Let's talk tech Thursday #3
We're back for another Let’s talk tech Thursday!
This week’s we're exploring a growing undercurrent in the tech world: a reckoning with how much control we really have over the tools we use - whether it's the apparent political leanings of AI models, or the creeping dominance of big-tech and subscription-based software.
- A revealing new study suggests that ChatGPT's political outputs are subtly shifting, reminding us that even "neutral" machines are reflections of the data behind them.
- Meanwhile, the resurgence of LibreOffice underscores a growing appetite for free, open-source, and unfettered alternatives in a world where every click seems to come with a monthly fee.
There's also a look at the nuances of phone-based distractions. Which this newsletter definitely isn't. Right?
Let’s dig in...
Top stories
ChatGPT is shifting rightwards politically
Summary
While earlier versions of ChatGPT showed values aligned with the libertarian-left, an analysis of 3,000 responses for each version of OpenAI's chatbot shows newer versions are shifting toward the political right.
So what?
This research showcases an important point to remember about how Large Language Models (LLMs) function. These LLMs are the basis for most Generative AI models we use today, and are an amalgamation of literally millions of data points. They include everything from news articles to blog posts, academic papers to film and TV scripts, and so much more besides.
Often (almost always, in fact) each individual data point will posit an opinion, and so shift the dial of the LLM as a whole one way or another. With a large enough dataset, you can mitigate for this. In theory. In practice, every LLM on the planet has multiple biases baked in to its core. This article is a real-world demonstration of that, showing a tangible difference in the output of a machine in the wake of a change in political sentiment.
Whenever I run workshops on AI, it's probably this point more than any other that I want my attendees to understand. Unless you have built the model yourself, you cannot know:
- what data is being fed into the LLM, or
- how data within the LLM is weighted
If you ask an AI chatbot to show you a picture of a clock it will really struggle to show you a time that isn't 10 past 10. Why? Do a quick Google Image search for clocks. Almost every single picture of a clock that exists on the internet shows 10:10 as the time. It's the most visually pleasing time to look at, so every watch manufacturer in the world sets their products to that time before they take their catalogue shots. As a result the LLM, which is trained on data from the internet, "knows" that clock should read 10 past 10.
While AI models like ChatGPT are technically apolitical, their outputs are shaped by the data they’re trained on. This means shifts in political sentiment online, or manipulation of training data sources, can directly influence the apparent political leanings of AI-generated content. This research underscores the importance of understanding that no model is neutral, and that control over data inputs can effectively shape public discourse through AI. With enough resource, say owning one of the world's most popular social media platforms, cosying up to the leader of the free world and his adoring press, or flooding the internet with your own politically motivated AI slop, it is possible to change the playing field.
Related story: Russia-Linked CopyCop Uses LLMs to Weaponize Influence Content at Scale
LibreOffice downloads on the rise as users look to avoid subscription costs
Summary
LibreOffice, a free alternative to Microsoft Office, is seeing nearly 1 million downloads each week, driven by rising interest since its recent version release. Many users prefer it for its straightforward interface and avoidance of subscription costs.
So what?
As the cost of... well, everything, continues to soar many organisations are looking to cut costs. And it doesn't get much cheaper than "free". Many Free and Open Source Software (FOSS) alternatives such as LibreOffice are as well maintained and secure as their commercial counterparts, and as the article points out many local government bodies from Germany to Ireland are rolling out LibreOffice to save thousands on licensing costs.
While cost is clearly an important factor in these decisions, there are other benefits to using products like LibreOffice. For one, there is no ambiguity about who owns or accesses your data. Most instances of LibreOffice are not cloud-based, and The Document Foundation (the company behind LibreOffice) has no physical way of seeing your work - a luxury most Microsoft 365 users don't have. Equally, while LibreOffice will let you install third party extensions for things like AI tools should you need them, the organisation has "zero plans" to dabble in AI itself. For a number of people - especially those working in government or other sensitive jobs - this is very welcome news.
Cloud-based subscription models for our technology has been the norm for a while now, in everything from how we consume music to how we pay for our Office suites. Increasingly though, I think people are more and more concerned about this. Whether it's people leaving X for Bluesky, or taking the plunge with LibreOffice, who we let into our own personal tech-stacks is more and more important. FOSS, including LibreOffice, might be something you want to look into.
More from the world of FOSS: France and Germany unveil Docs, a homegrown alternative to Google Docs
See also: Free and Open Source Software - A Crash Course
What else is worth reading?
Why Simply Putting Away Your Phone Won’t Help You Focus
A recent study suggests that everyone's favourite handheld distraction tool might not be as to blame for your lack of productivity as you think. The bad news? That means you're the problem...
This is the second week in a row I'm sharing a CAST link, but if you're looking to roll out AI in any capacity, then you should absolutely check this out. A one-page, at-a-glance view of the major things to consider. There's a Miro board as well, if that's more your style.
Tech can help workers with dementia - new study
A study from the University of Bath shows that digital technology can help people with dementia stay in their jobs. Simple changes, and the right tools, can help ensure workers with early-stage dementia aren't excluded. And if you want another MedTech story, check out: Brain waves become spoken words in AI breakthrough for paralysis
We're three for three on Let's talk tech Thursday emails! Missed out on the others? You can check them out here.
Do you have a favourite tech story from the week? I'd love to read it. Hit reply and send the link!
In the meantime, have a great week, and see you next time,
Will