[HN Gopher] The era of exploration
___________________________________________________________________
The era of exploration
Author : jxmorris12
Score : 101 points
Date : 2025-07-07 15:23 UTC (1 days ago)
(HTM) web link (yidingjiang.github.io)
(TXT) w3m dump (yidingjiang.github.io)
| webdevver wrote:
| speaking of exploration, has anyone ever thought about non-
| cartesian displays? say, pixels that are arranged in a hexagonal
| pattern. or pixels arranged in a radial pattern and addressed via
| polar coordinates.
|
| what pattern are human color/light sensors arranged in? maybe we
| should replicate that pattern? an organic arrangement discovered
| via simulated annealing?
|
| all this bitmap x/y display stuff is very pre-AI-ish. old tech.
| victorian era clockwork mechanism. built to make it easy to
| reason about for humans, before the advent of neuron-soup. maybe
| we can do better?
| AlotOfReading wrote:
| Yes, it's been well-explored and there's a number of clever
| ways to convert them back to Cartesian coordinates because
| those are much nicer to work with for humans.
|
| The very earliest color CRTs used triangular arrangements of
| subpixels for each color, for example.
| card_zero wrote:
| This pattern?
|
| https://commons.wikimedia.org/wiki/File:ConeMosaics.jpg
|
| Maybe we should put the electronics in a layer in front of the
| display, make them transparent, and use them to filter the
| light. The vertebrate retina does something like that, putting
| nerves in front of light sensors. Imitating evolved solutions
| isn't always a good idea. Nature's design process is like:
| start with the wrong parts for the job. Lay them out back to
| front. Wire them together the long way round. Adapt this design
| slightly and latch on to any convenient side effects. Iterate
| ten billion times. Now it works pretty good! Or it doesn't, but
| the species survives anyway, which is also fine since there is
| no goal.
| visarga wrote:
| I like this approach. It's not that intelligence is in the brain
| or the model, and it should not be a noun. It is a process, a
| search process based on exploration and learning.
|
| Think of a river - the water carves the banks, the banks channel
| the water. Which is the real river?
|
| A model is like the river banks, it is not intelligent in itself.
| Neither is activity by itself. Intelligence comes from their co-
| constitutive relation.
|
| You can see the structure+flow coupling in many places. The model
| weights (banks) vs the training set (water). The contextual
| tokens (banks) vs the mass of probability for the next token
| (water). Agent environment (banks) vs agent activity (water).
|
| The trick here is that neither is capable of explaining the
| process, or more fundamental than the other. It is a recursive
| process rewriting its own rules.
|
| And we know from Godel, Turing and Chaitin that recursive
| processes exhibit incompleteness, undecidability and
| incompressibility. There is no way to predict them from outside,
| they can only be known only through unrolling, you have to be
| inside, or you have to be it to know it.
| chrisweekly wrote:
| +1, Brilliant metaphor, thanks for your thought-provoking
| comment!
| patcon wrote:
| You are one of the few random HN users who have your own page
| in my personal knowledgebase, because I consistently see your
| wonderful comments and i can only hope to be swimming in
| parallel to your perspectives :)
|
| Thanks for your great contributions, especially your dedication
| to metaphor!
| close04 wrote:
| > Think of a river - the water carves the banks, the banks
| channel the water. Which is the real river?
|
| Fantastic metaphor.
|
| I also can't help but draw a parallel (or even an overlap) with
| how the process works for humans. Once you put aside the
| intrinsic "model" (brain) differences (nature vs. nurture),
| humans develop their intelligence the same way: relentless
| exploration, and _decision making_ to guide that exploration.
| We don 't just funnel new info into a child's brain, they're
| allowed to explore threads around that, and take different
| paths which then guide how to process the next bit of info.
|
| If we're looking at building anything like human intelligence,
| the only advanced general intelligence we know and understand
| to a degree, exploration will be critical.
| AndrewKemendo wrote:
| You might like my latest work id be curious your feedback
|
| https://kemendo.com/GTC.pdf
| dodomodo wrote:
| reminds me of a tweet
| https://x.com/nearcyan/status/1937640182863548827
| dangus wrote:
| "Exploration" is a funny word for "copyright infringement." I
| seem to recall that our society fined college kids $150,000 per
| song download.
|
| I suspect that the amount of freely available text, video, and
| audio will decrease once entities like Disney start winning
| lawsuits. Alternatively, anyone who doesn't want their work to be
| crawled and trained from will keep their work behind
| login/paywalls moving forward.
|
| I think it's entirely fair that there should be a robots.txt type
| of mechanism for sites to opt out of being crawled for training,
| and we might see places like the EU come up with something of
| that sort.
| frozenseven wrote:
| >"copyright infringement"
|
| Throwing around scary words means very little. You're using a
| highly unorthodox interpretation of that concept. Even the
| anti-AI lawsuits are typically focused on specific model
| outputs or some other technicality.
|
| >entities like Disney
|
| The same company that's now publishing papers on generative AI?
| See the HN front page right now.
|
| >keep their work behind login/paywalls moving forward
|
| At what cost? Raging against a new technology like this is sure
| to backfire. Not only are you crippling yourself in terms of
| regular search engine visibility, there's now a non-negligible
| amount of traffic coming from AI search. And the average person
| will have even less of a reason to interact with you.
| Logins/paywalls are a nuisance. Especially if AI is producing
| competitive results to whatever you have to offer.
___________________________________________________________________
(page generated 2025-07-08 23:02 UTC)