I just wanted to let you know that you have successfully driven me off of
Not just you, obviously, the site has been in steep decline for a long time, but you were such a huge asshole that you managed to be the last straw. So, congrats for that. I won't be back.
Yes, limited local AI models certainly are going to be a standard feature in the future.
I run both local and cloud AI. There is still a huge difference, because my local machine has to use smaller and quantized models. It takes considerably longer for lower quality results.
what's the point?
The point is to be able to use your hardware in the manner you deem appropriate, and to ensure it is not doing sneaky shit behind your back.
If, instead of a Playstation, we were talking about rooting LG televisions to run custom Linux builds stripped of pervasive privacy-violating code and "apps," I suspect your opinion might be rather different. (And no, when the entire industry has turned to the Dark Side, voting with your wallet doesn't work.)
...Modern ejector seats cause spinal injuries from which full recovery is "unusual". Smashing the spines of pilots might not be the most effective way to boost morale.
$22,000 per hour. The F35 will be nearer $42,000 per hour. These are under-estimates, as they do not include price hikes, crashes that close businesses, crashing into businesses, crashing into heavily-populated sports centres, etc.
My late wife did undergraduate research in photosystem II where there was quantum effects back in 2002. She did her PhD thesis on proton coupled electron transfer (PCET) in soybean lipoxogenase, then had a long career developing models for Fortran code that did quantum simulation of proteins (she was behind Tagdock in Amber MD). This is nothing new; several biological systems have quantum effects that are central to their mechanism.
local LLMs aren't there yet and probably won't for the forseable future for ordinary people. If you want to run a full model like DeepSeek v4.1 Flash, you need something like half a TB of RAM.
The open-source LLMs are a different threat: They enable a service industry independent of the large players. I could today get some venture capital, rent a few racks in a hosting center, install a bunch of beefy machines and offer the same service as Claude or ChatGPT, just on an open-source model. The differences in benchmarks are becoming irrelevant for 95% of common use cases. And if I can offer that service for $19/m while OpenAI wants $29/m then OpenAI has a problem.
And IMHO this is the chinese strategy: Undermine the US AI industry by breaking its business model.
Has everyone gone insane or are we witnessing a major PsyOps ?
Since AI isn't magical, for it to be able to "kill all humans", it would have to be able to acquire the means to do so. Which means that these already exist. It means someone is able to do it right now. And we're chill with that, but worried about AI?
No doubt a bunch of autonomous weapon systems linked into a C&C could cause considerable bloodshed. Given that most of the existing weapon systems of that kind are essentially suicide drones, it'll still be limited damage.
Human-killing virus? Still need a means of creating and delivering it.
Nuclear war? We have humans in the loop.
Drive us all insane with AI-slop TikToks? A few percent of humanity still don't use that.
So what is the actual scenario ? When someone shouts "AI will kill us all", why don't we ask them "how, exactly" ?
This is kind of like saying that car wrecks wouldn't happen if we had training and standards that excluded 95% of drivers
More like 99.99%. Maybe need some more nines in there. Large C codebases without regular critical CVEs are basically nonexistent.
Sam et al warn us to take actions that will save them from bankruptcy
What bankruptcy-prevention actions are they asking for?
15-25 kWh drawn by the utility is quite a bit! I hope they're paying him well for that.
Monopolist wants the competition to slow down while they speed up in secret.
They're bringing outside scrutiny into their processes; that's going to slow them down, too.
A few days ago there was a story about how an AI store was failing horribly. It can't do that, but at the same time will destroy us?
Not "at the same time". You're conflating what AI can do now with what it will likely be able to do in a year or five, or 20.
Amodei and others are worried about what AI that doesn't yet exist may do, not about what current-generation AI can do. AI is getting better, very quickly, and unless there's some blocking obstacle that no one has been able to identify will at some point become far, far smarter than humans, individually or collectively. So before that happens, we need to figure out how to make sure that superintelligent AI wants to help us, and wants to do it in the right way.
Considerably more difficult than you imagine. Between them, they have almost 70 command-line options or flags. Taking into account that many of those behave differently depending on the presence or absence of other flags, there's a couple hundred different code paths. Then you need to handle directories, files, sparse files, sym- and hard-links, on and across different filesystems, not to mention the various file-like things scattered around (pipes, streams, etc.)
It's not rocket science, but neither is it trivial.
Stinginess with privileges is kindness in disguise. -- Guide to VAX/VMS Security, Sep. 1984
