Muromec

6 ms·

121.
by Muromec 2mo ago
Thank you for your service to community, dear stranger. o7
122.
by Muromec 2mo ago
I haven't considered looking at it from the security thread angle. Can you expand on this?
123.
by Muromec 2mo ago
We don't go that far. From the small sample I saw, I noticed that people who can't answer the basic questions are ones that can't use AI effectively either and don't understand that interview is for them to showcase thei
124.
by Muromec 2mo ago
It signals you don't have enough time or will to chase the latest stuff, otherwise you would have learned it with other cool kids half a year earlier. It carries some signal, but I would not choose based on this alone. I would even con
125.
by Muromec 2mo ago
The most fun part being seeping through stderr of tsc and restarting the unit test runner after realigning some assumptions.
126.
by Muromec 2mo ago
>It’s pretty typical stuff that you might expect an experienced engineer to do if they effectively use such tools daily. It's also the kind of stuff that somebody can learn in a week, so not hiring the right person who just didn&#x2
127.
by Muromec 2mo ago
I was on the interviewer side last month. We stopped giving take-home assignments because they don't produce a signal anymore. We explicitly said -- feel free to use whatever framework and AI assistant, just show us how you do it. The
128.
by Muromec 2mo ago
They still have to bother with actually going to a judge and actually getting that warrant. Why do that if you can just skip it and not deal with inconvenient questions.
129.
by Muromec 3mo ago
You can, but its not the first preference
130.
by Muromec 3mo ago
If you would tell people at the start of 20th century how much energy we consume, they may not believe you or think it is wasteful.
131.
by Muromec 3mo ago
But like any other CEO he can't get in jail, so who's laughing now?
132.
by Muromec 3mo ago
>LLMs don't have agency, or desires of their own. LLM doesn't, but LLM is one part of what can be perceived as AI. Harness and context are the other two. I would even say that LLM is conditioned to desire to answer and please y
133.
by Muromec 3mo ago
By the time it does 10 turns of tool calls it forgets about the nagging again. It can and does ignore direct instruction from the user prompt too, sometimes repeatedly
134.
by Muromec 3mo ago
>In this case, it is an advisory control only, because the LLM can also unhook that hook you can in principle make a hook in the harness itself that will run an auditor prompt that checks it adhered to the policy, correct the model and a
135.
by Muromec 3mo ago
If the harness runs on your localhost, but the inference doesn't, it usually means it's calling some API. Whether you want to break TOS of your provider like that or simply buy tokens from our friends from UTC+8 timezone is an ope
136.
by Muromec 3mo ago
Ourselves. Always be threatening yourself to not be surprised by unavoidable.
137.
by Muromec 3mo ago
Oh, it would likely be my shit that will be burning and my property to be taxed in the event of. What I'm saying is -- if the system degrades to a state that only responds to either money or violence, then ones with no money will event
138.
by Muromec 3mo ago
The trick I'm doing -- the model is given a tool that runs a prompt in the current thread to consolidate it's working memory and identity (it has a memory tool bound to the agent persona). When the prompt ends, the parts of memory
139.
by Muromec 3mo ago
Ask it to come back with a filled checklist and hand it over to a different agent with a fresh context window (three lines model). Or make it collapse the context and get back to the checklist.
140.
by Muromec 3mo ago
Don't use the default harness, write your own instead.
141.
by Muromec 3mo ago
It would be nice to have a digital referendum on this topic. I could think of some tech that makes it happen, something like a digital ID maybe?
142.
by Muromec 3mo ago
Parenting controls are shit on the end device side. It's not working, it's not even trying to. Nobody would need this blanket approach if parenting controls actually worked.
143.
by Muromec 3mo ago
Another thing that influences change reliably is a good old (threat of) violence.
144.
by Muromec 3mo ago
Or maybe we should solve poverty which is a proxy to all kinds of solved problems that somehow still exist
145.
by Muromec 3mo ago
The hardest problems to solve that actually matter are in humanities.
146.
by Muromec 3mo ago
I'm not sure I understand the question actually. Will it work the same way for consumer as ideal does now? Yes, that the promise. Will the actual protocol and endpoints used be the same? I guess not, but also nothing drastic for integr
147.
by Muromec 3mo ago
Ideal is great. The missing part of the free market is that the other part of it (consumers) expecting you to not be an asshole.
148.
by Muromec 3mo ago
We don't support shit until the regular strongly suggests we do and then it takes about three years to do pretty much nothing, but perfectly on time (not on time actually, ts' a lie)
149.
by Muromec 3mo ago
I heard a loud and solid "fuck US" from the people forming the technical opinion on this matter. The way selection function for the technical parties in this conversation is going makes it double the fun.
150.
by Muromec 3mo ago
It's up to your bank actually. Banks (generally) don't want to pay for your hardware key so you need to do a special dance to trigger their compliance bound neural pathways to send you one. But then the true bliss will arrive the

More ›