Hey there!
Duru here from Novus, and May has been anything but slow in the AI world.
From Google dropping 100 AI announcements at I/O to Baidu working on a system to talk to animals (yes, really), this month’s updates have been equal parts ambitious and bizarre.
Meanwhile at Novus, we’ve been out in the world — from panels to summits — sharing what we’re building with Dot, and connecting with the growing AI community both locally and globally.
If you’re new here, welcome! I round up the biggest AI news, Novus updates, and personal insights in this monthly digest. You can also subscribe to our bi-weekly newsletter if you want the latest updates delivered straight to your inbox.
Let’s get into it.
May 2025 AI News Highlights
Google Goes Full Gemini at I/O 2025
Google I/O 2025 was an AI extravaganza. The company unveiled 100 new features, with Gemini models powering everything from search to smart glasses.
The most ambitious reveal? Gemini 2.5 Pro and Flash, which support 1 million-token context windows and a new “Deep Think” mode for tougher reasoning tasks.
Key Point: Google launched 100 AI features at I/O, showcasing Gemini 2.5 Pro, generative video tools, and a new $250/month AI Ultra plan.
Baidu Wants to Make Animals Talk
Baidu has filed a patent for a system that would use AI to interpret animal sounds and convert them into readable human language.
It’s early days, but the goal is to decode vocal patterns in pets and potentially understand what they’re trying to communicate.
Key Point: Baidu’s new AI patent aims to translate animal sounds into human language, opening up new frontiers in cross-species communication.
Duolingo’s AI-First Pivot Sparks Backlash
Duolingo cut 10% of its contractor workforce and announced an AI-first strategy for course content, leading to user complaints about quality.
Longtime learners, especially of Japanese, say cultural nuance is disappearing—and that AI can’t yet match the human touch in language learning.
Key Point: Duolingo’s AI-first shift led to layoffs and user backlash, raising concerns about quality and cultural accuracy in AI-generated education.
FutureHouse’s AI Tool Tackles Scientific Discovery
FutureHouse, a Google X spinout, launched a new AI platform that helps scientists explore biological research data without writing code.
Its first application in microbiome research allows researchers to model bacteria-human interactions more intuitively.
Key Point: FutureHouse launched an AI tool that turns biological data into interactive knowledge graphs, simplifying scientific exploration.
Novus Updates
The Novus team has been everywhere this month — from Istanbul to İzmir to Denver — sharing our vision for purposeful AI and showcasing how Dot is making it real.
- At Zorlu Holding’s “Geleceğini Yaz” event, our co-founders Egehan and Vorga spoke about solving real-world challenges with Dot, while Halit, Rehşan, and Doğa gave hands-on demos that sparked great conversations.
- At Webrazzi XYZ 2025, Egehan joined a panel on how AI agents are reshaping post-ChatGPT brand strategy, highlighting what sets Dot apart in this shift.
- During MEXT’s gathering for Denver’s startup ecosystem, Vorga introduced Dot to a global audience, while Duru and Sercan connected with leaders and innovators on the ground.
- At the AI Summit in İzmir, our CRO Vorga Can shared insights on building AI systems with real-world impact, in a session moderated by Ali Rıza Ersoy and hosted by ESİAD and EGİAD.
- Meanwhile, our CEO Rıza Egehan Asad attended the TRAI Mayıs Çalıştayı, exchanging ideas with local AI pioneers in a refreshing and forward-thinking environment.
Each event reminded us why we do what we do: to build technology that doesn’t just impress — but actually empowers.
Educational Insights from Duru’s AI Learning Journey
Why Telling Chatbots to Keep It Brief Can Backfire
We all want quick answers. But a new study from Paris-based Giskard reveals that asking AI for short replies might actually lead to more hallucinations. The research, tested on GPT-4o, Claude 3.7 Sonnet, and Mistral Large, found that when brevity is prioritized, accuracy suffers.
Why? Short answers leave no room for nuance. The AI might sense something’s off in the question, but without space to explain or clarify, it just blurts out an answer — confidently wrong.
This matches findings from another recent paper, Calibrating Verbal Uncertainty, where researchers reduced hallucinations by 32% simply by adding phrases like “it’s likely” or “this may be.”
So next time you prompt an AI, consider this: giving it space to elaborate might be the smartest move.
Is the Internet Still Human?
Have you ever read something online and thought, “This feels a little… machine-made”?
You’re not imagining things. According to Amazon Web Services, 57% of all online text is now generated or translated by AI. A separate arXiv study puts it at 30% minimum, and that number is growing fast.
Some analysts believe that by 2026, 90% of web content will be AI-generated (The Living Library).
That’s wild. It raises serious questions: How do we tell what's real? Can we still trust what we read? Is creativity getting flattened into algorithmic sameness?
I use AI in my own work, and I love what it can do — but I also try to stay aware. I still think it matters when something is written with intention, by an actual person. So for now, I’ll keep writing my own words — slowly, but meaningfully.
And I hope we all keep asking: who’s really behind what we’re reading?
Until Next Time, Stay Curious
That wraps up this month’s AI learning journey from my corner of Novus. If you’ve made it this far, I hope it gave you a fresh perspective or at least a few things to think about the next time you prompt your favorite model.
And if you want something a little more fun than a newsletter (yes, I said it), head over to our YouTube channel and check out Açık Kaynak — our podcast where we talk about AI, startups, and the messier side of building in public. It’s honest, unfiltered, and way more casual.
See you next month!
