The open internet is now a cesspit, with very little new good data. Expect everyone to train on user data always. They just got clever about whitening it.
The US just added a colony in Venezuela in cooperation with their local Reichskommissar and is looking to secure its oil. American perception of their country is quite something.
The systems jamming in Hormuz and Ukraine aren't exactly cheap and low powered, nor are those used by the US military in one of their many jamming exercises every year. Cheap GPS jammers fit in a cigarette lighter and are used by truckers who don't want their boss snooping on them, but they don't quite threaten aviation.
That aside, you don't need to give up on GPS just yet. As expensive as airplanes are, there is no reason not to have at least 8 element CRPA receivers in them. Mostly a military niche product in the past they are rapidly proliferating, and most fuselages make for a great ground plane.
> Cheap GPS jammers fit in a cigarette lighter and are used by truckers who don't want their boss snooping on them, but they don't quite threaten aviation.
They absolutely do threaten air and maritime navigation, because those trucks drive in and near ports and airports.
I don't think this is something anyone who has ever read a "thinking trace" would unironically say. Not that you can even see them in Claude.
When thinking first started and you would still see the whole "thinking process", I thought it was a ploy to 10x token use because it was just the most inane bullshit. "But wait, the user is asking me to" in loops.
You can see thinking summary transcripts in Claude Code and Desktop, and they are actually useful because they don't have those sort of thinking loops from the raw tokens.
Exactly, looking at a few of these traces is enough to realise that it's a waste of time to read them. A friend once described it as like reading a fever dream, which seems fitting. They're a necessary crutch for how the tools work, but probably should be considered an internal state representation that only sometimes accidentally seems to make sense.
Wasn't there a study recently that even found a model's performance was sometimes better when the "reasoning" was nonsense? As in, no clear correllation between what the reasoning says in a human's interpretation, and how the model actually did with the task.
I sometimes read the traces out of boredom or morbid curiosity. My favourite bit with Claude is how, even if you give a very comprehensive prompt in complete sentences, almost every trace will contain a variation of "the user asks me to X, but their thought cuts off mid-sentence."
It's crazy how much they have enshittified credit cards. It's not stuff like CVC, that's fine, but all the random failures! Nothing will turn me off your "payment system" more than having it randomly fail on me for utterly incomprehensible reasons when I'm simply trying to pay. It's five times worse when you are traveling, buying some small items in a shop and have it fail at the register.
I honestly don't know the last time I've had a credit card transaction fail, including internationally, unless I've fat-fingered some number. In fact, in my experience, random fraud failures are significantly less common than they used to be.
Less common but they still happen and except for a few banks that have modernized the failures are still much more cumbersome to deal with than necessary.
Worse I had stripe trip on me the other day which resulted in a "card declined" message but when I logged into the bank to authorize there was nothing there. Because apparently stripe decided not to even notify the bank of the suspected fraud. Thus there was seemingly no way for me to correct the error.
I can't even use tap to pay without either a physical chip from the bank or proprietary software. We've had ssh keys and oauth since ~forever there's no excuse for any of this.
You can point AI at any AI produced code and ask it to review it, get back 10 bullet points and a few pages of prose. And the fun part is, you can do that over and over and over again!
This happens all the time. Yesterday, I ran into an especially egregious case.
I had Fable add a new subcommand to our internal CLI tool. I reviewed and tested it locally and had to suggest several fixes that I feel like I wouldn't have had to tell a human senior engineer to do. When it finally submitted the PR, I had it on a loop waiting a few minutes for comments on the PR, then assessing/addressing/replying-to/resolving them, and then repeating again until all AI reviewers were okay with it. It ended up going through dozens of revisions and ended up with 160 comments left on the PR.
I think these "Flash" models are sort of an evolutionary dead end. Sure, there are some routine tasks and applications where they can be used. But for the actual novel development work? It's much better to run a big model at high power for 30 mins than watch the Flash model struggle for 2 hours and produce massive churn.
Same reason your phone has a few big CPU cores for real work, it's much better to "race to idle" than have an "efficient" core struggle. Shitty experience, shitty power efficiency.
It depends how you use the models. These small models work great for developers who prefer to stay more in the loop, and only task the model with things that can really only be interpreted in one way.
Not to mention, they’re great for self-hosting and getting yourself to not be dependent on some API that can go down or be altered at any time.
Big models seem to mostly be good for pushing ahead the frontier - the smaller models tend to gain the frontier’s capabilities after only a handful of months anyway. Many are perfectly content remaining a few months behind the bleeding edge.
If you have good feedback signals, like tests/benchmarks/etc, then it is potentially better to do multiple turns where model uses that to adjust code. Which might not need as smart a model.
The juniors I have seen don't even try to understand anymore what they are building. They don't understand the product, they don't understand the technology. So everything AI is building for them is nonsense. "Reviewing" it is impossible. Some of them think they are being extra clever having Claude process review comments for them. Even if you hire one, this way they are out the door within the month.
reply