Orbit currently uses a version of Mistral LLM (Mistral 7B) that is locally hosted on Mozilla’s Google Cloud Platform instance.
Hmm.
>locally hosted
>Google Cloud
Hmmmmmmmmmmmmmmmm.
Sounds like they’re running their own LLM instance on googles cloud infrastructure vs using something like OpenAI via API.
As web dev parlance it makes sense but for marketing it is definitely confusing and they should do better.
Remember how the cloud is someone else’s server? Now you can buy it (or lease) and bring it home, and it becomes only sorta someone else’s.
Amazon and Azure offer their own on-prem products.
“Locally hosted” means it’s running on the local host. In this case, that would mean on the same computer running Firefox.
Calling something that is only accessible over the internet “locally hosted” is outrageous doublespeak.
Information We Share.
We use third parties to provide the Service to you, and have contracted with these companies requiring them to protect your information (Third-Party Services):
Google Cloud Platform. Google Cloud Platform (GCP) is a cloud-computing platform. We use GCP to manage services that facilitate responses to user prompts and page summarization.
https://orbitbymozilla.com/terms
4. Content
A. Content You Share
By using the Services, you represent that you will only share material (including Inputs) that you own and/or have the legal right to share and sublicense to others, including without limitation, content and data contained in any web-page shared through the Services to generate Outputs. When you submit your own content through the Services, you continue to own the rights to that content. You grant Mozilla a worldwide, royalty-free, perpetual, irrevocable, non-exclusive, transferable license to use, copy, modify, adapt, sub-license, prepare derivative works from, distribute, perform, and display the Inputs for the purpose of operating the Services.
Thanks for the link to the privacy policy. You notice, at the bottom, it has links to both “About Mozilla” and “About FakeSpot”?
When you run the Orbit extension, it connects to two domains with every request:
- orbitbymozilla.com
- prod.orbit-ml-front-api.fakespot.prod.webservices.mozgcp.net
There’s FakeSpot again.
And FakeSpot has a terrible privacy policy that allows sale of private data directly to advertisers.
I don’t want that. I want full control and absolute privacy. I do not want your AI reading my emails. Look at that summary, it’s as long as the whole email, and you’re not going to be able to trust that it picked up on the most important part of the email. This is not efficiency, this is novelty.
I won’t trust the AI Mozilla uses until they show us the source data. Not the source code that consumes a massive binary blob; the stuff that generated the binary blob they are using.
Not far enough. I won’t trust it until I can build it myself and self-host it. Then if they provide reproducible builds and hashes of the currently running build, I can decide whether it’s better to use their hosted version or my own.
I won’t. But my concern is that Mozilla is heading in the wrong direction lately, and I have used Firefox for a very long time.
We always told them we want things to be optional, and now this is an extension so I dunno. Seems they’re listening?
Pretty sure the email is longer than is shown, hence why the last sentence is cut off
Here’s the summary of their example article (or perhaps the page?):
This email expresses a sarcastic and exaggerated perspective on the advancements and implications of Artificial Intelligence (AI). The author begins by expressing excitement about the technological marvels of AI, but then proceeds to poke fun at the complexity and convoluted nature of AI, its ability to predict our actions, and the replacement of human interaction with AI chatbots. The author also mocks the idea of AI-generated content and its ability to replicate human creativity, and the potential ethical concerns of relying on AI for decision-making. The email concludes with a sarcastic call to embrace the “glory” of AI and its potential to take over human autonomy. The tone of the email is light-hearted and humorous, but it also raises valid concerns about the role and impact of AI on our lives.
This isn’t really a summary, there’s some interpretation going on as well. I don’t want AI to do any form of interpretation, but if it does so, it should be as metadata below the actual summary.
And honestly, I almost never get an email that I actually want to summarize. Most of them I can either completely ignore (corporate BS), or they’re short and to the point. So it’s weird to me that email is the first thing they mention.
I did exactly this at work the other day. Someone had forwarded a full email thread to me and asked my opinion on it - they gave no summary or outline of the thread and expected me to read through it. I don’t have time to read through a full thread and work out what they want from me so I copied it into chatgpt and asked it to summarise and tell me points that might need my attention. It was pretty good
AI you can trust
Lost me there
Easily summarize emails
Haha “Give us access to all your emails for data and corporate espionage we pwomise nothing bad will be done with it!”
I sent in a support ticket asking them to save Firefox and stop all this AI bullshit
