Group creators can disable history sync if they wish, and when you join a group someone else created, you can see whether it’s enabled or not on the group details screen. I have it disabled in a bunch of my small, more intimate, groups for exactly this reason: so that nobody has to worry about someone else adding a less-trusted person to the group, who automatically gets a copy of history. But in larger groups that already include people who don’t know each other in real life, history sync makes threaded chat much more useful for new group members. And even in the absence of history sync, most people will already rationally treat these large groups as semi-public, knowing that any of the members you don’t already know and trust could easily leak what you’ve said to an audience outside the group.
We're very transparent in the UI about the fact that messages that mention @AI are not covered by e2ee. See https://imgur.com/a/5kPEdQP
Only messages that actually mention @AI are sent in plaintext to us, and then to OpenAI's API. All content shared between people without invoking the AI is fully end-to-end encrypted. Our servers don't have access to any users' private keys -- those are generated on each client and never transmitted, except in the case when a user logs in on secondary device and uses our QR-code transfer flow (in which case the sensitive information is communicated peer-to-per in that QR code). We'll share much more on this when we publish our tech blog in the coming weeks.
Regarding privacy around strangers joining a group -- Wavelength supports locking down groups to the degree the group creator chooses. You can disable history syncing, so that new members don't get a copy of prior history (which, when enabled, is reencrypted and sent by prior members of the group) when they join. And groups only grow to a size where there would be people you don't know in them, if someone chose to share an invite link semi-publicly, or allow all members to invite new members (which is also a setting the creator can toggle). Personally, I'm in a number of quite large groups including people I don't know in real life, as well as plenty of groups with 2-10 friends I know in real life, where I obviously trust everybody not to screenshot/share what I'm writing, and I appreciate having strong e2ee.
Yes. Each time you mention @AI in a thread, all prior mentions of @AI in the same thread, as well as its prior responses (up to the token limit for GPT-3.5) are resubmitted to the OpenAI API, so that it can continue the conversation including prior context.
The default setting when you join a group is that threads are not muted by default, but you can change a per-group setting to instead only be notified if you choose to reply, or explicitly subscribe to, each thread.
Hi, cofounder here. We're very transparent in the UI about the fact that messages that mention @AI are not covered by e2ee. See https://imgur.com/a/5kPEdQP
Only messages that actually mention @AI are sent in plaintext to us, and then to OpenAI's API. All content shared between people without invoking the AI is fully end-to-end encrypted.
How about you provide a way to generate an additional URL which would be associated with your own anonymous account rather than with an individual game? When this other URL is opened in a different browser (perhaps on a different device) you could associate that new session to the same anonymous account.
This way, if I know I'm about to terminate my browser session, or if I just want to move to a different device, I could do so without making an account.
The only problem is how to present this in the interface without confusing users and generating unnecessary clutter. I would probably put it near account creation options as an alternative.
I don't argue that AT&T's more recent decisions to stop offering unlimited plans were not both reasonable and rational. I'm just saying that, in light of Stephenson's comment that AT&T is happy with the iPhone deal for turning its business into a data-driven one, the unlimited plans may have accelerated that change. So if that was a serious goal for AT&T, maybe they shouldn't regret their initial offering of unlimited data, even if it makes much more sense now not to offer.
I would argue that it was the existence of unlimited data by default that "pushed the phone industry into a data-driven model," which Stephenson acknowledges to be a good thing. Without unlimted data, especially in the early days, you would have had customers fretting over every MB rather than engaging with the full app ecosystem and freely exploring the capabilities of the new technology.
Something tells me that NASA had a team devoted to watching Luna 15 as it orbited and watching out for funny business. Imagine the disaster that could have occurred if the Soviets had decided to use the probe to interfere with Apollo 11.
Of course, judging from the fates of later Luna missions, I doubt they could have pulled that off anyway.
Maybe you're right, but I think there's something to be said for teaching applied ML separate from the theory. I took a very in-depth theory-focused ML class at Caltech, in which lectures made up a thoroughly rigorous mathematical introduction, and the problem sets were only maybe 10% applications. I and my classmates came out of the class feeling super comfortable with the theory behind learning in general and how it applied to many of the standard ML algorithms, but without any experience actually working with very large data sets and building ML algorithm implementations that scale. That's what I'm hoping this book will help with, and I think it is appropriate to separate that kind of material from the theory.
Maybe financial industry salaries would change these rankings if they considered averages rather than medians, but a reasonable number of Caltech grads end up on Wall Street as well.
While this is definitely changing, Caltech graduates have tended to be heavily focused on academia in the past, and in that particular regard, the data definitely show success later on in life. The percentage of Caltech undergrads who earn PhD's is well above everyone else: http://www.nsf.gov/statistics/infbrief/nsf08311/?govDel=USNS...
In terms of other measures of success, Caltech might lag behind some other schools. As a current CS undergrad, I only know of a handful of startups run by Caltech grads. As I said, this is changing, but it goes to show the impact of choice of metrics.
Exactly. I actually think that Timeline is an improvement for user privacy, as it makes it easier for users to see exactly what information about them is accessible, and they can then decide if they want to remove it or strengthen privacy controls.
In the past, that ex-girlfriend or boss might have taken the time and effort to dig through content to find out about a drunken college night, but you would probably have completely forgotten it existed. Timeline offers you a way to double check old content comparatively quickly and easily.
The title is a bit misleading. One other lab's failure to reproduce the results does not prove invalidity of the original publication. It certainly raises some eyebrows, but it makes sense to wait and see what comes of Wolfe-Simon's further analysis before we declare that arsenic life definitely does not exist.