To proceed from that initial full page pop up it takes me to the App Store to download the app wherein I can sign up to get on the waitlist. There’s no apparent other way of doing it. Maybe if I signed into Bing on the webpage? 
Regardless, any tiny interest I had is gone now. I’ll wait for a version that respects my privacy. If ever.
Talk about waiting for unicorns…
Correct! And I use it only for Bing chat. 
fwiw I am using it on a PC, so that also might make a difference.
Maybe, although once trained these things can in principle run entirely offline, albeit without info on current events (would quickly get outdated). A company could sell a fully standalone “chatbox” that never connects to the internet. Kind of an appealing concept actually.
But what data is it feeding from for its answers?
Its training set. Whatever data they used to set up all the billions of model parameters is present in the model “as an echo”. Not everything is in there literally, but relations and phrases etc are in there implicitly.
I don’t think that’s enough for it to work offline, personally. It still uses searches to get most of the answers.
Just speaking of my own idea of privacy for this, it doesn’t have to be offline. It just has to allow me to use it without signing in and allow a browser that has ad blockers and tracking blockers etc.
Like any search engine.
The chat feature looks like the future of search. You should definitely give it a try.
Here you go. 
It doesn’t have to, if it doesn’t need to have date from after its training set. All answers can be generated without internet access, it’s all in there. That’s a big reason why these LLMs take so many gigabytes of data, they hold all the necessary information.
Bing Chat appears to be a kind of hybrid that can add current info, but the Facebook LLM that was “leaked” online can be installed and used offline.
Technically, yes. For the types of questions I have asked it though, it basically regurgitates information that I’ve found through various web searches myself from Medium, Stack Overflow, and the like. If it were to not use current searches, it wouldn’t be able to keep up with software development as a whole though. Things change far to fast in this industry.
Skipping down memory lane today I looked up ELIZA, which was created between 1964-66 and had a certain popularity after implementation on newfangled PCs in the 1980s.
“ELIZA’s creator, Weizenbaum, intended the program as a method to explore communication between humans and machines. He was surprised, and shocked, that individuals attributed human-like feelings to the computer program, including Weizenbaum’s secretary.”
Plus ça change, plus c’est la même chose.
I wonder how the “AI” enthusiast computer scientists of the time (‘60s) would feel about Chappy and how (not) far “AI” has progressed in nearly 60 years? I imagine they’d get depressed, get drunk, and possibly get a new career.
Many people are currently reporting seeing other people’s chats in ChatGPT. Again. Pretty worrisome that OpenAI can’t seem to keep chat caches separate. Not sure if related, but they also just had an outage related to capacity issues as reported on their status page.
This just in: @JoeS is obsessed with LLMs… 
Assuming this is real, this is a pretty great response from Bing/ChatGPT: (from Reddit)
Wow, look at this, we’re actually back on topic. Google’s “Bard” chat AI is “out”, aka available for waitlist.
I just want to know WHY Google associates their AI with The Bard???
Actually that association (with Shakespeare) is the only cool thing about the name. Maybe it was carefully chosen to sound friendly/dorky and not like a potentially evil AI.
I’d have preferred if they called it Bart, which conjures up an image of a good-natured but mischievous AI.
Also a comment on “answer quality” as well…
Bart would work for me. At least its answers would be funny.
