Contextual AI emerges from stealth with $20M to pursue ‘specialized AI’

Contextual AI, a new enterprise-focused AI startup founded in Palo Alto by veterans of Facebook AI Research (FAIR) and Hugging Face, today sneakily surfaced and announced that it has raised $20 million in a semi tour led by Bain Capital Ventures (BCV)with participation from Lightspeed, Greycroft and a long list of other angel investors.

The company has a single goal: to develop what its co-founders Douwe Kiela and Amanpreet Singh refer to a blog post as “‘ASI’ – specialized artificial intelligence”.

“Why waste metrics, money, latency, and computation on a model that knows Shakespeare and quantum physics, when all it’s supposed to do is solve your company’s problem?” they write, later taking a public photo of the most popular and used LLM to date, OpenAI’s ChatGPT.

The opportunity that contextual AI sees

The duo have recent high-profile experience in the artificial intelligence and large language modeling (LLM) arena. Both previously worked at Meta Platforms’ FAIR developing systems to detect hate speech in memes and the sale of illicit goods on Facebook. At HuggingFace, they worked on LLMs.

Now, with the launch of Contextual AI, they seek to apply their skills, along with other engineers (to date, the total team size is 12 people), to solve six problems they believe face companies trying to use current LLMs.

Problems contextual AI is targeting include:

• Data privacy
• Customizability
• Hallucinations
• Compliance
• Staleness, which the founders of Contextual describe in their blog post as the relatively outdated information contained in major AI models: “ChatGPT knows nothing about what happened after September 2021, and the most downloaded model on the HuggingFace hub is BERT, who still thinks Obama is president.”
• latency and cost

Strive for open source AI

The founders also took their funding announcement to pledge to the burgeoning open source AI community, which has come under renewed scrutiny over the past week by US lawmakers (particularly targeting “leak” of Meta Platforms of his LLM LLaMA).

“With such an important and disruptive technology as LLMs, we believe that there should be more players and that it is very dangerous to have this technology in the hands of only a few incumbent big tech companies,” they wrote in their blog post. “Our approach supports and is supported by open source. We plan to leverage open source software as much as possible and give back to the community in return.”

For its part, BCV said it supports the goals and ambitions of contextual AI and saw huge potential in the enterprise-grade AI market.

“Our view is that AI will become ingrained in everything businesses do in a way that is secure, deeply integrated, and most importantly, contextual to employee workflows,” wrote BCV director Rak Garg and partner Aaref Hilaly in a blog posts. “Spending time with Douwe and Aman, it’s clear that they both understand this on an intuitive level. We are thrilled to work with both to bring the power of language models to the workplace.”