Hi Synapticians!

Yesterday, OpenAI introduced their new models (o3 and o4-mini) during a livestream, and the opening line said it all: “There are some models that feel like a real step into the future. GPT-4 was one of them. Today is also one of those days.”

There are a few ways to react to this kind of announcement. You might think, “According to AI companies, we’re stepping into the future every other day,” and honestly, it’s starting to feel a bit exhausting. Or maybe you’re more like, “Nice, another tool to help me be even more productive 😁.” Either way, let’s put the hype aside for a second and actually look at what these models bring.

The biggest change is that these models are natively trained to use all of ChatGPT’s tools like browsing the web, running Python code, or analyzing files. In the demo, the model looked at an image that wasn’t super clear and just zoomed in by itself to get a better view. It might sound simple, but it’s actually really smart. These tools are becoming the main value, for example, if the model isn’t good at math, it just grabs a calculator.

And in terms of performance, it’s seriously impressive. Based on Codeforces rankings (a programming website), the “o4 mini” model would be ranked among the top 200 developers in the world. That’s honestly kind of amazing.

Top AI news

1. OpenAI’s o3 and o4-mini bring tool-using AI
OpenAI has released o3 and o4-mini, two new models that can autonomously use all ChatGPT tools—web search, Python, file analysis, and image generation. These models are trained to decide when and how to use tools to deliver accurate, well-formatted answers. Simon Willison has already integrated them into his LLM plugin, allowing users to test them easily. The models can even generate creative visuals, like a pelican riding a bike. This marks a significant step toward more agentic and useful AI systems.

2. OpenAI releases Codex, a terminal-based coding agent
OpenAI has released Codex, a lightweight coding agent that runs directly in the terminal. Designed to assist developers with code generation, explanation, and debugging, Codex offers a streamlined, integrated experience. It’s OpenAI’s alternative to Anthropic’s Claude Code, reflecting a broader trend toward specialized, embedded AI tools. Codex focuses on enhancing developer workflows without disrupting them.

3. AI Won’t Kill Creativity
The rise of television didn’t destroy cinema—it pushed it to innovate. Studios adapted through new formats, technologies, and collaborations. This historical resilience offers a valuable lesson for today’s creative industries facing AI disruption. Rather than fear AI, creators can embrace it as a catalyst for reinvention and growth. Adaptability, not resistance, is the key to thriving in the face of change.

Bonus. ChatGPT guesses photo locations
OpenAI’s latest models, o3 and o4-mini, can analyze images to deduce their location, even without metadata. Users are testing this by uploading photos of streets, menus, or buildings, and the AI often guesses the correct place. This trend, similar to GeoGuessr, highlights the power of visual reasoning in AI. However, it also raises serious privacy concerns, as there are currently no safeguards to prevent misuse. OpenAI has not yet addressed these risks in its safety documentation.

Meme of the Day

They are so good at producing models … but naming them seems the hardest task 😅 

Theme of the Week

AI for Education - Paper Review

What if AI could help teachers, not replace them? This paper dives into how to design AI tools that earn teachers’ trust—especially the skeptical ones. From real interviews to surprising insights, it’s a must-read for anyone building tech for education.

Stay Connected

Feel free to contact us with any feedback or suggestions—we’d love to hear from you !