Meet the Nordic Version of ChatGPT

The New AI Taught Itself a Language - Without Being Trained In It

Author

Peter Juel Jensen
October 10, 2023

Peter’s Pioneers: Checking the Nordic AI Pulse.
Welcome to the fifth edition of Peter’s AI Pioneers newsletter. Dive in for a roundup of this week's AI news, deals and insights covering the Nordic region.

📚 In This Issue:

  1. Meet the Nordic Version of ChatGPT

  2. Are Smaller Languages “Isolated” From AI?

  3. Why AI is Watching Vegetables Grow

  4. News Roundup

  5. Cool Tools and Companies

  6. Podcast Insights: The EU view on AI, AI and the Climate

  7. Upcoming: Nordic AI Events You Shouldn't Miss

Top Three News Stories

💻 Meet the Nordic Version of ChatGPT  

GPT-SW3, a product of AI Sweden, is a counterpart to Large Language Models (LLMs) like OpenAI’s GPT-3, which power generative AI solutions like ChatGPT.

GPT-SW3 owes its inception to state-funded efforts, specifically tailoring its capabilities to the Swedish language for reasons of national sovereignty. Addressing concerns of data transparency, which global counterparts often do not divulge, GPT-SW3’s creators pride themselves on being open about the datasets it's trained on, aiming for a representation as close to the Swedish populace as feasible.

The Swedish LLM is the result of a 72-day computational marathon using Berzelius, a Swedish supercomputer. After processing vast quantities of text in Swedish and its Nordic neighbour languages, GPT-SW3 was born. Interestingly, while the model exhibited proficiency in anticipated languages, it surprisingly showed capability in Faroese, a language they hadn't explicitly trained it on.

Ideas for similar LLM projects are percolating in other Nordic countries.

My take: I love the Swedish project. We should do the same in Denmark, but let's skip the state-owned language model.

Trying to build our own language models is a bold move. It’s great to see such strong ambition from our state, trying to be a leader in tech. But even as someone who’s all in on Nordic AI startups, I have to wave a yellow flag here and acknowledge the risks.

Competing with global tech giants means stepping into the ring with heavyweights. It requires vast resources, long-term commitment, and specialised knowledge. Just last week, the prime minister nearly shut down Aula — a public sector communication tool that’s had its struggles - with her words, causing some serious reflection.

Should we, instead, be focusing more on how to build up AI startups in our community? Maybe we should consider using existing models, even on-premise ones, in the public sector and planning how to gather data to refine and improve those models? In this context, it is worth mentioning that LLM open-source models are progressing by leaps and bounds. For example, this model was released on GitHub just last week.

Tech is changing fast, and our path forward needs to be smart, focusing on areas where we can make a meaningful and achievable impact.

🧱 Are Smaller Languages “Isolated” From AI?

Generative AI systems like ChatGPT are trained on enormous amounts of data. The quantity and quality of data are decisive for how good the models can be.

In both these respects, data from smaller languages, such as Swedish, is much less prevalent and easy to get to than for major languages. In addition, it is far from certain that a small country like Sweden can raise the resources required to train national language models.

This leads to a situation where the labour markets in countries with small native languages may be more “isolated” from the immediate effects of AI than larger countries with more popular languages - unless increasingly advanced algorithms and finding alternative data sets can compensate for these shortcomings.

My take: Nordic languages have an inherent disadvantage due to the relative scarcity of high-quality data compared to major languages such as English or Mandarin. Resources for bespoke AI models tailored to national, linguistic nuances can also be scarce. This may also have contributed to the initial performance of GPT-SW3 being somewhat below that of a ChatGPT.

One worry is that this could result in the Nordics having an AI literacy gap, at least where models fine-tuned for the Nordic languages are concerned.

There is a risk that this could slow AI-driven transformation in the region. If so, that might not be all bad. It could offer us more time to adapt, strategize, and potentially find niche areas of expertise or integration for AI, creating a differentiated value proposition.

However, the Nordics have distinct advantages, such as our high English proficiency. In the ever-evolving world of AI, it is also likely that increasingly sophisticated algorithms can efficiently learn and adapt even with limited datasets. This capability will become important not just on a language level but for companies too (a subject I will return to in future newsletters).

As I wrote about last week, AI researchers are also finding alternative sources for language content, such as Hestenet in Denmark.

These factors mean that smaller countries are poised to be able to quickly leapfrog into the AI landscape, reaping the benefits of tailored AI solutions without the setbacks associated with data scarcity.

🥦 Why AI is Watching Vegetables Grow

Nordic Greens, a renowned Danish horticulture company previously focused on tomatoes and cucumbers, has integrated advanced AI into its new venture into the salad market. Using cameras, they continuously monitor crop health, detect pests, diseases, and deficiencies, and amass data for crop growth and yield forecasts. Interestingly, the consistency of lettuce crops, as compared to cucumbers or tomatoes, simplifies the integration of AI into its cultivation strategy.

With a whopping investment of 250 million DKK, the company has pioneered a fully automated lettuce production facility, capable of producing 3,000 tonnes annually, making it weather-independent.

One of the standout features of Nordic Greens' lettuce operation is its state-of-the-art automation. From seeding to packaging, the entire production process is automated, minimizing human intervention. While there were many available systems on the market, the Nordic Greens team showcased their innovative spirit by opting to design their unique system.

My take: The adoption of AI in agriculture presents a paradigm shift, heralding a new era of precision farming. Add it to the region’s already strong reputation for high-quality and organic produce, and you have a winner.

Some areas where AI can increase yield and quality include soil analysis, predictive planting and harvesting, and field monitoring. AI-driven robotic systems can also be used to automate tasks like planting seeds, picking fruits, and even weeding.

For early adopters of AI in agriculture - and companies developing such solutions - the competitive advantages are palpable. Additionally, their proactive approach positions them as industry leaders, setting standards and benchmarks that others will later strive to achieve.

🎙Nordic News Roundup

  • New innovation hub aims to support SMEs’ use of AI. [DK] (Ritzau)

  • Is AI going to have a bigger impact than the internet? [SE] (Dagens Industri)

  • AI may have influenced one out of five Norwegians in a recent election. [NO] (Computer World)

  • Pain relief is about to get a boost from AI [SE] (Forskning)

🛠️ Cool Tools & Companies

  • Abtion from Denmark uses AI as an active part of its digital innovation/UX/App development services and solutions.

  • Pistachio from Norway uses an AI-driven security training platform to help companies counter cyber threats.

  • Responsibly from Denmark offers AI-driven sustainability tracking across the supply chain.

  • Saidot from Finland helps organizations harness the potential of generative AI safely.

  • MyGoodTape from Denmark uses AI in its secure, encrypted transcription solutions.

Do you know cool tools or companies I should feature in the newsletter? Let me know!

🎧 Listen To

  • AI-verkstan [SE] - Why SMEs should care about AI.

  • AI Revolutionen [DK] - What will AI mean for the future of your career?

  • I morgen [NO] - Are the Nordics becoming a global programming hub? And what does that mean for AI?

📅 Training, Meetups & Conferences

Got an AI story or insight from the Nordics? Share it with me!

Reply to this email or connect on LinkedIn.

Until next time,

Peterhe Future