Charitably speaking: I suspect the commenter above you was indicating that the government should have a stock portfolio you can transfer stocks to to pay taxes in a non-taxable event type scenerio.
If you're taxing wealth (and not income) then switching stocks into cash doesn't change the wealth. Then use the cash to pay the taxes which reduces the wealth.
I believe the argument was that for some people, switching stocks into cash actually do change the wealth, because they own such a large % they are meaningfully moving the market by selling (the "paper money" argument).
And the answer to that was "if you're so afraid that selling will tank the value of the stock, then we [the government] will happily take your taxes as stocks directly, and we take on the risk that selling it will reduce its value".
> And the answer to that was "if you're so afraid that selling will tank the value of the stock, then we [the government] will happily take your taxes as stocks directly, and we take on the risk that selling it will reduce its value".
Yes indeed. It's basically saying "don't bother starting everything because the government will gradually just own it all anyway".
Certainly this applies to AI... but was there equivalent dangerous knowledge or tech which existed back in ye olden days that spawned such tales to begin with?
History is filled with creators being killed by their creations. With peoples greed and ambition overcoming their intelligence to a bad end.
Look at the parable of Icarus, a story of ambition and greed. But it seems very likely to be somewhat based on people experimenting with flying and learning about gravity the hard way. Not that they ever got close to the sun.
There is a KISS way to do this with our current models:
use a time string that updates:
every second (that it is not responding) erase a byte string at the end, then send a new byte string (that represents time from last response): bot either outputs a null terminating token, or if the number gets high enough, instruct/train it to take initiative then.
That only really gets you a limited sense, certainly better than nothing though.
The approach I would like to see would be more expensive token wise, but I don't think that can be avoided for fully interactive real time AI.
Add another dimension to position encoding to allow for multiple simultaneous streams. Run a model on each stream with an additional "I must speak" output value. A simple adjudicator model (maybe just softmax would work) decides which token(s) are emitted, each tagged by the channel that emitted them, which in-turn goes to the additional position encoding dimension. Then reinforcement learn them all at once with attention being allowed to look at all streams.
Then the model is always emitting tokens but while waiting it might just be emitting dum-dee-dum twiddle thumbs, on a channel that doesn't go out to the user.
It needs a way to focus attention though because you need a much larger potential context.
Is there any work on attention being limited to looking at a subset of embeddings defined by a parameterisable function because if a model could emit special tokens to change those parameters, it would potentially be able to scan it's own memory. Could be tricky to train but a training process that reduced the number of locations attended to over time might force it to compensate for the loss in scope by focusing.
I'm the weird one here. In JS at least, I reach for reduce before map and filter in most cases. Often it is because I want the accumulator, particularly when I have a list of objects with various properties that I wish to sum together in a reduced object.
I think it's funny how often "Folding Laundry" is brought up as the critical use case as if that's a labor function that needs to be done in the first place!
Some of us already live in the Utopia of not folding laundry!
I'm not advocating, but pointing out how silly it is that the supposed linchpin of the end of human labor is a completely voluntary and unnecessary task. I get why it can be a good example, because it's common and "hard" in that computery way that seems like it shouldn't be that hard, and that it's supposed to represent all the other hard problems of housework.
But we know what the household robot industry considers "solved" to look like, and I'm not that satisfied with roombas, and knowing that, is $6000 for something that could fold your laundry really that much better a proposition than just being wrinkly sometimes? Downey produces a product that you spray on a clothing item and stretch it and it stops being wrinkly. Totally unnecessary, but way cheaper than $6000.
folding laundry is an exemplary case of a lot of household work which involves carefully collecting items through the house, putting them into a machine to run a cleaning cycle, then putting them away in their right location.
folding is a hard middle step in the laundry process, but the job is end to end
Digging down through the source code, I see "const char *system_prompt_append;" in the config.
A lot of people have multiple files for their injected system prompts (a.la openclaw or hermes), I think it would be a good idea to either add or modify this point to be able to handle multiple file injections (system_prompt_append_folder or the like). Fitting that shape would make it easy to compare to those systems and make it easier for people to transition from those systems to yours.
low effort work though, that's nothing new... accelerating that low effort: that is new... but the low effort work hasn't been changed by ai: only illuminated more clearly.
> you can't be trapped in a sub-optimal situation and be rational at the same time.
In order for this statement to be true, you must also have perfect knowledge: something that is unachievable in most circumstances. Getting trapped in a local-optimal solution is perfectly possible if you are blindly following stochastic decent along a gradient: as it is too with life... getting trapped like this looks like being in the situation where there is no rational action (only irrational ones) that (eventually) lead to anything other that that status quo. a local optimum is where you have to do increasingly (in proportion to the level of optimization) non-positive actions to escape (likely to another local-optimal).
> In order for this statement to be true, you must also have perfect knowledge: something that is unachievable in most circumstances.
if the optimal situation requires perfect knowledge that can't be had then it's not the optimal situation, it's a fantasy. it's like saying the fastest route through a maze is a straight line from the start to the finish-- that would be true if the maze wasn't there.
there are two ends to what i'm arguing but people seem to only be able to hold one in their mind at a time.
It occurs to me: we have latent embedding giving 'general knowledge' to an LLM. What if we use a 'blank' LLM as well as an agent and train that blank LLM on personal context to query that as memory?
Can an LLM be trained to understand language without remembering anything else from its training data? I thought the intrinsic knowledge and the ability to understand language were tied together.
It would be closer to using an LLM as a RAG for memory, as the reasoning LLM in injected with the return of the 'memory llm' (maybe with a defined number of 'slots' for easy clean up).
reply