bearjaws
Recent example of a breach, they are now using AI to decomp APKs and iOS apps, mainly old builds of apps that nobody would bother spending the time on previously to look for secret keys or accidental information disclosure.
This would have been immensely time consuming with each app usually doing things a bit differently (e.g. encrypted + compiled settings, keys in database, settings in a settings.json).
Now you can automate the entire pipeline, this is actively being used to look for legacy API endpoints that may not be hardened or monitored, or if you don't rotate your keys and at one point disclosed them.
I had a Dell OLED 4k laptop 4 years ago that depending on screen brightness would absolutely murder the battery.
Now, just 5 years later I squarely believe nobody should be touching this stuff, it's basically worse than Blackjack or Craps in every way.
Never thought I would see myself aligning with the more "Puritan/Christian" values.
Plan mode tends not to ask as many questions and go as deep, so often edge cases and other details get missed.
The absurdity of every week another AI feature, another way for data to leak to their services, another crash report.
They are exhausting and abusing their positions of power.
For example Duke is for profit here in Florida, and charges about 25% more per KWH than OUC that is a public utility.
This leads to pricing where on one street your power bill is $250, but one street over its $200, often coming from the exact same power plant.
That being said, they are definitely even less reliable now due to demand.
I feel bad for the team, but will be great growth opportunities for all.
When Claude added it, I immediately hated it since you click into the terminal and accidentally click something.
They even negotiate with all three aggressively.
So not really.
"Friend-slop" has already proven wildly successful, and significantly lower effort than AAA games.
Combine this with the fact your game will be stolen and put into Roblox, it really feels like we're approaching a era of just rapidly produced games that live for weeks rather than years.
They already have BOTW torrents that are pre-setup and as easy as it gets.
Someone at work recorded a demo of a new feature using Claude + https://github.com/veedstudio/open-edit
It was pretty slick and he said he only did two passes on it. Not sure what it means for Adobe, but hopefully it lights a fire under them.
Extremely needy, demanding, and simultaneously spineless.
Riot releases a $500 skin and make a million dollars, GTA6 has a $400 edition that doesn't include the game, and of course the OG horse armor.
Outrage after outrage, yet they keep on buying.
Additionally, 99% of them have no clue how hard it is to build a game, at any scale really. They don't understand where corners have been cut, and that it's possible to use AI without cutting corners.
They quite literally demand the world despite having never created a single thing themselves, blindly consuming, never producing.
Personally, it makes no sense to build a game without AI. A lot of code is needed and you need to iterate on it, perfect use case for AI.
Art, buttons, sprites, and models, often need rounds of iteration, why not start with a basic set and work on it?
There's nothing wrong with iterating, there is plenty wrong with slop (like the 6 fingers in the trailer). So let's be mad at the slop, not the AI itself.
Honestly still enjoy writing automation today because of it, I've gotten to use a lot of the skills I learned from OG RS at work.
For whatever reason people have clung onto this idea that the "Gotcha" to slow down AI is to worry about water consumption, and it's an idiotic gotcha.
Reality is if a regulation is too broad it tends to get overwhelmed and dies when it requires 100 people to enforce.
Even a basic in house data center can push 100kw with a few servers.
Every time a new model comes out, people come out in droves "oh I don't notice anything different".
People have been saying this about <currentModel-1> for 2 years now, and the entire state of AI has changed dramatically.
It cannot be that the next AI model isn't better, but also suddenly what they are capable of is on an entirely different level.
AIs are very good at writing tools that convince other AI that your resume is the best, when in fact its complete bullshit.
I do agree with his idea overall, recently at work we've had QA (as in pharmacy QA) wanting to use Claude to help automate some pipelines between PowerBI and Sharepoint. Was cool to see people who are not very technical pick up Claude Code. Of course agent safety is the big problem really..
How many observation drones can you build for the cost of this program? 10,000? 100,000?
What is the point of this in modern warfare?
I have it retry on failure, but you should do that with any LLM really.