That's one of the reasons to keep it as far as possible from anywhere you need trust (like prod): software, like soylent green, is made of people. And you can't trust people like him.
Multiple teams and efforts are now making Qwen 3.8 27b work much faster on Mac and CUDA and the inference speed is doubling almost every week.
There's now ongoing competition about who will make it work faster with exactly the same byte output at https://mlx.fast . TensorFold is the project to follow.
All companies should at this point invest a lot of money into defensive AI. In ~July the AI hit crypto and bitcoin, where it's the easiest to extract the most money.
Now the AI powered attacks are hitting infra and SaaS. Next (~December), it will hit classic infrastructure, banks, small businesses, etc.
By mid next year, we will be over this hump with defensive AI being required part of everything.
I find it breathlessly funny how people think it's going to be AIs vs AIs fighting it out with light sabers and plasma lasers in the cyber space or some shit.
Your intuition that this can be balanced out is wrong. When you have a CVE in your TCP stack, your vault unlocks and you're fucked, it's done. My dude, there is never going to be AIs scanning every incoming packet for signs of a 0day, and deciding whether to, uhh, cut it up with a lightsaber and shoot it with a laser pistol. Your options are to have a stack so secure that it can't be targeted, or to disconnect all network cables.
I specifically meant red teaming with AI across your infrastructure to find vulnerabilities before these are exploited. This is what I have been actively doing and it works.
Its already happening to classic infrastructure, small and medium size businesses. Absolutely anyone can summon a gremlin capable of exploiting stuff like this or the PHP thing last week. There are clankers probing this stuff all over the place within a day of it going public.
People may have specific hardware issue that's blocking them and a new kernel release fixes the issue. In that situation they see this announcement and they may be technically brave enough to want to install and check it. Currently they have to find and go to something like https://kernel.ubuntu.com/~kernel-ppa/mainline/ , find the right versions, download multiple files and then `sudo dpkg -i linux-*.deb` and then confirm if their issue is fixed.
I'm not saying it's a great idea, but I'm explaining why people may expect to have a smoother path here.
Agreed that centralizing identity is critical (which implies strong reconciliation systems, that can both do well on merges/splits/identity-shifts, but also need to enable backtracing to sources for each triple), but then I don't think it makes sense anymore to force ontology across all systems in the pipeline.
As long as you give AI ability to do toolcalls/skills for each individual system and let it build up memory, then AI does much better job piecing together the actual practical ontology. You can give the observability AI to ask each system about it's state for the given piece of information and the AI should be free to make any toolcalls (e.g. call the specific system appropriately via CLI) it needs to gather the precise state.
I think human establishing the ontology gets in a way here, just let the AI poke the systems on its own via clean small interfaces.
What context did you get those numbers on?
I'm always surprised that in these reports people have like 8k context. How can you survive with 8k context, doing what?
reply