Here at Sopra Steria Bulgaria, we are a team of 120 people, who span various domains including software development, project management, and consultancy services. AI can be utilized in almost every aspect of our daily work life. As a business, we add significant value to our clients by reaching creative solutions together. But what is more intriguing is how our employees use the technology to generate even more value.
Our OneTeam Day sessions are a place where our colleagues present topics, and we explore different areas as a team such as finance, technology, or design, conduct project overviews, or simply play games together with a team-building effect.
We come together to chat about things we find interesting, away from our usual projects. It’s a time for lively talks that keep our curiosity alive and our thirst for new knowledge satisfied.
During the last OneTeam Day, we had so many deep and insightful conversations, where AI took center stage that were too compelling not to share. With this post, we would like to bring you some key takeaways.
Hardware Advancement
As we delved into the topic, we had to know how good the hardware that underpins the foundation of AI became. We talked about how computing power changes as a prediction technology and directly impacts decision-making. With an abundance of sensitive data, comes the main threat of how the business protects their data and positions. Robust security protocols and privacy safeguards are the way to go, to prevent data breaches and regulatory non-compliance.
“In Goes Data, Out Goes Magic”
Then we dived into Generative AI and LLM – the world of “In Goes Data, Out Goes Magic,” we explored how these models generate language on a grand scale, leading to emergent behavior that can often surprise us. We all love how AI is great with the language, very fluent, very creative, and very convincing but is this enough to trust it? That is why Fact-checking is primary when using AI for social copy, video scripting, or event collateral. Hallucination* and overreliance on AI are two other considerations of why users must check if the information is true or false. (*Hallucinations can stem from the AI erroneously connecting inputs to another idea.)
The concept of singularity
We had a look at some intriguing breakthrough device interactions like Rabbit, a single device that will replace multiple ones. The AI goggles that blend the digital and real world into a believable world. And here we begin to question ourselves, “How are confidence thresholds determined?” At the mention of limitations and boundaries, the inevitable other side of the story emerged – the concept of singularity.
“We cannot offload our ethical, professional, or moral obligations to the AI”.
Emphasis was placed on ethical responsibilities, model adjustments, and the nuanced context-setting process. A characteristic of AI is that it relies on a large amount of data collected over the years. Natural language works in this way: when you ask a question, it’s usually ambiguous and you always get an answer. AI needs context to give a correct answer. But do we let AI make those judgments?
Garbage in – garbage out: What you train the AI on
We highlighted the challenges of ensuring unbiased outcomes. And when speaking of bias – we can all think – is AI an accurate representation of the real world? And if so, then is the data a reflection of our own biases? The discussion centered on the impact of data on AI, how the models are trained and retrained, and the importance of giving feedback. Poor quality or limited data can lead to inaccurate predictions or insights. Moreover, the relevance and accuracy of data can change over time.
Side Hustles
When used for side hustles, opportunities were discussed particularly in prompt engineering, GPTs, and creative work. AI unveils an unimaginable exploration path, and it is really exciting. We touched upon how one can profit from it and still protect their intellectual property so that the work does not just get integrated into the next great development.
To end with we had a practical demonstration showcased AI Copilot’s efficiency in peer programming, emphasizing its role as a code-writing assistant. The tool’s ability to respond dynamically to prompt changes was explored, opening avenues for both coding and side projects.