So the Swiss are at it again…
Getting ready to release an open-source ‘ethical LLM’.
It comes in two flavors:
• 8B parameters (snack-size)
• 70B parameters (a heavyweight contender among open models)
What makes it different? It’s being built in compliance with:
• Swiss data protection and copyright laws
• Transparency obligations under the EU AI Act
• A fully carbon-neutral training process
And in a related study, they found that respecting opt-outs during web crawling led to no meaningful drop in performance. Imagine that: doing the right thing and still getting good results.
In speaking with Team Bubbly and a few other agency folks, and the same issue keeps coming up:
“We can spin up an MVP with AI, but we don’t know what’s legally safe to use. Our clients don’t want to risk it.”
This could make it the go-to AI for businesses not willing to wade into the murky depths of other Studio Ghibli, Disney Princess-trained models that could bite them in the ass later.
AI’s biggest problem has always been the wielder.
t’s a massively transformative tech that is hiding behind more and more paywalls. The lusty push to replace every worker with AI and squeeze out the artists is a problem with whose hands are on the levers.
(See: Microsoft laying off 9,000 and then Outlook going down for 400 million users.)
You can blame the robots or blame priorities.
But you can’t fumble email.
If this model delivers, it might be the shot in the arm the space needs:
Not just open source.
Not just performant.
But ethical, accessible, and people-first.
When this LLM drops in a couple months I’ll be interested to see how it is adopted.
AI for the people?
