• By Dave Emerson
• Cover Story

Summary

A detailed look at O-PAS™ Standard, Version 1.0

By Dave Emerson

Process automation end users and suppliers have expressed interest in a standard that will make the industry much more open and modular. In response, the Open Process Automation™ Forum (OPAF) has worked diligently at this task since November 2016 to develop process automation standards. The scope of the initiative is wide-reaching, as it aims to address the issues associated with the process automation systems found in most industrial automation plants and facilities today (figure 1).

It is easy to see why a variety of end users and suppliers are involved in the project, because the following systems are affected:

• manufacturing execution system (MES)
• distributed control system (DCS)
• human-machine interface (HMI)
• programmable logic controller (PLC)
• input/output (I/O)

In June 2018, OPAF released a technical reference model (TRM) snapshot as industry guidance of the technical direction being taken for the development of this new standard. The organization followed the TRM snapshot with the release of the OPAS™ Version 1.0 in January 2019. Version 1.0 addresses the interoperability of components in federated process automation systems. This is a first stop along a three-year road map with annual releases targeting the themes listed in table 1.

Table 1. The O-PAS Standard three-year release road map addresses progressively more detailed themes.

By publishing versions of the standard annually, OPAF intends to make its work available to industry expeditiously. This will allow suppliers to start building products and returning feedback on technical issues, and this feedback-along with end user input-will steer OPAS development. O-PAS Version 1.0 was released as a preliminary standard of The Open Group to allow time for industry feedback.

The OPAF interoperability workshop in May 2019 is expected to produce feedback to help finalize the standard. The workshop allows member organizations to bring hardware and software that support O-PAS Version 1.0, testing it to verify the correctness and clarity of this preliminary standard. The results will not be published but will be used to update and finalize the standard.

Figure 1. A broad sampling of suppliers and end users are highly interested in the scope of the OPAS under development by OPAF, because it touches on all the key components of industrial automation systems: hardware (I/O), the communication network, system software (e.g., run time, namespace), application software, and the data model. 

Some terminology

For clarity, a summary of the terminology associated with the OPAF initiative is:

• The Open Group: The Open Group is a global consortium that helps organizations achieve business objectives through technology standards. The membership of more than 625 organizations includes customers, systems and solutions suppliers, tool vendors, integrators, academics, and consultants across multiple industries.
• Open Process Automation Forum: OPAF is an international forum of end users, system integrators, suppliers, academia, and other standards organizations working together to develop a standards-based, open, secure, and interoperable process control architecture. Open Process Automation is a trademark of The Open Group.
• O-PAS Standard, Version 1.0 (O-PAS): OPAF is producing the OPAS Standard under the guidance of The Open Group to define a vendor-neutral reference architecture for construction of scalable, reliable, interoperable, and secure process automation systems.

Standard of standards

Creating a "standard of standards" for open, interoperable, and secure automation is a complex undertaking. OPAF intends to speed up the process by leveraging the valuable work of various groups in a confederated manner.

The OPAS Standard will reference existing and applicable standards where possible. Where gaps are identified, OPAF will work with associated organizations to update the underlying standard or add OPAS requirements to achieve proper definition. Therefore, OPAF has already established liaison agreements with the following organizations:

• Control System Integrators Association (CSIA)
• Distributed Management Task Force (DMTF), specifically for the Redfish API
• FieldComm Group
• Industrial Internet Consortium (IIC)
• International Society of Automation (ISA)
• NAMUR
• OPC Foundation
• PLCopen
• ZVEI

Additionally, OPAF is in discussions with AutomationML and the ISA Security Compliance Institute (ISCI) as an ISA/IEC 62443 validation authority. In addition to these groups, the OPC Foundation has joined OPAF as a member, so no liaison agreement is required.

As an example of this cooperation in practice, OPAS Version 1.0 was created with significant input from three existing standards, including:

• ISA/IEC 62443 (adopted by IEC as IEC 62443) for security
• OPC UA adopted by IEC as IEC 62541 for connectivity
• DMTF Redfish for systems management (see www.dmtf.org/standards/redfish)

Next step: Configuration portability

Configuration portability, now under development for OPAS Version 2.0, will address the requirement to move control strategies among different automation components and systems. This has been identified by end users as a requirement to allow their intellectual property (IP), in the form of control strategies, to be portable. Existing standards under evaluation for use in Version 2.0 include:

• IEC 61131-3 for control functions
• IEC 16499 for execution coordination
• IEC 61804 for function blocks

O-PAS Version 3.0 will address application portability, which is the ability to take applications purchased from software suppliers and move them among systems within a company in accordance with applicable licenses. This release will also include the first specifications for hardware interfaces.

Under the OPAS hood

The five parts that make up O-PAS Version 1.0 are listed below with a brief summary of how compliance will be Tested (if applicable):

• Part 1 — Technical Architecture Overview (informative)
• Part 2 — Security (informative)
• Part 3 — Profiles
• Part 4 — Connectivity Framework (OCF)
• Part 5 — System Management

Part 1 - Technical Architecture Overview (informative) describes an OPAS-conformant system through a set of interfaces to the components. Read this section to understand the technical approach OPAF is following to create the O-PAS.

Part 2 - Security (informative) addresses the necessary cybersecurity functionality of components that are conformant to OPAS. It is important to point out that security is built into the standard and permeates it, as opposed to being bolted on as an afterthought. This part of the standard is an explanation of the security principles and guidelines that are built into the interfaces. More specific security requirements are detailed in normative parts of the standards. The detailed normative interface specifications are defined in Parts 3, 4, and 5. These parts also contain the associated conformance criteria.

Part 3 - Profiles  defines sets of hardware and software interfaces for which OPAF will develop conformance tests to make sure products interoperate properly. The O-PAS Version 1 profiles are:

• Level 1 Interoperability Hardware Profile: A certified product claiming conformance to this profile shall implement OSM-Redfish.
• Level 2 Interoperability Hardware Profile: A certified product claiming conformance to this profile shall implement OSM-Redfish BMC.
• Level 1 Interoperability Software Profile: Software claiming conformance to this profile shall implement OCF-001: OPC UA Client/Server Profile.
• Level 2 Interoperability Software Profile: Software claiming conformance to this profile shall implement OCF-002: OPC UA Client/Server and Pub/Sub Profile.

The term "Level" in the profile names refers to profile levels.

Part 4 - Connectivity Framework (OCF) forms the interoperable core of the system. The OCF is more than just a network, it is the underlying structure allowing disparate components to interoperate as a system. The OCF will use OPC UA for OPAS Versions 1.0, 2.0, and 3.0.

Part 5 - System Management covers foundational functionality and interface standards to allow the management and monitoring of components using a common interface. This part will address hardware, operating systems and platform software, applications, and networks-although at this point Version 1.0 only addresses hardware management.

Conformance criteria are identified by the verb "shall" within the O-PAS text. An OPAF committee is working on a conformance guide document that will be published later this year, which explains the conformance program and requirements for suppliers to obtain a certification of conformance.

Technical architecture

The OPAS Standard supports communication interactions that are required within a service-oriented architecture (SOA) for automation systems by outlining the specific interfaces the hardware and software components will use. These components will be used to architect, build, and start up automation systems for end users.

The vision for the OPAS Standard is to allow the interfaces to be used in an unlimited number of architectures, thereby enabling each process automation system to be "fit for purpose" to meet specific objectives. The standard will not define a system architecture, but it will use examples to illustrate how the component-level interfaces are intended to be used. System architectures (figure 2) contain the following elements:

Distributed control node (DCN): A DCN is expected to be a microprocessor-based controller, I/O, or gateway device that can handle inputs and outputs and computing functions. A key feature of O-PAS is that hardware and control software are decoupled. So, the exact function of any single DCN is up to the system architect. A DCN consists of hardware and some system software that enables the DCN to communicate on the O-PAS network, called the OCF, and also allows it to run control software.

Distributed control platform (DCP): A DCP is the hardware and standard software interfaces required in all DCNs. The standard software interfaces are a common platform on top of which control software programs run. This provides the physical infrastructure and interchangeability capability so end users can control software and hardware from multiple suppliers.

Distributed control framework (DCF): A DCF is the standard set of software interfaces that provides an environment for executing applications, such as control software. The DCF is a layer on top of the DCP that provides applications with a consistent set of O-PAS related functions no matter which DCN they run in. This is important for creating an efficient marketplace for O-PAS applications.

OPAS connectivity framework (OCF): The OCF is a royalty-free, secure, and interoperable communication framework specification. In O-PAS Version 1, the OCF uses OPC UA.

Advanced computing platform (ACP): An ACP is a computing platform that implements DCN functionality but has scalable computing resources (memory, disk, CPU cores) to handle applications or services that require more resources than are typically available on a small profile DCP. ACPs may also be used for applications that cannot be easily or efficiently distributed. ACPs are envisioned to be installed within on-premise servers or clouds.

Within the OPAS Standard, DCNs represent a fundamental computing building block (figure 3). They may be hardware or virtual (when virtual they are shown as a DCF as in figure 2), big or small, with no I/O or various amounts. At the moment, allowable I/O density per DCN is not settled, so some standardization in conjunction with the market may drive the final configuration.

DCNs also act as a gateway to other networks or systems, such as legacy systems, wireless gateways, digital field networks, I/O, and controllers like DCS or PLC systems. Industrial Internet of Things (IIoT) devices can also be accessed via any of these systems.

Figure 2. OPAS establishes a system architecture organizing process automation elements into interoperable groupings.

Building a system

End users today must work with and integrate multiple systems in most every process plant or facility. Therefore, the OPAS Standard was designed so users can construct systems from components and subsystems supplied by multiple vendors, without requiring custom integration. With the OPAS Standard it becomes feasible to assimilate multiple systems, enabling them to work together as one OPAS-compliant whole. This reduces work on capital projects and during the lifetime of the facility or plant, leading to a lower total cost of ownership.

By decoupling hardware and software and employing an SOA, the necessary software functions can be situated in many different locations or processors. Not only can software applications run in all hardware, but they can also access any I/O to increase flexibility when designing a system.

One set of components can be used to create many different systems using centralized architectures, distributed architectures, or a hybrid of the two. System sizes may range from small to large and can include best-in-class elements of DCS, PLC, SCADA, and IIoT systems and devices as needed.

Information technology (IT) can also be incorporated deeper into industrial automation operational technology (OT). For example, DMTF Redfish is an IT technology for securely managing data center platforms. OPAF is adopting this technology to meet OPAS system management requirements.

Comprehensive and open

Each industrial automation supplier offers a variety of devices and systems, most of which are proprietary and incompatible with similar products from other vendors and sometimes with earlier versions of their own products. End users and system integrators trying to integrate automation systems of varying vintages from different suppliers therefore have a challenging job.

To address these issues, OPAF is making great strides toward assembling a comprehensive, open process automation standard. Partially built on other established industry standards, and extending to incorporate most aspects of industrial automation, the O-PAS Standard will greatly Strengthen interoperability among industrial automation systems and components. This will lower implementation and support costs for end users, while allowing vendors to innovate around an open standard.

For more information on OPAS Version 1.0, please download the standard at https://publications.opengroup.org/p190. Submit feedback by emailing ogspecs@opengroup.org. 

Figure 3. DCNs are conceived as modular elements containing DCP (hardware) and DCF (software), both of which are used to interface field devices to the OCF.

Monthly Program Design is exclusively for repurchasing clients. To qualify for Monthly Program Design, clients must work with one of RecWell's Nationally Accredited Personal Trainers for at least 5 individual sessions.

This two-session package offers an exercise program based on a client's personal goals, workout preferences and fitness level. Each month, includes an individualized four-week fitness program based on the included Fitness Assessment as well as two optional consultations with a Wellness Coach (for students) at no additional cost. The qualified team of trainers and coaches support clients in their pursuit of an active, healthy lifestyle.

Within the eVTOL landscape, regulatory compliance and airworthiness certification present unique challenges. Bhavesh Mandalia, Chief Airworthiness Officer and Deputy CTO at Lilium, in a accurate interview with Avionics International, discussed how the company navigates this complex environment, detailing its innovative certification approach, dedication to safety, distinctive technology, and the journey towards obtaining Design Organization Approval (DOA). As the industry continues to take shape, Lilium's pioneering efforts offer a glimpse into the future of air mobility.

Lilium's Bhavesh Mandalia talks about the team’s strategies for achieving airworthiness certification and compliance with evolving aviation regulations. He also highlights their unique technology, a strong commitment to safety, and future goals in the dynamic eVTOL industry. (Photos: Lilium)

Avionics: Could you provide an overview of Lilium's approach to achieving airworthiness certification and ensuring compliance with aviation regulations? Are there any unique challenges Lilium has faced in this process?

Bhavesh Mandalia: With certification, we've been engaged with two different authorities—EASA, the authority here in Europe, and the Federal Aviation Administration in the U.S. This was quite important for us, because both authorities have been developing a relatively new regulatory landscape for electric VTOL aircraft. It's important for us to establish a good relationship as early as possible with them, which takes some years for a new applicant. Lilium applied for EASA type certification very early, back in 2017, and applied for type certification validation with the FAA shortly after.

EASA has broken down the certification process into 18 different disciplines, which is good for us. All disciplines come together to form the aircraft but enable us to group certification activity with focus on a particular area. Therefore, we can work on parallel certification streams with different EASA experts. 

We've constructed our teams to work with these individual areas of discipline for certification, which has been important for us. It's helped us progress well with the evolving requirements. When we compare what we're doing now with conventional aviation, developing an aircraft when you still have some fluidity with the regulations is quite a unique challenge to have. 

Equally, establishing an early relationship with the regulator has helped us to learn with them and to use our product and some of our key learnings to actually influence the regulations, which is always positive. It's difficult, I think, for regulators to put new regulations together without actually seeing what they look like when they're applied. So we've used a lot of our research and development activities to help them evolve as well. 

Avionics: Lilium recently completed the fourth and final Design Organization Approval audit by EASA. Could you share some details about this achievement and the key factors that contributed to the team’s success?

Mandalia: The DOA process in the European regulation framework is a prerequisite to develop and certify a new aircraft, and also for organizations that want to keep aircraft airworthy by implementing changes and repairs once they're in service. EASA has recently made a number of updates to the regulation governing initial airworthiness known as EASA Part 21, to introduce a Safety Management System and a risk-based system to help determine their level of involvement within certification activities. 

We've been working with EASA since 2017 on our DOA approval. We've had four audits. The fourth and final one was this year, and each audit has progressively focused on a different area of our organization, such as the company structure, how we approve our people, and how we approve our suppliers. As we've evolved as an organization and our product has evolved, the types of audit are changing. The second audit was more about certification activity and how we manage configurations of the aircraft. The third one was more on how we start the genuine compliance demonstration. The final one was about some of the deliverables that we would provide as part of the aircraft, including manuals on how to train people to operate and maintain the aircraft, and also to keep the aircraft airworthy.

We passed the audit, and as the regulations have evolved, we've actually tailored our DOA to be more specific for developing this type of electrically propelled, vertical take-off and landing aircraft. That's been quite important for us instead of just putting together something that was more generic. 

I think successfully completing this last audit is a really good testament that we have competent people within the organization but also a structure that's credible enough to certify such an aircraft. This is an important milestone for us because ultimately having a DOA approval will demonstrate to people that we're a credible aerospace company; that is quite key in everything that we're doing. 

For me, the DOA process is not new. This is actually the fourth approval I've been involved in with my career starting from scratch, and every single one is different, even though the regulations you have to comply with are the same. The approach is based on the type of product that you're working with. Therefore, there's a lot more emphasis on electrification and the use of new technology within our DOA, which is somewhat different to conventional aviation. The next steps now are to work to close off any open actions that we have and then get our DOA later this year.

Avionics: As Chief Airworthiness Officer, you play a crucial role in ensuring the safety and reliability of Lilium's eVTOL aircraft. How does the company address safety considerations?

Mandalia: Like any aerospace company developing an aircraft that people will be using as a mode of transport, safety is a crucial component in order for our aircraft to meet the highest safety objectives. Within Europe, EASA has decided to prescribe the same level of safety as they do for any commercial aircraft. Essentially, the probability of what we call a catastrophic failure is one in a billion flight hours, or 1x10^-9. 

The other important element for us as an organization is the implementation of the safety culture. EASA have recently introduced what they're calling a safety management system. Most of the leadership team we have, including myself, are from large aerospace organizations where we're already used to working with such a safety culture, and we already have systems implemented. For most of us, this was just a way of formalizing how we work with safety, but we consider multiple different areas within safety in our organization, and we also promote this and train our people to follow the same mantra. That includes design—the safety of the design, meeting the standards for airworthiness—and safety of the product, so that within the production organization system, it also has its own safety management element. Safety in operations—as we design, develop, and manufacture the product, we also have to ensure it's safe to operate the product by aircrew and for people traveling on it. Safety for our employees—we operate in a safe environment, we make sure the office space is safe for them and we also make sure that the environment that they operate in is safe. 

We employ something called a "just culture." We encourage people to report anything that they feel has gone wrong or could potentially go wrong and could impact either other people or the safety of the product that we're developing. 

We also have safety of our customers; that's the end goal, and we have organizational safety as well. These are important aspects of what we do. We make it very easy for our people to actually report items of safety as well and ensure that they have a good area where they can learn about why safety is so important, because we have a number of people in our organization that are from outside of aerospace. 

We've created what we call the Lilium Safety Hub, which is a repository on our intranet system, where we have a number of resources, including our safety policy, and also self-learning and training that people can take in addition to those that we mandate for safety—all of those elements of the Safety Management System which we have introduced under our DOA. For my position, because I'm something they call a regulatory postholder, if something goes wrong, I'm accountable for the certification and safety of the product. That's quite key, and it's taken very seriously, even as a new company.

Avionics: What differentiates Lilium's technology from others in the eVTOL industry, and how does the company maintain a competitive edge in this dynamic landscape?

Mandalia: I'll mention something that I think is quite unique, which you don't necessarily hear other people talking about: customer comfort and customer experience. More recently, we've been advertising the quality of our cabin, and in addition to meeting the important safety requirements for the cabin, we also need to ensure that the customer experience is positive. This is why we have essentially the largest cabin on the aircraft with the ability to configure the cabin in multiple ways to suit the needs of the customer—the four-seater plush VIP-style club configuration, or the six-seater shuttle configuration. We're also offering different color schemes and various other bits and pieces, and it's quite a comfortable environment. I think that's one of the key differentiators that's quite important. 

We have multiple Electric Ducted Jet Engines, which is different to what the competitors have; most of them have open rotors. The use of multiple Electric Ducted Jet Engines introduces a number of benefits for our aircraft. Firstly, they are quieter when compared to open rotors, which will enable us to operate our aircraft around the clock and in areas where they have noise restrictions.

Secondly, having multiple ducted fans provides redundancy to meet the highest safety standards but also adds resilience against things like bird strikes which tend to be more common when flying at lower altitudes like eVTOL aircraft will. 

Thirdly, this architecture provides great cruise efficiency which is why we are targeting regional air mobility for our product.

We have 30 engines on the aircraft in total. It means we have a lot of redundancy. So if we were to have any failures on the aircraft, we have multiple engines that keep the aircraft airborne and continue to its original destination or to an alternative. The aircraft can still continue to fly with certain engines damaged. That was quite important for us as well—having this kind of architecture. 

We're actually developing our own battery cells instead of opting to procure something that's already out there and off the shelf. Some of this is because we want to develop something that is unique and something that is specific to our needs and the operational needs of our aircraft. Secondly, the regulators—particularly EASA—has set the bar very high with regard to safety standards for batteries and for technology. We don't feel, from what we've seen out there, that automotive batteries or lithium-ion batteries used for other technologies are meeting the standards that are necessary for airworthiness. Compare it to automotive: with a car, if you have an incident with the battery system, the car comes rolling to a stop. With an aircraft, if you have a major failure of the battery cells or the electrical system, you've got more of an issue at hand. As a result of that, EASA has set the bar quite high for cells. So we have to develop our own technology to meet the requirements.

Avionics: Looking ahead, what are Lilium's main priorities and focus areas for the upcoming year?

Mandalia: For my team in particular, one of the most important things is closing out the actions that remain for the DOA and receiving our DOA approval, which we're expecting later on this year. That one is quite big on the radar. 

The next thing for us is actually a certification program. The 18 different areas that I mentioned within EASA, we have certification plans for each of these areas which we've already shared with EASA. It provides EASA with an indicative level of involvement as well within these activities. We need to basically formalize the agreement with EASA and then move on to the next phase, which is the genuine demonstration of compliance and certification. 

Also within the next year, we will be going into production of our first prototype aircraft, which will be used for performing flight testing activities and ground testing, as well as other verification activities. In parallel with this, we'll also be continuing other analysis and demonstrations of compliance for the authority in alignment with the certification plans that we've already submitted to them. Those are some of our objectives as an organization. More importantly, aside from that is ensuring we have the right people. We've already demonstrated to EASA that we have highly competent people, but making sure we maintain that engagement of our workforce, and also continue to find the best talent out there—all of this is important for us to continue the momentum.

Avionics: Could you share some details about Lilium’s accurate wind tunnel testing?

Mandalia: We performed one of the most complex wind tunnel testing campaigns recently where we had an active scale model of our aircraft in one of the wind tunnels here in the region. This is one of the models that we're using to provide us with data to validate that engineering is on the right track. This is something that we do to de-risk activities before we go into production and manufacture the full-scale aircraft. We used an active model with the engines running because having so many engines on the aircraft does have an impact on the airflow around the aircraft. So it's important that it's a representative model of what we're doing. We'll be using data from that to validate some of our other simulations that were put in place like, for example, for fluid dynamic models. For other simulations, we're using aerodynamics. We're working to demonstrate that the control laws and things that we've put together for the aircraft work before we actually put them into place on a real-life aircraft.

Q2 Shareholder Letter Updates

Lilium published its Q2 2023 shareholder letter this week. The company reports that it has hit key development and certification milestones to keep on track for the first manned flight of its type-conforming aircraft in late 2024. Lilium secured around $192 million in fresh financing, including a successful capital raise of $117 million mostly from new investors, leading to a $75 million pre-funding commitment from Aceville. The new financial resources, combined with existing funds, result in approximately $386 million of liquidity as Lilium steps into the second half of 2023. The funds will be instrumental for the development of the Lilium Jet. Future financing will prioritize non-dilutive funding. 

The accurate capital raise is a significant endorsement of Lilium's technology, attracting more investors and accelerating commercial engagement. The firm is set to begin assembling the first Lilium Jet for systems integration validation. Meanwhile, progress on the aircraft design and testing and marketing initiatives continue. The Lilium Jet's cabin was a major draw at the 2023 Paris Air Show. Lilium also received a G-1 certification from the FAA, indicating the regulatory acceptance of its jet, and reports that interest from global markets, including China, is on the rise.

FITCHBURG — The local Leaders for Equitable Local Economies (LELE) team joined with city officials and state legislators, among others, to celebrate the positive impact of their childcare entrepreneur training program during an event last Friday.

The program is designed to assist aspiring family childcare entrepreneurs to go through and navigate the complex licensing process to expand access to childcare in Fitchburg and the surrounding areas.

The activities during celebration July 21 included remarks from Fitchburg LELE program director Marites MacLean, Mayor Stephen DiNatale and Ines Palmarin, the co-leader of the LELE at the Federal Reserve Bank of Boston.

The LELE is a pilot program at the Federal Reserve Bank of Boston that aims to support leaders who are working to make their local economies more inclusive. The goal of the Fitchburg LELE team is to create a more equitable childcare system.

The program has graduated 80 participants and will be starting its fifth cohort in August 2023. Event attendees include elected officials, philanthropic leaders and leaders from various organizations who support the early education and care industry.

The Fitchburg LELE team designed and implemented its bilingual (Spanish/English) childcare entrepreneur training program to help women from immigrant communities and communities of color open up their own small businesses. The program also provided financial support to graduates who have successfully obtained their family childcare licenses through its partnership with Empower Children for Success, a nonprofit organization.

The training program was made possible by the grant funds from the Federal Reserve Bank of Boston through the team’s fiscal sponsor, Seven Hills.

Fitchburg’s Economic Development Department also set up its own FCC Start Up Grant Funds and graduates who obtained their family childcare licenses from the state Department of Early Education and Care are encouraged to apply for the grant.

The celebration also acknowledged the team’s partnership and collaborative work with the Federal Reserve Bank of Boston, NewVue Communities, Fitchburg State University, Fitchburg Public School, Seven Hills, MOC, the City of Fitchburg’s Economic Development Department, and Mount Wachusett Community College, its coaches, and other collaborators.

This release was provided by the Fitchburg Leaders for Equitable Local Economies (LELE) Team.

Language models like GPT-4 and Claude are powerful and useful, but the data on which they are trained is a closely guarded secret. The Allen Institute for AI (AI2) aims to reverse this trend with a new, huge text dataset that’s free to use and open to inspection.

Dolma, as the dataset is called, is intended to be the basis for the research group’s planned open language model, or OLMo (Dolma is short for “Data to feed OLMo’s Appetite). As the model is intended to be free to use and modify by the AI research community, so too (argue AI2 researchers) should be the dataset they use to create it.

This is the first “data artifact” AI2 is making available pertaining to OLMo, and in a blog post, the organization’s Luca Soldaini explains the choice of sources and rationale behind various processes the team used to render it palatable for AI consumption. (“A more comprehensive paper is in the works,” they note at the outset.)

Although companies like OpenAI and Meta publish some of the vital statistics of the datasets they use to build their language models, a lot of that information is treated as proprietary. Apart from the known consequence of discouraging scrutiny and improvement at large, there is speculation that perhaps this closed approach is due to the data not being ethically or legally obtained: for instance, that pirated copies of many authors’ books are ingested.

You can see in this chart created by AI2 that the largest and most accurate models only provide some of the information that a researcher would likely want to know about a given dataset. What information was removed, and why? What was considered high versus low-quality text? Were personal details appropriately excised?

Chart showing different datasets’ openness or lack thereof. Image Credits: AI2

Of course it is these companies’ prerogative, in the context of a fiercely competitive AI landscape, to guard the secrets of their models’ training processes. But for researchers outside the companies, it makes those datasets and models more opaque and difficult to study or replicate.

AI2’s Dolma is intended to be the opposite of these, with all its sources and processes — say, how and why it was trimmed to original English language texts — publicly documented.

It’s not the first to try the open dataset thing, but it is the largest by far (3 billion tokens, an AI-native measure of content volume) and, they claim, the most straightforward in terms of use and permissions. It uses the “ImpACT license for medium-risk artifacts,” which you can see the details about here. But essentially it requires prospective users of Dolma to:

• Provide contact information and intended use cases
• Disclose any Dolma-derivative creations
• Distribute those derivatives under the same license
• Agree not to apply Dolma to various prohibited areas, such as surveillance or disinformation

For those who worry that despite AI2’s best efforts, some personal data of theirs may have made it into the database, there’s a removal request form available here. It’s for specific cases, not just a general “don’t use me” thing.

If that all sounds good to you, access to Dolma is available via Hugging Face.

One of the most prominent pirated book repositories used for training AI, Books3, has been kicked out from the online nest it had been roosting in for nearly three years. Rights-holders have been at war with online pirates for decades, but artificial intelligence is like oil seeping into copyright law’s water. The two simply do not mix, and the fumes rising from the surface just need a spark to set the entire concept of intellectual property rights alight.

As first reported by TorrentFreak, the large pirate repository The Eye took down the Books3 dataset after the Danish anti-piracy group Rights Alliance sent the site a DMCA takedown. Now trying to access that dataset gives a 404 error. The Eye still hosts other training data for AI, but the portion allotted for books has vanished.

Rights Alliance told Gizmodo it sent The Eye a takedown request, and the site took down the content last month. The group said the Books3 dataset contained around 150 titles published by their member companies. Rights Alliance also reached out to AI model hosting site Hugging Face (which hosted a datacard and link to the Books3 download) as well as EleutherAI. Both organizations pointed the anti-piracy group toward The Eye.

The nonprofit research group EleutherAI originally released Books3 as a part of the AI training set The Pile, an 800 GB open source chunk of training data comprising 22 other datasets specifically designed for training language models. Rights Group said the organization “denied responsibility” for Books3. Gizmodo reached out to EleutherAI for comment, but we did not receive a response.

The Eye claims it regularly complies with all valid DMCA requests, though that data set was originally uploaded by AI developer and prominent open source AI proponent Shawn Presser back in 2020. His stated goal at the time was to open up AI development beyond companies like OpenAI, which trained its earlier large language models on the still-unknown “Books1” and “Books2” repositories. The Books3 repository contained 196,640 books all in plain.txt format and was supposed to deliver fledgling AI projects a leg up against the likes of ChatGPT-maker OpenAI.

Over Twitter DM, Presser called the attack on Books3 a travesty for open source AI. While other major companies and VC-funded startups get away with including copyrighted data in their training data, grassroots projects need something to compete—and that’s what Books3 was for.

“The only way to replicate models like ChatGPT is to create datasets like Books3,” Presser said. “And every for-profit company does this secretly, without releasing the datasets to the public... without Books3, we live in a world where nobody except OpenAI and other billion-dollar companies have access to those books—meaning you can’t make your own ChatGPT. No one can. Only billion-dollar companies would have the resources to do that.”

For as long as the media industry groups have fought against piracy, few expected the next front to the neverending copyright war would be AI. In a phone interview with Gizmodo, Rights Alliance CEO Maria Fredenslund said the organization is actively working to take down other copies of Books3. But this is just the start, and anti-piracy groups now have a new target to focus on compared to the usual boogeymen of file-sharing services and pirate libraries.

“We are very worried. It’s a really huge development in technology and how the content is used,” Fredenslund said. “In a way, we see it as the same as 10 years ago when we discussed file sharing, and governments were very afraid of regulating the internet because, in their eyes, everything had to be free. It turned out that copyright also needed to be regulated on the internet as well as in any other aspect.”

It’s not like there are no more copies of Books3 being hosted on the internet. After the books were taken down last week, Presser posted two new Books3 download links on his Twitter profile. Rights Group said it will continue to pursue sites that host the dataset, but as any old salt of an internet pirate would tell you, once a file’s out and available, it never truly goes away.

Comedian Sarah Silverman was just one of several authors who signed on to a class action lawsuit against Meta, claiming the company stole their books in order to train their LlaMA AI. The lawsuit mentions that Meta used the Books3 repository for training its AI, but added that Meta did not mention what works were contained within those gigabytes of data.

In its whitepaper describing the original LlaMA language model, Meta researchers described Books3 as a “publicly available dataset for training large language models.” Meta referenced this dataset coming from The Pile.

Growing AI models requires an enormous amount of information, and for close to a decade the technology’s development has depended on using protected text. Earlier versions of OpenAI’s language model from just two or three years ago were trained on datasets like BookCorpus, which contained thousands of scraped-up scraps of book text from sites like Smashwords. That dataset was only a few gigabytes of data, but researchers found that it included works that were copyrighted, or required payment to access.

OpenAI’s GPT-3 model used the Books2 training set to train its AI. Both Books1 and Books2 make up close to 15% of GPT-3's training data, though there’s little to no precise information on what’s contained in it. Some have speculated the Books2 data was scraped from Libgen, the open source pirate library also called Library Genesis. There’s even less information on what’s contained in GPT-4's 45 terabytes worth of training data.

Big tech companies are increasingly uninterested in sharing this data, knowing the more they do, the more other people can build similar AI models, or tangle them up in lawsuits. Then again, the costs for training these massive models are staggering, especially for larger models.

But while OpenAI has been revealing less of its training data over the years, we know exactly what’s gone into the Books3 repository. The dataset was derived from a copy of the Bibliotik library. Bibliotik is a so-called “shadow library” akin to other, industry-derided sources like Libgen, Z-Library, and Sci-Hub. Presser had to build scripts that managed to turn PDFs and images into usable .txt files, a very labor-intensive task.

“My goal was to make it so that anybody could [create these models.] It felt crucial that you and I could create our own ChatGPT if we wanted to,” Presser said. “Unless authors intend to somehow take ChatGPT offline, or sue them out of existence, then it’s crucial that you and I can make our own ChatGPTs, for the same reason it was crucial that anybody could make their own website back in the ‘90s.”

Fredenslund said their group was looking to “reach out” to Meta about this copyrighted content being used to train its AI. While the tech giant that is Meta is unlikely to retrain its entire AI model to placate copyright holders, there’s little worldwide regulation mandating transparency for AI models. While the European Union is currently working on an AI Act that will force companies to have some model transparency, Fredenslund said AI developers need to be forced to share the specifics of their training data, including what precise works were used to create their AI models.

“We hope this attitude toward using illegal content will change, that they will not do that in the future,” she said. “We want to be able to actually control the copyright in this aspect, then we actually need to know what the models are trained on.”

As noted in past forum comments, Presser actively worked with EleutherAI to add the Books3 dataset to The Pile. EleutherAI has used The Pile and other data to craft its own AI models, including one called GPT-J that was originally meant to compete with OpenAI’s GPT-3.

Meta went as far as to claim that the original LlaMA-65B model didn’t perform as well as some other, larger models like the PaLM-540B because it “used a limited amount of books and academic papers” in its pre-training data. The original LlaMA was also formatted on C4, a version of Common Crawl that was a large dataset of mass amounts of internet data. Researchers found that the C4 training set included mass amounts of published work, including propaganda and far-right websites. Those researchers told the Washington Post the copyright symbol appeared more than 200 million times in the C4 training set.

Since then, Meta has clammed up hard about what goes into its language models. Last month, Meta released a newer, bigger language model called LlaMA 2. This time, Meta worked with Microsoft to add 40% more data than its previous model, though in its whitepaper the company was much more hesitant to outright state what data its latest LM was trained on. The only reference to its training data was that it’s “a new mix of publicly available online data.” As the friction between AI and copyright grows hotter, companies are less and less likely to share exactly what’s contained in the morass of AI training data.

BATAVIA — Batavia Downs was the fourth different track that the recently landed Australian import Mirragon A has competed over since starting his North American career in June, but it didn’t dissuade this Aussie from winning the $15,000 Open I Handicap pace on Saturday night.

Despite being assigned the outside in the field of six, Jim Morrill Jr. wasted no time pushing Mirragon A off the gate and to the lead, a position he controlled at the :27.2 quarter. With the field in single file, Mirragon A took the group to the half in :56.3 and three-quarters in 1:24.2 without any challenges to that point. But trouble was coming and his name was The Longest Yard (Kevin Cummings), who had pulled from fifth at the five-eighths and was making up ground very quickly heading into the final turn. As they entered the stretch, Mirragon A was in control by 1-¼ lengths and The Longest Yard was still over three lengths behind, but he persevered and kept closing the gap as the wire drew near. Not giving into the pressure, Mirragon A maintained his lead while holding off The Longest Yard and a late charging Stranger Things (Shawn McDonough) to win by one-quarter length in 1:53.

It was the fourth win in only seven North American starts for Mirragon A ($13.00) who is owned by his trainer Mike Deters, in partnership with Joel Warner and John Manning.

Morrill ended the night with a driving grand slam after also winning with Rock N Blue (1:55.3, $2.30), Reggiano (1:56.3, $4.20) and Just A Wrangler (1:57, $3.20).

The $13,000 Open II pace provided the biggest upset of the day when the overlooked Sporty Deal took full advantage of a class drop to score his sixth win of the year and reward his backers with a premium payout.

Sporty Deal (Keith Kash Jr.) was almost 10 lengths off at the quarter while Passe-Grille Beach (Ray Fisher Jr.) skated unabated through quarters of :27.1, :55.3 and 1:24.4. It was at that last station where Kash tipped Sporty Deal three-wide from sixth, circled the field and took the lead just as they hit the turn. Kash high-lined Sporty Deal and tried to put some real estate between themselves and the competition and it worked. Sporty Deal hit the stretch and paced away to a convincing 1:54.1 victory, which was a new seasonal mark.

L’Emmur Stable owns Sporty Deal ($43.20) who is trained by his driver, Kash.

Kash had a training/driving double as he also won with Starsnstripes GB (1:55.4, $7.10).

When live racing resumes at Batavia Downs on Wednesday (Aug. 23) at 6 p.m. there will be a $5,601 carryover in the Jackpot Hi-5 pentafecta in the 13th race.

Free full past performance programs for every live card of racing at Batavia can always be downloaded at the Downs’ website (bataviadownsgaming.com) under the “Live Racing” tab and all the racing action can be viewed as it happens for free at the Batavia Downs Live Facebook page.

The Chicago Bears held their third practice of training camp and the second open to fans Friday morning at Halas Hall with the 100-minute session being mostly assignment-driven and slower tempo.

That was intentional as they don’t want to push players too hard this early in camp. The scaled-back tempo helps players mentally grasp the schemes as well.

Advertisement

The Bears will practice Saturday morning before taking Sunday off and returning to work Monday, two days before the first workout in full pads. Here’s our camp rundown.

Jaylon Johnson sat down in front of reporters and fired first.

Advertisement

“Contracts?” the cornerback said. “What are we doing? Let’s get into it. I know we’re going to talk about it.”

The 2020 second-round pick is entering the final year of his rookie contract, and two days after the team completed a $50 million, four-year extension with tight end Cole Kmet, it’s natural to wonder if a pay day could be in the works for Johnson.

The cornerback recently changed agents, hiring Chris Ellison, and it will be interesting to see what position general manager Ryan Poles takes. The Bears have used second-round picks in consecutive drafts on cornerbacks — Kyler Gordon and Tyrique Stevenson — and drafted Terell Smith in the fifth round this year. Poles has spoken fondly of Johnson. The great unknown is whether the sides will be able discover common ground.

[ [Don't miss] Cole Kmet’s contract sends a positive message to the Chicago Bears locker room. Who could be next for an extension? ]

Without a lot of full-speed action, there wasn’t an abundance of plays to sift through and overanalyze. Fields threaded a nice pass to DJ Moore on an intermediate dig route in tight coverage against Johnson in one-on-one drills. His best throw, also in one-on-ones, might have been a corner route to tight end Robert Tonyan against safety Elijah Hicks. The majority of throws in 7-on-7 action were checkdowns, so much so that it could have been by design.

The Bears believe Fields is improved when it comes to rhythm and timing, two elements that begin with footwork. That all gets tested at another level when the team shifts to padded practices next week. You don’t think of practices with pads affecting quarterbacks in the same manner it does linemen, but there’s a change for everyone.

[ [Don't miss] Chicago Bears WR Darnell Mooney works to return to form after 8-month recovery from ankle surgery: ‘It will be better this year’ ]

“There’s that element that the pocket has to move and change a lot more when there’s a real rush,” offensive coordinator Luke Getsy said Thursday. “That’s playing this position at the highest level when you’re able to manipulate the pocket, not necessarily always scramble but manipulate for an angle of the throw or anything like that. So that’s what I’m more talking about.”

The Bears won’t have any truly live pass rushes in pads, but there will be traffic for Fields to work around on occasion, which will test some of the elements he has been working on.

First-round draft pick Darnell Wright revealed the secret to crushing the conditioning test when the right tackle reported last weekend. Wright, whom the team lists at 6-foot-6, 335 pounds, was mistakably training over the summer for the test given to wide receivers — not offensive linemen. The Bears evaluate position groups differently with the benchmarks they are asked to meet to prove their fitness and Wright, well, he was fixing to run with players significantly smaller and a lot faster.

Advertisement

It paid off as GM Ryan Poles noted that Wright crushed the test upon his arrival. Wright said he hired a personal chef with a meal plan in the offseason and dropped 16 pounds, getting down to 328 from where he was in the spring when joining the club for rookie minicamp.

“I thought we had like more,” Wright said of the test for linemen. “I was looking at the wide receivers’ running portion of the workout, so I was doing theirs. Obviously we have different stuff.”

Wright, the No. 10 pick in the draft out of Tennessee, will be one of the players to watch when pads go on. Young linemen are prone to making mistakes, but as long as Wright can learn from his as he grows, he could ascend rapidly.

[ [Don't miss] Chicago Bears president and Mayor Johnson continue talking as NFL team keeps stadium options open ]

After Stevenson spent most of the first two practices running with the starting unit, fellow rookie Smith rotated in with the ones Friday. It wasn’t a reflection on Stevenson’s early work in camp.

“Don’t read anything into who goes in first, second or third,” defensive coordinator Alan Williams said. “It’s a rep plan, not a game plan. The lineups will change daily. They’ll change by period. So I wouldn’t read anything into that. That’s just maybe yesterday he went with the thirds, today he went with the seconds, tomorrow maybe he’ll go with the firsts. So nothing about that in terms of when they’re in there and who they’re working with.”

One of the smoothest plays of the day came in the one-on-one period when Gordon matched wide receiver Nsimba Webster in the slot and carried his coverage over the middle before diving to break up a pass. It was a great effort by Gordon from start to finish and notable from the standpoint the team is asking the 2022 second-round pick to keep his focus on playing the nickel position this summer.

Advertisement

Gordon played inside and on the outside a year ago and has a little less on his plate now.

“I do believe that just with everything, less is more,” Williams said. “So when you reduce the amount of things that he has to do in terms of technique assignments, they can’t help but get better. He works at it unbelievably. He is smart. He is instinctive. I don’t know if you remember last year, but he didn’t get into that spot (nickel) until this time last year. We didn’t rep him there in OTAs. So he’s got all those reps banked from the season underneath his belt on top of a little bit less on his plate in terms of being outside, so I would say yes, it has helped him on top of one more year in the system.

“But if you’re just talking about Kyler from what his skill set is, gosh, man, the guy’s have nicknamed him Spider-Man because he is so quick. He’s so agile. He is instinctive. He’s got spidey senses. When he makes a play and you go, ‘Wow, how did he make that?’ Ding, ding, ding, the spidey senses are going off. The quickness, the instincts, they’ve been showing up in a big way. Usually that happens Year 2. I’ve seen guys come in and Year 1, they’re getting their feet up underneath them. Year 2, they feel more comfortable, they know how to get lined up. They’re not just going through memorization. Now they’re starting to play football. That’s what you’re seeing out there.”

Linebacker Dylan Cole, who left Thursday’s practice about midway through with a member of the athletic training staff, was not spotted and the team has not provided an update. Wide receiver Dante Pettis remains on the non-football injury list.

Originally Published: Jul 28, 2023 at 4:13 pm