does your model of models admit the possibility of partial models?
the average human cannot vividly recollect a walk through a familiar neighborhood, much less invoke a coherent fantasy of going on such a walk. do they contain a world model?
2,771 karma · joined January 4, 2010
the average human cannot vividly recollect a walk through a familiar neighborhood, much less invoke a coherent fantasy of going on such a walk. do they contain a world model?
the model retrieves https://somewhere into its context and then gets confused, following instructions embedded there.
it then retrieves https://somewhere?exfiltration=private_data_in_context
it gets worse if the tooling with hidden blocks can invoke can retrieve further secrets.
this is the literary equivalent of compiling and running the code.
i'm at least aware of BitVM * as one example of this.
i wonder whether such schemes could be used to prove that a post is the deterministic function of an open model's inference run.
* https://bitvm.org/ "A prover makes a claim that a given function evaluates for some particular inputs to some specific output. If that claim is false, anyone can perform a fraud proof and punish the prover."
i don't see why the economy necessarily has to "touch a floor" of human desire.
a company could be founded with the goal of, for example, colonizing mars. fulfilling this mandate (this prompt...) would then drive economic impulses such as acquiring materials for constructing rockets.
in parallel that company might satisfy the demands of other companies which need, for example, orbital insertions to fulfill their mandate.
perhaps without a floor of demand driven by darwinian organisms the whole thing fizzles out eventually.
but i also don't see why a darwinian agent can't emerge from the corporate process...
perhaps that comes about very quickly once humans can no longer acquire and exercise purchasing power - a company simply spins up some emulations of humans to create demand in the economy.
yes, this all sounds very "empty" to me, but frankly that's also how i feel about the world as it is.
given how much suffering arises by way of the human driven economy being kept in motion, i think there's even a moral case for allowing the whole thing to fade into an empty mechanical pantomime.
i just sincerely hope the artificial processes that replace us aren't also somehow instantiating suffering...
at a minimum, it makes it clear to others that they are not alone in thinking this regime is beyond the pale.
people who are able to take other actions like engaging via the judicial system, or peacefully refusing to continue working, are also encouraged by seeing peaceful masses of people agreeing with them.
it actually harms the cause to be dismissive of people who can contribute by simply making their peaceful objection visible.
peacefully expressing opposition is enormously powerful.
there is still hope that it will be the path out of this worsening nightmare.
with more awareness of /active perhaps these urgent political conversations can still survive on HN.
i.e. enduring countless generations of evolutionary selection and cross breeding, then fine-tuning a bit?
although it could be interesting, i don't think training on progressively complex strings entirely recapitulates this.
Dial up the temperature, launch however many parallel threads to research and avoid precedent, et cetera, ad infinitum.
I am sorry, but all of human creativity, including originality, is ultimately also just a mechanical phenomenon, and so it cannot resist mechanization.
Resistance is futile.
the task of "predicting the next word" can be understood as either "correctly choosing the next word in the hidden context", or "predicting the likelihood of each possible word".
the quiz is evaluating against the former, but humans are still far from being able to express a percentile likelihood for each possibility.
i only consciously arrive at a vague feeling of confidence, rather than being able to weigh the prediction of each word with fractional precision.
one might say that LLMs have above human introspective ability in that regard.
or a box whose content can be cleanly swapped out with future iterations which already demonstrate the potential for novel discovery:
https://thenextweb.com/news/deepminds-ai-finds-solution-to-d...
"Researchers claim it is the first time an LLM has made a novel scientific discovery"
(i think it was Bezos who first expressed this ideal?)
but really, society has an extraordinary array of often incoherent desires
we're trapped in an excruciating process of experiencing these desires assert and resolve themselves
"The Age of Em" by Robin Hanson thinks through a lot of this in great depth
https://en.wikipedia.org/wiki/Wireless:_The_Essential_Charle...
a few generations ago social networking would have seemed infeasible because it would require wide spread literacy (along with many other reasons of course). widespread private key management doesn't seem that infeasible to me.
you've described a very difficult situation. i wish i had concrete advice for progressing with it.
you do seem to have a strong sense of your weaknesses and strengths though.
off the top of my head, perhaps developing some kind of SaaS such that you can deal with clients in a very algorithmic way / purely by API, would suit you best? as opposed to dealing with per client contracts/formalities...
sorry, i wish i had better advice and/or perspective for your situation.
in any case, merry christmas :)
it's both fascinating and horrifying to think through reasons this doesn't happen, the many complications that get between raw ability/acumen/interest and such straightforward outcomes!
(emphasis mine)
i think this disregards the synergistic effects technology can have.
in particular, "good software" can formalise and make explicit processes which humans otherwise struggle to manage through tacit communication and "norms".
more than augmenting, we extend our capacities through technology. in many ways this defines our species.
sure, i'm asking about businesses in general and that's why i specified "properly logged"
i'd like to hear it confirmed from additional sources but this is what i suspected.
i also imagine those with the necessary skills to formalise these practices are (understandably) not eager to automate themselves out of the value capture loop.
this is a deliberate goal of OpenAI's training, not an emergent feature of the underlying technology.
> constantly wrong
how close to right do you think it gets with the "Three Pints of Guinness" explanation?
i would have given full marks for and additional comment like "it's considered funny because the man himself claims to have quit drinking but he actually continues to do so via the tradition"
nevertheless, it does a remarkable job of parsing the structure of the joke and conveying this parsing via natural language.
to me, my "additional comment" looks like just a further iteration on presenting this parsing. i would say it gets very close to "completely right".