Check your idleCompaction line in /config to make sure it's off -- several times they've toggled new features on by default for me. This one isn't in the VS Code extension so I don't know if it shipped default-on.
Yes, it should absolutely change (both should be freely available), since ignorance of the law is not a defense, and case law is law (here, in the US, since we're a common law country). But the flip side of that is that the law has to be knowable, if ignorance is not a defense, or you're just going "lol too bad for you for not knowing the secret law".
Also, because how the judicial branch decides case law needs to be reviewable by the People, so that we can inform the various legislatures to adjust the law if the judicial system is not ruling to the will of the People.
Even without AI it should be. Honestly the less AI has to do with giving people access to our laws and court records the better since the value of having that access would decline with hallucinated laws/interpretations/attribution. A post about how AI isn't even fit to write wikipedia articles just fell off the front page a few hours ago, and providing something like this to the public seems at least as important.
There are benefits to specialization, and will be for a long time even with AI. It's worth a premium to not limit all my software to what I and my agents know, and to be able to rely on other people to fix it if there's a problem.
Investment size for software plays is going to drop, and people will still invest for the same reason they do now: solving problems for people is valuable. Marginal costs dropping will bring down a lot of costs in various places, but the software itself is only part of what SaaS provides.
> I think everyone has known that inference is profitable overall for the major providers for a while
A lot of AI haters like Ed Zitron and their readers have refused to update on this. I don't know if any of them have yet, but it'll take a year or two for the idea to fully filter out of their podcasts anyway.
Seems a bit of a disingenuous framing. Most of the arguments I see are some mix of
1. Their figures have been pretty dodgy or excluded training costs so why is it trustworthy this time?
2. API inference charged at a profitable level rather than heavily subsidised subscriptions means everyone tightens up their usage + governance Which we very much saw with the rise of openrouter et al.
Personally I think there's a lot of the bait and switch of 2 when people talk about profitability of the 10-20% of API usage for Anthropic while also wanting to lump in subsidised chat users as if they're profitable too.
I think we know if it was this clear cut then the behaviours of companies would be wildly different and more consistent with profitability.
There are thoughtful versions of these arguments, sure. There are also thoughtless versions, and it's not disingenuous to note it.
It's not actually clear that subscription users are unprofitable as a group (and the group level is the only important one).
We do know the companies are not currently profitable, because they're open and intentional about that, but that tells us nothing about their margins without infrastructure buildout and investment in training.
If anthropic were making money on subscriptions they wouldn’t be pushing people onto API + enterprise packages.
> We do know the companies are not currently profitable, because they're open and intentional about that
Why do they keep engaging in stunts like excluding infra costs then?
You are picking choosing to ignore evidence and actions when it suits you and then saying “oh we can’t possibly know” when we do absolutely know the problems of capital and debt raising on infrastructure problems (which are mostly not the Big Frontier labs either) there’s hard arithmetic for what’s needed to make these projects profitable and right now all the inputs are getting more expensive between chips and capital.
They were coined in the extremely influential foundational research the "pop science book" was based on, but that doesn't make it a marketing ploy; System 1 is very fast non-verbal subconscious heuristic operation, and a lot of our brain's classification takes place at that stage. Anyone familiar with the usage immediately had clear ideas about how it'd ideally be used (e.g. heuristics that reduce search space, manage context elements dynamically, etc) vs not (e.g. as a 1:1 replacement for something that could already be easily solved with a classifier).
Sure, researchers get paid when they either do the research under contract or productize the research and sell it themselves. But they don't usually do the latter because that's quite difficult.
I've seen them talk about it a bit on Twitter -- it seems to be fairly well-calibrated in general, but obviously you need to test it on your use case and dial it in comparison with known data for best results.
Contrived false positives and negatives could (and should!) always be possible. That doesn't tell us anything about the natural false positive and negative rates, and a lot of people who have actively tried to use voice instruction to get models to consistently fool the detector without iterating against it directly have failed.
reply