AI Updates - April 2025

GPT-4o Image Generation

Two weeks ago, OpenAI launched their new image generation model, and it is truly impressive! Everyone is creating Ghibli-style avatars and action figures for their profiles. The standout feature of this model is its ability to generate highly realistic images with accurate text 🎉.

Here is an example I created with the prompt:
“Generate an image of a 2021 Skoda Octavia Combi in dark blue with a transparent background”:

Currently, it is only available in ChatGPT, but the API is expected to follow in a few weeks.
Learn more about GPT-4o Image Generation

AI 2027 - A Scenario of AI’s Future

The AI Futures Project has published a scenario (or essay) on the future of AI from now until 2030. The main author is well known to predict the AI future very well and the project even offer bounties for better future predictions.

You can find the full scenario here: AI 2027
For a summary, visit: AI 2027 Summary

If you’d like to predict the AI future yourself, try this handy tool:
AI Futures Prediction Tool

AI Tools – What About Data Protection?

Vischer AG has provided a great overview of the data protection implications of various AI models and their usage: Part 25: AI Tools – What About Data Protection?

Azure OpenAI and Google Gemini stand out as the best options in terms of data protection (when using hosted models).

Meta AI Launched Llama 4

Meta AI launched Llama 4 on a Saturday 🤔. It is a multimodal model that can handle text, images and video. The initial feedback from testers has been underwhelming. While it performs well in benchmarks, its real-world performance seems to fall short. There are claims of “tricks” being used, as highlighted in the article:
LMArena Tightens Rules After Llama 4 Incident. Let’s see what the future of this model family holds.

For the official launch post, visit: Meta AI Blog - Llama 4 Multimodal Intelligence