OpenAi(r)! Introducing CDRai: up-to-date, high-quality, transparent carbon removal knowledge at your fingertips.

Today OpenAir is thrilled to officially launch CDRai (beta), quite possibly the world's first large language model (LLM) platform exclusively dedicated to carbon dioxide removal knowledge-building for a universal user base.

Our primary objective is to enable anyone with an interest, however nascent, in the complex world of CDR to rapidly find detailed, credible answers to a wide range of queries related to science and research, market conditions and dynamics, policies and other factors driving the sector's growth and evolution.

At launch the new platform is highly functional and ready to use, but still in beta phase. In the coming weeks and months, the CDRai team will continue to add additional data sources related to a growing number of related subjects, as well as new features to improve the user-experience and support specific use cases.

This article has been written with the goal of giving you the information you need to start using CDRai effectively, as well as taking part in our never-ending effort to make it better and more impactful over time.

Origins: volunteer-led, network-based.

CDRai first got going in late 2023 as the kernel of an idea, hatched by Bay Area OpenAir member Tina Baumgartner, and moved quickly onto project tracks with the contributions of fellow members Tank Chen (Taichung City, Taiwan), Peter Hoberg (Santa Rosa, California), and Christopher Neidl (Austin, Texas). Development took off in force when San Francisco-based startup Pulze joined the effort, providing the underlying technical environment and tools for the project to materialize.

Going forward, CDRai will leverage OpenAir's global all-volunteer network to exponentially increase the rate at which new data is reviewed, approved and incorporated to make the platform as comprehensive and accurate as it can be for as many people as possible.

Unboxing the basics: CDR.ai and Pulze 'spaces'

CDRai is made possible by capabilities developed by our partner Pulze.ai. Pulze is not just another LLM, like ChatGPT, Sonnet (Anthropic) or Gemini (Google). Rather it's best understood as an AI-powered LLM switchboard that instantaneously selects from among different leading LLMs that are likely to deliver the best responses based on the nature of the user's query, or 'prompt'.

More significantly, Pulze enables its partners to curate their own chat instances in the form of 'spaces'. With spaces one has the power to directly control and easily modify over time the types of data and content that a chat instance both trains with and draws responses from. For example, if a subscriber wants to create an AI chat tool that only pertains to subjects related to, say, beekeeping, then they can train that chat using only documents (or websites) of their own choosing, related to that subject.

Finally, Pulze - unlike many leading LLM platforms - makes the sources, and specific data extracted from sources, visible in the prompt response itself. This feature comes with a number of advantages, but most importantly, for CDRai's goals, it creates a high level of transparency for the (sometimes AI skeptical) user. This is a critical feature for a tool like CDRai, which aims to shine light on a very novel, complex and important subject that still remains not only obscure, but also in some corners, controversial and highly subject to misunderstanding and misinformation. Pulze solves this by turning the blackbox of AI into an open window, thus preempting potential concerns regarding source credibility. In short, if a user wants to track down and validate the source data behind a response, they can easily do that.

.... it's the quality of sources, even more than the number, that matters most to CDRai's ultimate objective: expanding access to knowledge through trust.

Our data: credible, current, unbiased, transparent

At launch, CDRai's data pool consists of over 620 individual sources, with new documents being reviewed and added almost everyday. And we have and will always make these sources completely public. We aim to reach at least 1,000 individual sources by the end of August, and keep growing from there - powered by the time and effort put in by a growing number of OpenAir volunteers.

But it's the quality of sources, even more than the number, that matters most to CDRai's ultimate objective: expanding access to knowledge through trust.

This has been our core priority from the earliest stages of concept development. And with the Pulze functions described above we are able to pull it off. It starts with establishing, following and making transparent non-arbitrary criteria for source selection. While our criteria may change over time, based on feedback from users and our own learning, we think we've started with an approach that puts us on very solid ground right out of the gate.

• Selection based on search terms, not point of view.

OpenAir is a community that exists to support the rapid expansion of carbon dioxide removal as a necessary component of addressing the climate emergency. Full stop, no apologies. But, in the interest of gaining trust, we can't fall prey to the temptation to cherry-pick sources that simply reflect our own, admittedly pro-CDR biases. Indeed a meta objective of CDRai is to provide users with a perspective on CDR that reflects the net direction of data and perspective from diverse sources, regardless of what they reveal or the conclusions they arrive at. If the net direction from all sources points towards conclusions that generally reinforce the necessity and ultimate viability of CDR (as we believe it clearly does) then that creates a level of meta validation that simply drawing from data that could be characterized as 'pro-CDR' would not.

To achieve this we narrow down our potential sources using a number of search terms that pertain to different subjects related to CDR. These include very general terms (ex. 'carbon dioxide removal' 'carbon removal' 'greenhouse gas removal', 'negative emissions', etc.); the common names of different CDR pathways (ex. 'enhanced rock weathering' 'biochar', 'direct air capture', 'ocean alkalinity enhancement', etc.); as well as cross-cutting concepts relevant in areas such as commerce, economics, policy, and social acceptance (ex. carbon dioxide removal + 'monitoring, reporting and verification', 'carbon markets', 'regulation', 'public attitudes', 'life cycle assessment', 'finance', etc.). By using search terms, with the help of Google Scholar notifications and searches, we cast a wide, values-blind net to identify sources.

• Credible source categories

But search terms aren't enough. We also narrow the field around general source categories that we believe promote quality and credibility. These include:

• Peer-reviewed articles. We include only journal articles that have been subjected to peer-review prior to publication. Selections are limited to journals that have attained a minimum SCImago Journal Rank (SJR) for the year that the article was published. SJR is a robust and widely accepted standard for evaluating academic journals that crunches the number of citations, impact factor and other prestige markers. Only SJR Q1 (top 20%) journals are eligible for inclusion. Roughly 90% of all CDRai sources are peer-reviewed journal articles.
• Publications by accredited academic institutions, university presses, and other independent research bodies. Many quality resources are published directly by academic institutions, and other research bodies in the form of reports and books. CDRai includes such resources, but at present, restricts selections to those published by institutions that have attained a minimum score of 75 for academic reputation in the Quacquarelli Symonds (QS) World University Rankings. QS World University Rankings is widely accepted globally as a standard for assessing relative quality and reputation of universities. In addition to university resources, CDRai also includes relevant materials published by different national academies of sciences and engineering.?
• Government produced, sanctioned, or funded resources. Federal and state governments support significant high-quality research through a variety of public or quasi-public institutions, including national laboratories and research centers, and relevant government agencies and departments. For example, in the context of the United States, examples of acceptable source institutions include the National Oceanic and Atmospheric Administration (NOAA), the U.S. Geological Survey, the National Science Foundation, and the U.S. Department of Energy’s National Laboratories.

In the near future we intend to add new categories, including:

• Verified information about real world CDR projects (type, location) and companies.
• CDR related laws, regulations and legislation.
• Published methodologies by leading third-party bodies related to CDR certification, monitoring, reporting and verification.

So stay tuned - more to come soon!

• Current: Only recent publications from 2022 and after.

CDR represents a rapidly evolving and increasingly complex field, and the rate of data and publications related to it is growing at an exponential rate. For this reason, we have limited inclusion to those sources that were published no later than 2022, and this cutoff will be adjusted over time. A lot of rigorous, important knowledge was of course published prior to 2022, and they indirectly influence our data pool because so many of them are highly cited in the documents that we do include.

Use Cases for all levels of learning and exploration

CDRai is meant to be useful for all varieties of people, and to meet different user requirements across a range of levels of expertise and prior engagement in the subject. We have been particularly attuned to (READ: obsessed with!) the needs of folks who are brand new to this vast subject. If it doesn't work for this largest and most important of segments, then to our mind it just isn't really doing its job. As with all AI platforms, ultimately users will figure out CDRai's value better than we can. However, we are launching the platform with a few general use cases that we know it can powerfully support:

• CDR 101. Our platform is great for people who are just trying to wrap their heads around the fundamentals: terms and concepts, major scientific findings and positions, and clarifying what CDR is and isn't, and what it's for and not for? If you fall into this category, we think a good place to start your searches is with simple prompts like "what is carbon dioxide removal?" or "Is CDR necessary?" - and then proceed down the rabbit hole from there.
• Deeper Dives. CDRai is also handy for users who want to take a much deeper dive into specific CDR pathways and related concepts. Let's say you want to understand emerging trends in direct air capture chemistry; or the lifecycle analysis of BECCS; or the different things biochar can be made out of and what it can be used for; . . . or information about public attitudes concerning any or all of these and more technologies. You can use prompts that correspond to these types of inquiries, and even at this early beta phase with just 600+ sources, CDRai will give you pretty solid answers.
• Fact checking/ correcting the record. There is a lot of questionable chatter and opinion about CDR online and in the media. Some of it is partially correct, while some is just outright false, rooted in misunderstanding, oversimplification, or even deliberate attempts to distort. This needs to change if CDR is to advance, and our platform provides an excellent tool for rapidly vetting and responding to incorrect or misleading statements about CDR on social media, in media outlets of record, and other sources. And correcting the record by sharing back to the world accurate data from CDRai is extra powerful because you can say "I checked your statement against hundreds of peer-reviewed journal articles using AI, and this is what it had to say....".

Setting expectations, and tips for effective use

We urge our early adopters to be patient with the course and pace of this new venture, and seek your help in our efforts to constantly make it better. When you first visit CDRai to give it whirl, please try and bear the following in mind:

AI is a work in progress. That means CDR.ai is too.

CDRai is new. But, hey, AI in general is also still kind of new, too. As tools like ChatGPT become important, even essential parts of our lives and work, most of us have come to experience firsthand some of the technology's current limitations and quirks. LLMs are not perfect. They make mistakes, sometimes even crazy ones, aptly referred to as 'hallucinations'. As a result we need to manage our expectations, be aware of current limitations, and adopt strategies that can help us get the most from these new world-changing but still evolving tools. Among the latter we recommend:

• Using simple, clear prompts. Communicating what you are after in clear, straight forward ways tends to lead to the best and most accurate results.
• Varying prompt language to ask the same thing a few times. Getting the best learning experience from an LLM sometimes means phrasing your questions in a few different ways to see what comes up. Sometimes it's a matter of shifting some words around to get the most lucid response.

CDRai doesn't know everything - it's only as smart as its current sources.

There is a tradeoff we have to accept in controlling and selecting our own data sources. This path weeds out questionable information, and promotes quality and trust. But it also means a lot of work has to go into finding, reviewing and incorporating new sources. We are today at just over 600 sources, but we aim to reach thousands in the not too distant future. And as our data grows, we want to ensure that it covers as full of a spectrum of related subjects as possible. But, inevitably, it won't have answers for every subject that is of interest to every user.

Fortunately, CDRai will let you know when it doesn't have enough data to offer a minimal response. Our volunteer base will grow over time, and that will mean more sources reviewed and added at a faster rate. And we also hope to use the crowd to ramp up our sources. On the CDRai webpage, the feedback form also includes an option to upload articles for our review and consideration. Please use it, and short of that let us know what areas of knowledge that you are interested in, but are not getting sufficient responses for. Our team can turn our attention to those first, and start chipping away at it.

Also, while we believe our quality guidelines are very sound, we also don't claim that responses represent unassailable truth. The source categories we have selected increase the chance of accuracy and depth. However, academics get things wrong, and most subjects, general and specific, come with a range of opinions, findings and even heated debates. So, don't surrender your own judgement, or sell your own knowledge short when assessing the platform's responses.

This is the just the beginning, so we appreciate your patience!

Also, as we hope we've made clear, CDRai is right now at its starting line, not its final destination. By implementing the guidelines described above and reaching a critical mass of sources, we believe we've reached the right time to release it into the world for general public use, and productive scrutiny. There will be bugs, most small we hope, but some inevitably big. We will continue to fix, modify and improve as we go - which tends to be the way all great and ultimately useful things are made. In the interest of accelerating and improving our error correction, we have included a feedback form on the very bottom of the CDRai webpage. User ideas, problem identification, and suggestions will be essential to making this platform the best it can be. So don't be shy! Let us know what you think.

So, that's all for now! Thank you for joining us on this new journey, and we appreciate your interest and patience in advance. We hope CDRai brings our users great value, and contributes to building awareness and knowledge in the world about this critical, necessary, fascinating and ultimately inspiring part of the climate challenge.

Happy prompting everyone!