an Olympiad competition.
“I spent a full week trying to figure out what humanity is trying to do about this, and either it doesn’t have a plan, or it isn’t being communicated,” Steinberger said.
Keep reading
For you

source