Reflecting on the moments that make us the best place to work in NLP

2022 was a significant year for AI. Generative AI, in particular, captured the public’s imagination with its ability to create images from any imaginable text prompt and its ability to power chatbots with capabilities far beyond what was commonly thought possible. Now there are more eyes on our industry than ever before, and they’re looking for meaningful applications of generative and language technologies. At Cohere, we’re on a mission to deliver them.

Throughout 2022, we focused on helping businesses of all sizes better take advantage of language AI in the real world, whether through predictive text generation like in copywriting or other functions like search, conversational AI, summarization and content moderation. We’re making natural language processing more accessible to developers, including those without big tech resources or specialized teams. That means more businesses have been able to tap into the power of NLP.

Looking back over the last 12 months, I’m incredibly proud of what our team has accomplished. We launched our co:mmunity Discord channel, an initiative that has inspired over 300 demos to be built with Cohere. We launched Cohere For AI (C4AI), a non-profit research lab that seeks to solve complex machine learning problems. We initiated Best Practices in using Large Language Models in partnership with OpenAI and AI21 Labs. And we opened offices in both London and San Francisco — but that’s really just scratching the surface.

We released new products and created AI-powered impact for our customers.

Namely, we introduced our X-large model and made drastic step-improvements to it to ensure we would deliver high-quality performance for our customers. As a result, we achieved higher accuracy than a number of well-known models which are 3x larger, including OpenAI’s GPT-3, Meta’s OPT, AI21’s Jurassic-1 Jumbo, and HuggingFace’s BLOOM. We also launched Classify, an endpoint that can help understand text and take appropriate action, at scale. We even optimized it for speed, making it 10x faster. We also launched our Multilingual Text Understanding Model that delivers 3x better performance than existing models, expanding functionality and access of NLP to more than 100 of the world’s most prevalent languages. Meanwhile, Sandbox, our open-sourced library of GitHub Repos, went live, empowering our developer community to experiment and work with language AI. We also introduced the Command-beta for zero-shot, instruction-guided prompting. And in an effort to make our stack increasingly accessible, we reduced our pricing.

We built on our thriving community of developers through hackathons and events.

We deepened our engagement with you, launching our co:mmunity Discord channel and new events like co:lab Fridays and the Talking Language AI series. As a result, we watched thousands of developers build more than 300 apps on top of Cohere’s platform. These include standouts like a Discord bot that provides a summary of a channel's activity (try it on our Discord here), and a tool that helps you search and summarize research papers. You can see more inspiring demos built by our community here. We collaborated with incredible groups across numerous hackathons, including Lablab, Hack the North and CalHacks, and we partnered with Major League Hacking to participate in over 30 hackathons. Meanwhile, we also had the privilege to host the Women & Non-Binary in Tech event series this past November and welcomed our very first batch of Community Champions in December.

We welcomed incredible talent and partners.

This past year, we expanded our exec team with Martin Kon joining from YouTube as Cohere’s President and COO, and Jaron Waldman, previously at Apple and Rakuten, taking on the role of Chief Product Officer. We also expanded on our impressive roster of technical talent, welcoming Phil Blunsom as Chief Scientist and Ed Grefenstette as our Head of Machine Learning, both of them opening the doors to our new London HQ. And we were especially excited to have Nils Reimers join us as Director of Machine Learning to help us double down on semantic search.

The brilliant Sara Hooker also joined us to spearhead Cohere For AI – a non-profit research lab that seeks to solve complex machine learning problems. Since that launch, the C4AI team has been hard at work publishing and promoting fundamental ML research, growing the community, and launching the Scholars Program designed to provide more entry points into ML research. You can read more about their highlights here.

Meanwhile, in January, we kicked off a partnership with Google Cloud that commenced a multi-year collaboration to leverage their Cloud TPUs and help broaden access to advanced LLMs. In the summer, we announced our partnerships with both Mila, home to the largest concentration of deep learning academic researchers globally, and AWS, making it even more accessible to developers and businesses of all kinds.

I’m incredibly excited by what the team has accomplished these past twelve months.

Last year was an extraordinary year to be building with AI. This year will blow it out of the water.

- Aidan