Wow, thanks for the link. How prescient was that writer - even the parts about using GPUs weren't too far off the mark! A depressing vision of where we're headed if something isn't done about it.
My mental model is to treat AI agents running on your system as a form of malware that is running in a honeypot you control. You don't want to just get rid of it as you want to observe its behaviour (in the case of malware) or hopefully do something useful (AI). But you certainly shouldn't assume it won't do anything bad to your system.
I remember deterministic/stable multi threading was an active area in systems research a while back. As I recall it never saw much adoption, most likely due to performance. Haven't watched the full video but will be interesting to see their views on that.
I think the next big thing will be brain computer interfaces - effectively everyone has always on access to a frontier model just by thinking. This raises some interesting questions in terms of what actually constitutes our identity - will our 'wet brain' component always be in charge of what/when information is allowed to influence our thinking? How can we avoid becoming overwhelmed with effectively always on access to super-intelligence?
I would probably recommend the art of multiprocessor programming (herlihy and shavitz) as a good starting point for concurrent programming. There is also " A primer on memory consistency and cache coherence" (Nagarajan et al) if you want to get more into the interaction between memory consistency and coherence.
Not the first time Microsoft employees invented a formal methods tool, Microsoft ignored them, and AWS scooped them up. Not the second or third time, either.
I don't think it's entirely fair to say they ignored them. As far as I know they are still employed at Microsoft Research and working on developing and improving the tool?
Sorry not an AI specialist, what exactly do you mean by 'custom harness engineering'? Some way of informing the model whether the output it is producing is good or bad based on the specific task in question?
Harnesses are the control surface the model interacts with. How it invokes tools, the tools it has access to, how agents are coordinated. It's like an interface or a shell. It's the magic that lets an LLM operate a computer. You can read more here: https://learn.microsoft.com/en-us/agent-framework/concepts/h...
> Some way of informing the model whether the output it is producing is good or bad
That's what post-training is for. At its most basic, you're giving it examples of inputs and outputs and then doing reinforcement learning to calibrate its adaptation to your examples. You need much less to fine tune a model than you do to pretrain the weights. You can build a really excellent LoRA for a stable diffusion model, for example, with 50 high quality images. LLMs are "a bit" more complicated and costly to fine tune, and you have to be mindful of the agentic loop, but the principle is the same. There's more to it than just LoRAs. Steering vectors, projection layers, custom encoders, etc. There's a fair amount to learn, but it sounds a lot scarier than it is.
Here's something to chew on: chain-of-thought doesn't exist until after pretraining! It's basically created by having ... blocks directly in some example outputs, and this is fine-tuned into stability. It's literally not much more than a parlor trick and some careful calibration. A powerful parlor trick to be sure, though.
I'm not chastising them for anything. I'm explaining my view on the rationale of their leadership.
That being said, if you're talking about the 1953 coup, ask the British about that. Also, it doesn't take away from the fact that the 1979 revolution didn't install any sort of representative government, either.
reply