Welcome to TechFreedom Dispatch #6
Tom is back from his summer break, full of optimism, energy and a distinct lack of remembering quite what it is he does. So naturally, I have my attention to getting another Dispatch out.
In this one: some things from us, some things from elsewhere, a slightly alarming Microsoft story, communities building their own places on the internet, more evidence that ATProto might be interesting well beyond Bluesky, and the curious case of LibreOffice hiding inside an AI tool.
Let's go.
Things from us
Free TechFreedom taster session
Join us next week Wednesday 9th September at 2pm for a free, 45-minute TechFreedom taster session based on the session we ran at TechNExt.
Bits of content, bits of discussion, nice and relaxed. We'll explore some of the less obvious risks hiding in our organisational technology stacks, and what we might actually do about them.
Sign up here
TechFreedom cohort - new dates and a (shh) special discount code
We've added new dates for our next full TechFreedom cohort, starting 21st October.
Over three sessions we'll help you map the technology your organisation depends on, explore where the risks and dependencies actually sit, and work out practical steps towards having a bit more freedom of movement.
Find out more about the programme
And, because you read the newsletter, here's a not-very-secret secret:
Use discount code TF100 when you sign up for £100 off.
FIELD STATION experiment - Shared Inference
As you may or may not know, TechFreedom is now part of FIELD STATION and we're ramping up to run our first live experiment.
The question behind this one is deceptively simple:
How might we run AI differently?
For Experiment 001 we're looking at Shared Inference: what happens if, rather than assuming AI needs to happen in a giant corporate data centre somewhere, we try pooling the compute available across several ordinary machines?
Can a collection of smaller devices collectively run something more capable than any one device could on its own? Is it useful? Is it practical? Is it ridiculous?
We don't know yet. That's why it's an experiment.
Come and find out with us on Tuesday 15th September, 2–3pm.
Read about the experiment
Join the live experiment
Gatherings
We're heading to London in a couple of weeks for the AI, Communities & Coops — CoTech Autumn Gathering.
If you're going too or fancy meeting up the evening before let us know.
We're also heading to MozFest next month. Again, if you're going and fancy a TechFreedom-ish gathering over a drink/coffee/appropriately sovereign beverage, give us a shout.
Now onto things we've seen that caught our attention
1. Start a Bonfire?
We've talked quite a lot about moving away from big centralised platforms. But the interesting question isn't really "what is the open-source version of Facebook?" It's "what might we build instead if Facebook wasn't the starting point?"
Bonfire is maybe becoming an answer?
They've shifted from building another social feed towards tools for communities to organise themselves: groups, events, governance, decision-making, discussions and shared spaces, all built on open and federated infrastructure.
"people weren't turning up asking Bonfire for a better microblog. They were asking for somewhere their community could have a home."
TechFreedom isn't just about substituting one product for another but that different technology allows us to organise in different ways.
Maybe don't build another feed. Start a Bonfire.
Bonfire for communities
2. And this is why continuity is on the risk register
This one is particularly relevant to the charities and nonprofits among you.
Slate has reported on small nonprofit organisations which say they lost access to, and in some cases permanently lost, years of data held in Microsoft 365 following changes to Microsoft's nonprofit grant programme.
There appears to be a messy story around notices, licence expiry dates and exactly what different organisations were told. Microsoft says it had been notifying nonprofit customers about the changes; affected organisations describe warnings that were missed, unclear or contradicted what they subsequently experienced. There are now multiple public reports from nonprofits saying data could not be recovered.
We aren't sharing this to say DON'T USE MICROSOFT (although...).
We're sharing it because it is an unusually brutal illustration of one of the questions we ask organisations:
What happens if a supplier changes the terms of the relationship?
Continuity, cost exposure and lock-in can sound like fairly dry risk-management terms, right up until they aren't.
Read the Slate article
3. So what actually is ATProto?
Heard of Bluesky? EuroSky? Well underneath much of that is the ATProtocol and Krzysztof Kowalczyk has written an excellent, detailed introduction to how ATProto actually works.
One of the important distinctions is that Bluesky the company, Bluesky the social network and ATProto the underlying protocol are not the same thing. ATProto is intended as infrastructure on which lots of different applications can exist, sharing common approaches to identity and data.
Particularly interesting from a TechFreedom point of view is that protocols are becoming increasingly important to the future of technology and, in our opinion, TechFreedom Organisations.
A very thorough introduction to ATProto
4. Why is there a copy of LibreOffice inside ChatGPT?
Simon Willison spotted something curious while poking around the runtime installed by OpenAI's coding tools. Alongside Python, Node.js, Git and various other utilities sits an installation of LibreOffice, which is pretty fascinating.
Fascinating in a way of "Is this good? Bad? Not really sure?! One interpretation is simply practical: LibreOffice is an excellent open-source tool for manipulating office documents, so of course you would use it rather than reinventing all of that functionality, but it raises a question we don't quite have an answer to yet...
What does it mean when open-source software increasingly becomes the invisible infrastructure underneath proprietary AI products?
Is that a win for the open ecosystem? Does AI make open tools such as LibreOffice more strategically important because agents can use them without caring about brand or interface or does it just make the open-source layer even less visible while value and power accumulate somewhere above it?
We don't know...but we're interested.
Codex bundles LibreOffice — Simon Willison
5. How much AI can your computer actually run?
One of our recurring themes is just enough AI (or Frugal AI as some call it).
If you can run a useful model locally, perhaps the question shouldn't always be "what's the biggest model I can access?" but "what's the smallest model that can do this job well?"
The trouble is that figuring out what will actually run on your laptop can be surprisingly annoying.
llmfit is a lovely little open-source tool that tries to answer exactly that. It inspects your CPU, RAM, GPU and available memory and recommends models that should realistically run on your hardware, scoring them for things like fit, speed, quality and context length. It can also benchmark real performance and feed those measurements back into the project.
llmfit on GitHub
Random shot: cooperative email built on ATProto
And, having just told you a little bit about ATProto, well the thing about protocols is they can be used for more than just one thing like Bluesky and so
Here's comail.
It's an experiment in cooperative email infrastructure built around ATProto identity. Members use their own domains but share an SMTP relay and its sending reputation, rather than that reputation living entirely inside a commercial email provider. The project is open source and ultimately wants multiple cooperative relays to be able to work together.
This is early and niche and quite possibly not something you're going to deploy on Monday morning.
Have a look at comail
That's all for now. Thanks for reading.
Doug & Tom :)