๐ค You have a keylogger on your phone
Dear curious mind,
This week, the shared insight dives into a topic that might make you uncomfortable: the keylogger potentially running on your phone. While we often protect our privacy on computers and the services we use, the keyboard on our phones can be a significant blind spot, sending every keystroke to tech giants. I share a discovery of a promising alternative for Android users that allows for true privacy while typing and transcribing your voice. Read on to learn how you can regain control over your data.
In this issue:
๐ก Shared Insight
Type and Transcribe on Your Android Phone with True Privacy
๐ฐ AI Update
DeepSeek releases updated R1 with performance boosts
Image Editing with Flux
๐ Media Recommendation
Book: Tiny Experiments to Promote Personal Growth
๐ก Shared Insight
Type and Transcribe on Your Android Phone with True Privacy
Most keyboards from big tech companies are essentially keyloggers. Every keystroke you enter is sent to their servers. Even if these companies claim they do not store or misuse this data, it does not feel good to provide everything I type on my phone to a company I cannot fully trust. Most would never accept this level of surveillance on a laptop, yet on phones it seems to be accepted by nearly everyone. I wonder: do most people realize this is happening?
Some time ago, I switched back to Google's Gboard from a privacy-friendly keyboard because I wanted to also have a powerful voice transcription on my phone, which became my daily driver on my PC as I describe in the Voice-Powered Chatbot Search: A Game-Changer for Information Discovery. Despite valuing my privacy, I found myself making this compromise because I was not aware of any solution that could perform quality voice-to-text processing locally on my device.
Recently, I discovered a promising alternative that respects privacy without sacrificing functionality. The FUTO keyboard, currently in alpha release but with already more than 100k installations from the Google Play Store, offers on-device voice transcription powered by optimized Whisper models. What makes this special is that FUTO solved a key limitation of Whisper for mobile devices - the original implementation of the voice transcription Whisper from OpenAI processes audio in 30-second chunks regardless of input length, making it inefficient for quick voice commands, especially on devices with limited compute power like mobile phones.
However, the quality of transcription with the FUTO keyboard app on my Pixel 8 Pro has been impressive. While struggling with specialized terminology, it performs quite well for everyday use. What amazes me is that all this processing happens locally without an annoying time delay. The keyboard also employs transformer models for text predictions - essentially using local LLMs for what they do best: predicting the next word.
There are some trade-offs compared to Google's offering. For me, the main drawback is that I constantly have to manually switch between English and German as input language. In contrast to the Gboard, this improves the typing and transcription results a lot. However, this was getting way more convenient when I discovered that I can tap a dedicated button instead of long-pressing the space bar to switch my input language.
If you are looking for an even larger change: Besides the FUTO keyboard itself, the transcription can also be integrated into the open-source Thumb-key keyboard app.
The open-source community continues to demonstrate that we do not need to sacrifice privacy for functionality. By running AI models locally on our devices, we can enjoy modern conveniences without exposing our personal data. If you value privacy and use an Android phone, I encourage you to try the FUTO keyboard.
๐ฐ AI Update
DeepSeek releases updated R1 with performance boosts [DeepSeek ๐ post]
DeepSeek has announced an update to their open-source model, DeepSeek-R1, with the 0528 release. According to the company, this new version brings improved benchmark performance and reduced hallucinations. It also adds support for JSON output and function calling, which are valuable features for using the model in applications or agentic approaches.
It is exciting to see an open-weight model continue to close the gap with the top proprietary models according to benchmark results. This progress is truly appreciated. On the other hand, many, myself included, are still eagerly awaiting the teased DeepSeek R2, which is rumoured to feature an entirely new architecture. Despite the wait for R2, this R1 update is a solid step forward. Furthermore, there are multiple voices stating that the distilled 8B version which is based on the Qwen3 model is rather good. You can test the latter on your on hardware as it only needs 6 GB of memory in the 4bit version (deepseek-r1:8b-0528-qwen3-q4_K_M in Ollama) and it even runs quite decent on CPU only systems.
Image Editing with Flux [BFL blog article]
The German company Blackforest Labs released FLUX.1 Kontext, a new suite of generative image models for in-context image generation and editing. These models unify instant text-based editing with classic text-to-image generation, delivering strong character consistency, context understanding, and editing capabilities. Two models named FLUX.1 Kontext [pro] and [max] are currently available via their API and partner platforms. Furthermore, they released a [dev] version as an open-weight variant in public beta.
An appreciated new release competing with the closed source gpt-image-1 from OpenAI and the just last week released open-weight model Bagle from ByteDance with image editing capabilities. We are on a path to replacing Photoshop and with that, the experts needed to do the image manipulation.
๐ Media Recommendation
Book: Tiny Experiments to Promote Personal Growth
I really love the book โtiny experiments: How to Live Freely in a Goal-Obsessed Worldโ by Anne-Laure Le Cunff. The main concept of doing something multiple times, documenting the process and evaluating the results afterwards is surprisingly simple, yet incredibly powerful. One example experiment highlighted in the book is to track for two weeks your energy level. Afterwards, you will understand at which time of the day you should do your most difficult tasks. The method is a game changer, and Iโve already started to implement tiny experiments in my own life. My current one is to reach out to at least one new solopreneurs, creators and AI enthusiasts and connect with them on a more personal level.
The book is packed with valuable knowledge and insights. I found myself taking notes constantly while reading. Itโs especially helpful if you are on a path that is not yet fully defined, like I am currently.
What made this book even more special was the unexpected references to people I previously discovered and value in the creator and solopreneur AI space: Pieter Levels, Arvid Kahl, and Ben Tossel. It was a nice surprise to see their names within the pages.
If youโre interested in exploring new opportunities and living a more intentional life, I highly recommend giving the book tiny experiments a read. And if you do, share your thoughts after reading it with me. I would love to hear from you.
Disclaimer: This newsletter is written with the aid of AI. I use AI as an assistant to generate and optimize the text. However, the amount of AI used varies depending on the topic and the content. I always curate and edit the text myself to ensure quality and accuracy. The opinions and views expressed in this newsletter are my own and do not necessarily reflect those of the sources or the AI models.