When I was there Google used multiple Sun/StorageTek tape libraries at each datacenter. The robots handled much of it, but there was still a lot of manual tape loading and unloading performed by technicians. Once tapes were full they were unloaded and hauled off to a secure third-party facility.
If anyone needs a Bluesky invite, there are three in the About section of my profile page here (note: you will need to prepend "bsky-social-" to the code).
I asked the Bluesky devs about this back in May (of 2023).
Me: "if the network is intended to be public, why are user profiles and posts currently hidden behind a login wall?"
Paul Frazee: "it was a kind of bad artifact of how we set things up initially (just trying to ship). once we realized it communicated the wrong idea it was too late, and we now need to spend a heavy bit of effort communicating before we spring it on everybody."
Oof. That is not a fantastic answer for them to give. I'm not even 100% sure what that means.
> Once we realized it communicated the wrong idea it was too late
For what? Too late to change the technical side of things? Is there a major technical barrier to having a public interface that matches the public firehose APIs? Because I can't figure out what that barrier would be.
What magical deadline or restriction was in place that would have prevented fixing an obvious barrier to the network?
And "once we realized it communicated the wrong idea"? People aren't misinterpreting the message, they're accurately assessing that Bluesky is not an Open network even though it is marketed as one and pretends to be one. This isn't a communication problem, it's not that blocking public access communicates the wrong idea to the public, it communicates correctly that the network isn't Open. It's complete nonsense to try and phrase a failure to fulfill the basic promises of the network as if it's actually just a PR problem.
----
> and we now need to spend a heavy bit of communicating before we spring it on everybody.
Communicating to whom? The users? Is this an admission that Bluesky users don't view the network as public or that they don't want the network to be public?
This is phrased like "we need to spend a bunch of time clarifying and explaining how this will all work before we pull the rug out from under people's feet" but who on earth would this be pulling the rug out from under? Who would be confused about this change? This isn't actually complicated; if a user without an account looks at a post it will either be visible or it won't be visible. That doesn't require a FAQ.
If the idea of that post being visible is contrary to community expectations and if the devs feel they literally can't make open decisions because the community would oppose those changes, then that's a pretty heckin big problem and it sounds like they should stop advertising that this is intended to be an open network or that federation is coming any day now, because it doesn't sound like the community is on board with that idea.
Or is it a communication problem for people outside the network? But how? What would that even mean?
Who outside of Bluesky would be confused if the devs took measures to fulfill the promises they've been publicly making since day 1 of the network? This isn't some complicated thing where people will be misinformed or they'll be confused by the idea that they can view without an account but can't post without one. That's how most networks work, locking viewing behind a login is the abnormal confusing decision to outsiders.
Ultimately, the network will be publicly viewable or it won't be. I do not understand what about that would require a PR campaign. Were people signing up for Bluesky thinking that the network was going to be permanently private? Because if so, that is something the owners should be horribly embarrassed about.
----
It's just a fundamentally weird statement. The only thing that a closed-down network communicates is that it's closed down and exclusive. The only reason that fixing the network to reflect their own marketing would be a problem is if the network doesn't want to reflect the marketing. In which case, they should stop pretending that this is a temporary limit on growth to help prevent out-of-control scaling.
The most charitable take I can have about the response is that it's corporate bullcrap from people trying to take a simple decision that was made for marketing reasons (or has accidentally been found to be extremely valuable for marketing) and to after-the-fact justify it as something complicated and difficult so that they have an excuse to avoid actually making the change.
The less charitable take I could have is that they're being honest, and they're unable to make changes to make the network more open because their userbase would be hostile to those changes or without a PR campaign would view opening up the network as if it was an attack -- and if that's the case, that sure as heck is not making me feel confident that this network has any potential at all as an Open platform. If the users aren't on board with Bluesky as a federated and Open network for everyone, then y'all don't have an Open network and it doesn't matter what your plans are.
And in either case, it's clearly not a temporary technical restriction to help with scaling so I don't know why devs are jumping into threads pretending that it is. It's clearly a deeper problem or else the devs wouldn't be giving you this kind of a nonsense response as soon as you tried to dig into it more.
I mean, you're not wrong, but you're also missing the point. We don't need "perfect recall" in this case. It's not difficult to get any of the ChatGPT models to divulge their knowledge cutoff date. It's also not hard to verify with a handful of crafted prompts.
No he means from the Console app on macOS. It's not the same as the Terminal app, which it sounds like you're referencing. It's a log viewer that's installed by default but tucked away in Applications/Utilities. It's similar to dmesg I guess.
You realize the Sunshade is science fiction, right?
Picking two examples from the Wikipedia article you linked, the Fresnel lens is a proposed object that would float in space between the Earth and Sun, approximately 1000 km wide and only a few mm thick. The article notes "at a science fiction convention in 2004, Benford estimated that it would cost about US$10 billion up front, and another $10 billion in supportive cost during its lifespan." Lol. This idea is pretty hilarious, just in general. A lens so wide as to span the entire US state of South Dakota. Please explain how you'll launch and/or assemble that at the L1 point. (This was a subplot of a Simpsons Episode by the way: https://en.wikipedia.org/wiki/Who_Shot_Mr._Burns%3F#Part_One)
Or there's the "Lightweight solution", "a distributed sunshade with a mass on the order of 100000 tons, composed of ultra-thin polymeric films and SiO2 nanotubes", which "estimates that launching such mass would require 399 yearly launches of a vehicle such as SpaceX Starship for 10 years." Also lol!
Yup. There was information about the NSA's Echelon program floating around for decades before Snowden. Check out the Public Disclosures section of this Wikipedia article:
Despite multiple whistleblowers dating back to 1972, no one took it seriously until the Snowden leaks confirmed it without any doubt. For forty years this information remained below the surface.
Eventually, maybe. In the keynote, there was a vague outline of a virtual keyboard, but (unless I missed it) we never saw that virtual keyboard in use. Instead, the demo pivoted to using a paired Magic Keyboard.
Would someone please explain like I'm five which components of LLM's like ChatGPT are still closed source? What are the specific technologies that OpenAI is holding on to?
I know a lot of LLM stuff has either been released or leaked out, but don't have enough expertise in this area to understand the competitive advantages or breakthroughs OpenAI has obtained.
As far as I understand, it's mostly the weights. If you only have the models, you're still gonna need to get a massive amount of training material, huge costs for training it and fine tuning the hyper params until it works.
