Week 11: GPT 4 & new Python tooling

A thrilling week with many topics and opinions to research after the launch of GPT 4.

This week was a thrilling week: Lots of cool stuff happened at work, that I still can’t disclose. I explored a new dinning place in my hometown with friends. GPT 4 was released. Eventually, I discovered a new Python tool for dependency and package management, that I really enjoy.


I am still on the waiting list, to get access to the GPT 4 API. But I am already keen to learn myself how much better it is compared to GPT 3.5, which already provided great results for some use cases I have in mind for one of our flagship products. I really enjoyed the live stream where Greg Brockman showcased the new features, but also talked about the limitations.

Rewatch the live stream on YouTube.

Two days later, Microsoft shared its view on the future of work and provided some insights into the AI solutions, that will be integrated into Microsoft 365 apps, Teams and its business tools in general. Impressive and the logical move after their high invest into OpenAI and after showcasing how they want to integrate this technology into Bing and Edge.

The Future of Work With AI – Microsoft, March 2023 Event on YouTube.

Besides these bold announcements by OpenAI and Microsoft, they also had to deal with a story published by the Platformer, that stated Microsoft has closed down their ethics and environment team. Not the best story you want to read as the CEO between two large announcements. So, it was no surprise that they reacted with a response, where they confirmed the close down of the team, but also explained that they built up a team for responsible AI. We will see what they will really do in the future.

Personally, I think every company doing AI or using AI services should have an office to take care of responsible use of AI. Otherwise, we might outperform the climate catastrophe on the left lane. Hopefully, we won’t push the climate catastrophe forward with the enormous amounts of power required for AI services, but develop more efficient systems that need less power and resources.

Microsoft built a large super computer to train the GPT 4 model.

AI criticism & evangelism

Besides all the hype around the launch of GPT 4 and the Microsoft Office integrations, I also read some really fair and good articles criticizing all the AI hype. These articles are from German publications covering data privacy and technology. But I guess, you will find an AI, that will help with a translation. 😉

  • Wie Millionen Menschen für die KI schuften – Netzpolitik.org published an interesting interview with Milagros Miceli, who is leading a team at the Weizenbaum institute. In this interview, they were discussing the dark side of AI – people working in underpaid and bad jobs to tag and verify the data used to train the AI. If you read this interview and pent some time researching this topic, you discover that we are talking about one of the worst incarnations of the gig economy.
  • Bullshit, der (e)skaliertJürgen Geuter published a long essay on the German tech news site Golem. I really like how he takes an intense look at the full spectrum between criticism and evangelism. It is a long read with a ton of well-placed links in it – every link is worth checking out. The point that is most intriguing for me is, “The space of our possible visions of the future is larger than the straitjacket that belief in AI imposes on us.” We are currently heading again into a position where everything can be solved with the hammer – AI – that turns every problem into a nail.
  • ChatGPT Is a Blurry JPEG of the Web – another interesting long read, talking about the mediocre quality of texts generated by AI systems.


Some more interesting links on AI topics.

PDM – fresh tooling for my projects

I love good tooling and automating the boring stuff. I always used tools like pipenv, npm, mix and lately poetry to handle package and dependency management in my work projects and side projects. This week, I discovered pdm after I had some issues with the fish shell integration of poetry. Pdm has some nice features – PEP-621 compliant project metadata, a PEP-517 compliant build backend, support for PEP-582, and it supports a pnpm like centralized cache. After playing around a bit with it, I was sold on it. I converted django-tailwind-cli and my django startproject template to use pdm.

The conversion from poetry to pdm was super easy using the command pdm import -vf poetry ./pyproject.toml. The only thing I had to fix manually was the dev dependencies, which didn’t get converted from the poetry format to the pdm format.

Currently, I am planning to introduce pdm at my company for a large project, that is still managed by manually editing a big requirements.txt file. Pdm will be part of the process to modernize the whole project by introducing proper dependency management, pre-commit, black and ruff.

We will stick to poetry, pre-commit, black and ruff. Some changes might be too spontaneous to undertake them. 😉

Cover Image: The picture for this week’s edition was published by Jukan Tateisi on Unsplash. I picked it because it shows dedication and focus even though a big and may be overwhelming adventure lies ahead of the small child. Occasionally, I wish to be able to focus like that and take one step after another.