What Dolly 2.0 Means for the Digital Workplace

The frenetic pace of new AI tools and services continues to set the news pages alight with stories and a barrage of text and image AI-generated content. But for most businesses, daily tasks are performed traditionally as always. Will Databricks' Dolly 2.0 and commercial-use large language model (LLM) change how typical organizations think about adopting AI?

Databricks is another generative AI player working on the fast-release track, with Dolly 2.0 coming out a couple of weeks after the first edition. Billed as the “first open, instruction-following LLM for commercial use,” Dolly 2.0 has been crafted with Databricks’ own in-house-generated learning dataset, and it encourages businesses to modify that training data to deliver more relevant insights for your organization.

You can try Dolly 2.0 over on GitHub or deploy it from here, asking Dolly questions and testing its performance. The current web engine runs on Nvidia A100 GPU hardware with 40GB VRAM storage. Predictions should complete within 6 seconds, informing your own hardware requirements when planning to use Dolly locally.

Dolly 2.0 Is Not Another AI Clone

To make it open-source and distributable, Dolly 2.0 uses a 12-billion-parameter language model. This is based on an open-source Eleuther AI Pythia model family but fine-tuned on a small (only 15,000) set of instruction records, created by Databricks’ employees, delivering an open-source and accessible solution to all.

That gives Dolly 2.0 strong appeal for enterprises, researchers and college users, enabling any organization to create, own and customize its own powerful LLM. That’s without the need to share proprietary data with third parties or pay for API access.

So, what can the open-source (including training, weights and dataset) Dolly 2.0 do for your business?

Interest is certainly high. Dolly V1 has been downloaded 13,000 times at the time of writing, and Dolly 2.0 is already up to 4,500 downloads in a matter of days. The immediate implications for tech leaders and operators are:

• Enterprises can fine-tune the instruction set to meet their needs, assuming they have the resources and knowledge to edit and create new models.
• Through open source, enterprises have fine control over their AI tools and are better able to manage compliance, governance and data security needs.
• As with ChatGTP and contemporary products, businesses can use AI for chatbots, knowledge centers, application development, computer vision use cases and scientific research.
• All generative AIs are still fallible, capable of producing incorrect inferences and passing on wrong information, so they should still be moderated by humans where possible.

Related Article: Why Meta's Investment in Generative AI Is an Investment in the Metaverse

Dolly 2.0 Could Control the Horizontal and the Vertical

Note that Dolly isn’t the only open-source ChatGTP-alike. TechTalks' Ben Dickson provides a useful list of other models (LLaMa, Alpaca and others — what’s with the wooly obsession, creators?) that may offer broader features or scope for innovation, as well as future models that will focus on specific markets (for example, health or retail).

While experimenting with ChatGTP and Dolly is fun, most businesses are waiting until they can create their own practical applications with a solid use case. Dolly 2.0 (and 3.0) and contemporaries will help round out the final element of the GAI market, helping businesses create services and tools for their operations.

There will also be a race to develop new apps for vertical and horizontal businesses. As Gokul Rajaram, product helper at DoorDash, notes; “The AI stack has 4 layers. From bottom to top:

Learning Opportunities

ON DEMAND

Collaboration in a Distributed Workplace: Tools and Techniques of Top Performing Teams

Watch Now

ON DEMAND

Improve Knowledge Sharing: The Key to Employee Productivity

Streamline team collaboration and engagement

Watch Now

ON DEMAND

How McDonald’s Drove Productivity Through an Elevated Employee Experience

In the new remote/hybrid workplace, work/life boundaries are blurred and workplace stress is a top driver of mental health needs.

Watch Now

ON DEMAND

How to Future-Proof Your Employee Experience Strategy in 2023

A framework to navigate through economic uncertainty

Watch Now

ON DEMAND

Challenges to Efficiency in 2023: Your Employees Need the Digital Workplace of the Future

The era of asking employees to do more with less is upon us

Watch Now

ON DEMAND

The Essential Role of Communicators in Fostering Wellbeing in the Digital Workplace

Join us for practical insights on how digital communicators can support employees to thrive in the digital workplace

Watch Now

ON DEMAND

Addressing Employee Needs and Wants with a Digital Workplace

The workplace is getting more and more digital – both in how we work and where we work

Watch Now

ON DEMAND

Maintaining a Human-Centered Approach During Digital Transformation

When it comes to digital transformation - people drive change, not technology

Watch Now

ON DEMAND

The Evolution of Employee Recognition

Leveraging the power of appreciation to improve the employee experience

Watch Now

ON DEMAND

Are You Blocking Your Internal Comms Team Without Knowing It

Best practices & tips to help IC teams thrive

Watch Now

ON DEMAND

Cyber Resilience In The Cloud Is Not Optional

In today’s age of ransomware and widespread cyber attacks, your cloud data needs to be protected.

Watch Now

View All

• Infrastructure (e.g., Nvidia)
• Foundational models (e.g., GPT)
• AI tooling (e.g., Hugging Face)
• Apps (e.g., Harvey for legal)

There will be mega winners in both horizontal and vertical AI apps.”  I include functional (sales/marketing/etc.) AI apps within the horizontal bucket; vertical here refers to an industry focus (e.g., healthcare, legal).

Related Article: How Smaller Digital Workplace Vendors Are Integrating Generative AI

The Future of Dolly 2.0 and Friends

Dolly 2.0 is just the starting point for most organizations looking to start building their own generative and creative AI tools.

While the race is on for LLMs as generative agents that act with human-like behavior, most businesses will be looking for more pragmatic solutions.

While the hype moves incredibly fast, real businesses take a little longer to get going. And it is very early days for Dolly 2.0, so the burgeoning market of AI reports and discussion posts, like the “State of AI” are still getting to grips with it.

And, as with most hyped solutions, there will be teething issues. For example, be aware that some of the information used to build and train Dolly 2.0 comes from Wikipedia and other similar sources, so there may be some inaccuracies in its output.

For a broader understanding of where businesses plan to commercialize the likes of Dolly, I suggest you watch “The AI Race — Training & Commercializing 2T Parameters”  by Dr. Alan Thompson in which he discusses where these products are headed and how products 10-times as powerful will influence accelerating development — and what the creative possibilities might be.

Have a tip to share with our editorial team? Drop us a line:

About the Author

Geoff Spick is a Bournemouth, England-based tech writer with over 25 years of experience across Gartner, Infor and a host of startups and SMBs as a freelance marketing content creator. With fingers on the pulse of digital business, automation, no/low-code and productivity solutions, he is familiar with most bumps on the road to digital transformation. Connect with Geoff Spick: