I'm slightly miffed that the pictured computer is technically an Apple IIc (though I'm sure they meant the Apple II family). Mainly because there actually was an Apple II computer. I see the original article got this detail correct.
A bit pedantic, I realize, but the original Apple II had a max of 48k of RAM, while the IIc had significantly more ram at 128k. Software is (AFAIK) otherwise compatible between them.
I was surprised when (finally) trying out Claude how much I preferred Gemini's way of communicating. I wont argue Claude is better at coding, but for knowledge work, I had to dig through Claude output to find what I actually wanted. At times, it even felt borderline incomprehensible.
Just today I had Sonnet 5 generate this (asking about always-on display in the iPhone e-versions):
> This mirrors how Apple has always segmented Pro vs. non-Pro iPhones: base models got LTPS panels while Pro models got LTPO, and only with the mainline iPhone 17/17 Plus did that gap close the standard versions previously lacked the smoother 120Hz ProMotion technology and the always-on display feature, unlike the Pro models — the 17e is the one model line still using the older, cheaper panel.
(emphasis mine)
I mean, I can guess what it is trying to say, but who RL'd this nonsense?
Fable 5.1 even today inundates prose with "it's not this, it's that" type of garbage. I had to rewrite two paragraphs from a generic class announcement which I was planning to post on the LMS. Not sure where that productivity gain is that everyone is talking about.
I couldn't even stand 4.6 by the end. 4.5 was very usable, 4.6 started alright and somehow got more annoying. 4.7 made me quit my subscription and ditch their services entirely - it's an honourary member of my quite short "Coworkers I'd like to throttle if I didn't work remotely" list. Infuriating to instruct or communicate with.
Flock records public events. None of these apply. There is no reasonable expectation of privacy in a public space and the courts have been quite clear on this. Actual legislation would need to be passed to restrict what can be recorded in public settings. A redefinition of "public", basically. And what about the public? Would they also be barred from making their own recordings of public events? Would they be barred from sharing them with law enforcement?
Not necessarily. One could argue my cell phone's signal is in the public when I use it (home or otherwise; it is rather omnidirectional), and the records thereof aren't subject to it the same way as say a search of my phone or my home.
However, in Carpenter v. United States, government entities were found to be in violation of the fourth when accessing historical CSLI records containing the physical locations of cellphones without a search warrant (at least over a period of 7 days or more, if I understood it right). Previously, it was considered fair game under the basis of records technically turned over to a 3rd party.
Then there's United States v. Jones (2012), where it was established a GPS tracking device was attached to a car constitutes a search.
It remains to be seen how Flock will ultimately hold up in court. Although a few cases have popped up, I think it's too early to tell.
The police officer is not standing on the street corner, observing public events. They are searching a privately-held database of sensitive information whose use is governed by certain restrictions, which are being violated by the police officer. Whether or not those violations rise to the level of crime does not erase the fact that they unethical, and likely against both Flock's terms of service and the department's policy.
I, uh, would be more careful of making that statement given our public comment history. I know I can't cast any stones here.
More than interesting though, I will vouch for many of the points they presented in practice. Spreading partitions and taking advantage of the flexibility of Linux with how and where they are mounted has been a part of best practices for decades now. Many cloud-based solutions for Linux workstations I've used take full advantage of this for easier management of space (i.e. volume expansion) as well as security considerations, to say nothing of keeping your /var/log separate to avoid running out of space because you turned on verbose mode for something.
Part of the reason Meta is scrambling for purpose is how much privacy laws and changes have hurt them while their platforms fail to attract new generations. Ads are still selling, but they can no longer tie them to actual sales. Users are still logging in, but aging (i.e. failures to capture the younger generation).
Personal AGI has massive hurdles to overcome to be anywhere near the levels of profitability their ailing ads network is. This will bring in new data for them to mine for sure. But it will basically have to outrun new legislature that will likely limit just how "personal" these systems can get as they seek the profit they need to sustain growth via said data.
I mean, sentences themselves are graphs of data and context with connectors of various types. Embeddings within these models help form the links to the underlying semantic concepts.
Directly representing things as a graph would likely reduce some of the 'translation overhead', done correctly.
I'm not so sure on the latter -- that introduces a bunch of tool calls, while the text patches are much closer to the semantic space imo and easy for harnesses
If we rephrase this as instruction following alignment, what is the concern here in practice vs a skill? (Which a model eventually internalizes)
It doesn't detract from the research - as a paper, it shows more crisply the structure is useful. I'm just not sure how necessary the encoding, and given durable tasks, novel the insight. Is there new alpha here somewhere, esp given the similarity?
I am but an armchair expert here, but I do think most of what's missing is the complexity of real-life ecosystems and "connections". Having biomes that are substantially large on a given planet and diversity among them, animals unique to each, and some kind of connections between the sparse settlements (even just like 3-5 actual cities per planet would be a start) would be a big difference. It would be a bit less about the 'wonder' and more about understanding the interplay.
For now, all I can see is "optimization" and "system limits on numbers of unique objects for things to work at 60+ FPS".
I assume you mean modern cities? NMS has had settlements for a long time now. What kind of computer would be required to even generate and render 18 billion planets with 3-5 actual cities per planet? Most open world/sandbox games are challenged just to have several large cities in a game.
> even just like 3-5 actual cities per planet would be a start
This often takes me out of immersion in reading sci fi books. Every planet has one culture, one ruler, one major city. I know it's usually just a convenience but has the author never seen earth? Why would future human colonized planets not have hundreds of thousands of cities and competing factions?
Finding only one village on each planet is kind of laughable in the same way. The stage of human evolution where there was one village is such a tiny slice of time (if it ever even existed at all).
A bit pedantic, I realize, but the original Apple II had a max of 48k of RAM, while the IIc had significantly more ram at 128k. Software is (AFAIK) otherwise compatible between them.
reply