I'm guessing diminishing gains are setting in on training frontier models and the capital isn't there to chase them. Even if OpenAI/Anthropic and Chinese companies said they would slow research nothing can stop the NSA and Chinese intelligence services from continuing on in the dark.
Is there a plausible scenario where having 2 weeks of water does anything more than having a weeks worth? I feel like after a week you're either getting evacuated, the government is delivering supplies, or large parts of the country are not functioning.
Sat in CERT and emergency coordination meetings in the Pacific Northwest. If the > 9.0 earthquake happens that is predicted, it will take a week to get temporary shelters working. All supplies will get flown to the few large airports east of the cascades (Redmond OR, Tri-Cities WA, Yakima, WA, etc). Because all the small bridges on moutain passes will collapse (creeks, small rivers, etc) all supplies will be helicoptered over the mountains until they can get Army Corps of Engineers teams in to rebuild runways.
They suggested having at LEAST a weeks worth of supplies, and more if you don't want to be waiting in a very impatient line with a bunch of scared, hungry people that may be desperate.
The other data is harder to quantify, but kind of insane. Like how long DHS estimates it will take to rebuild I-5 (minus major bridges) as a passable road, along with some key railroads, and basic power.
This annoyed me to the point that I created a python script that could search opencode's database and then a skill that uses the script to find the session I was thinking of. It's nice to be able to say "I think 4 days ago I had an issue with a dropdown in Qt6, find that session." Saves a lot of time and tokens.
I've been using the "experimental teams" in claude code forever and its great for this. I never start a claude code session manually, I just ask the main agent to open a new tmux window for me. The lead agent knows all the agents I've ever interacted with and built its own system to keep track of things between compactions.
Not in my experience. Tasks that would normally cost $0.08 on DSV4-Flash have cost me $0.30+ on GLM-5.3-Flash. These costs are after Deepseek's recent increase. Also GLM-5.3-Flash is so slow compared to DSV4-Flash. I would be fine with GLM-5.3-Flash if it was cheaper and at the same speed as DSV4.
I use DSV4-Flash on Max through Deepseek's API. I have been using GLM-5.3-Flash on High through Openrouter which I thought had a 50% discount. I must be doing something wrong for the costs to be off this much.
If I'm really studying a book I'll have an office supply store cut the spine out and make the book spiral bound. It's pretty cheap to do it too, like $10 or something.
It reads like marketing speak is the problem. Why is it written in such a urgent, compelling manner with those short punchy sentences? It may not be AI, but no one writes like that naturally. Maybe the author picked it up from reading a bunch of AI writing? The repetition of using you as a subject is also a red flag.
A camera in 2026 is not the camera of 1996. Cameras today are connected to active and passive monitoring systems which are much more invasive to privacy. What will a camera be able to do in 2036?
One of the things that is really interesting to me is when people don't update their opinions of some thing when the level of difficulty and cost associated with that thing change. A few decades ago, if a police officer wanted to find the whereabouts of some individual, they would have to do some serious leg work, even with ALPRs, potentially watch many hours of videos, etc.
With Flock that can be reduced to a mere search query across many locales (maybe not even their locale!) with effectively no effort spent.
When the cost changes so dramatically it effectively changes the balance of power and that is something we should, at the very least, deeply consider.
> A few decades ago, if a police officer wanted to find the whereabouts of some individual, they would have to do some serious leg work, even with ALPRs, potentially watch many hours of videos, etc.
The whole point about ALPRs is that the license plate reading is automatic. Even during the 90s, all a police officer had to do was query a database.
In the US, up until the mid to late 2000s, the majority of ALPRs were mounted on the police cruisers. Even when they moved on to installing them in fixed locations, they had two major limitations:
• There were few of them
• The databases were managed by the jurisdiction
To my exact point, by increasing the number of readers and SaaS-ifying the process to enable jurisdictions to opt-in to information sharing, you have meaningfully changed what the technology is capable of and how it can be used.
Motorola built the first national ALPR database in the mid 2000s. Again, what Flock is doing is decades old. It doesn't even look like you dispute this fact.
> Again, what Flock is doing is decades old. It doesn't even look like you dispute this fact.
Scale and level of effort involved matter. How many cameras were reporting to the Motorola database in 2000? 2010? Flock today?
How easy was it to type in “white Chevrolet Tahoe at [intersection]” until you see the one you want, get its license plate and run through the rest of the database?
Who was able to do that? How easy was it?
The sum of many small changes can create a meaningfully different result that we need to evaluate different than the thing that came before.
It is, and always was as straightforward as typing in the license plate number. Do you think indexing a bunch of 7vcharacter strings was beyond the capability of databases in the 2000s?
reply