Hacker Newsnew | past | comments | ask | show | jobs | submit | wumms's commentslogin

You’ve set up artificial selection in idea-space with yourself as the designer. What if AI could learn the rules of coherence from examples instead?

Even if coherence is learned implicitly, choosing a language, model, or representation already defines a bounded idea-space. The bias doesn’t disappear, it just moves into data and architecture. So my question isn’t whether coherence should be explicit or learned, but whether absolute exploration of an abstract idea-space is possible at all once any boundaries are imposed.

Guess absolute exploration hits the heat-death limit. You are hinting at a Drake-equation for bounded idea-space to guide AI: anchors x pressures x connectors x depth. Shift the boundaries for novelty.

Yeah, that’s a good way to put it. Absolute coverage feels like heat death, but changing the factors changes the space itself. That’s the part I’m still stuck on.

NASA uses "dark side" (meaning far side, not night side) when facing the public [0]:

> A series of test images shows the fully illuminated “dark side” of the Moon that is not visible from Earth.

> The far side of the Moon was first observed in 1959

[0] https://science.nasa.gov/earth/earth-observatory/the-dark-si...


> the fully illuminated “dark side”

Personally, I don't find the phrase 'fully illuminated “dark side”' to be a convincing alternative to the physically more accurate term 'far side'. Of course NASA has only just emerged from the Earth's dark side as I write this (UK here, mid-morning), so I'm not expecting an immediate response from them.

And yes, I do know that 'side' is itself not entirely accurate because of libration [0] but that's a different hill to die on.

[0] https://en.wikipedia.org/wiki/Libration#Lunar_libration


Agreed.

The far side is the darker side, though, at lunar night. Poetic proof: "The Earth shine might illuminate the light side of the Moon a little during the long night" (from Jules Verne, All Around the Moon https://www.gutenberg.org/cache/epub/16457/pg16457-images.ht...)


I do like poetry, but if we are looking at a crescent moon, in our night, it means that the bulk of farside is facing toward the sun, and will therefore be brighter than nearside

This describes lunar day on the far side of the moon, right? Excuse my ambiguity; I was comparing lunar nights only (inspired by the Jules Verne quote):

The far side is darker during lunar night (lit by starshine only; Full Moon on Earth) than the near side during lunar night (New Moon on Earth), because it receives both star- and max. Earthshine.

I'm not sure about Crescent Moon though: that only narrows the brightness gap slightly, right? Or I’ll have to ask if there’s an astronomer on board our flight.


> NASA uses "dark side" (meaning far side, not night side) when facing the public [0]:

That's not helpful, at all.


> hardware switch that physically disconnects your microphone

Might need disconnecting motion sensors, too:

AccEar: Accelerometer Acoustic Eavesdropping with Unconstrained Vocabulary https://arxiv.org/pdf/2212.01042

Spearphone: A Lightweight Speech Privacy Exploit via Accelerometer-Sensed Reverberations from Smartphone Loudspeakers https://www.winlab.rutgers.edu/~yychen/papers/%28WiSec%2721%...

A survey of acoustic eavesdropping attacks: Principle, methods, and progress https://www.sciencedirect.com/science/article/pii/S266729522...


Love it. Needs a moon.

Title could use (in Britain)

I see them on the leaderboard, but not on the main page.

https://www.ismypubfucked.com/leaderboard


> Like the conversation history is introducing noise rather than helpful context.

From https://docs.github.com/en/copilot/concepts/prompting/prompt...:

Copilot Chat uses the chat history to get context about your request. To give Copilot only the relevant history:

- Use threads to start a new conversation for a new task

- Delete requests that are no longer relevant or that didn’t give you the desired result


> LLM does not always cooperate

Getting from Visible Spectrum to Rainbow seemed impossible (tried all actions)


Interesting...if you see my comment/path above, I was able to do "Visible spectrum → Rainbow Chromaticity (places)."

Under the hood, the prompts are all something like this "Generate nouns that are broader, more general categories that encompass 'Piano'. Return exactly three distinct Wikipedia-style article titles that are nouns, formatted as a comma-separated list with no extra text. Do not include descriptions or adjectives. Focus on specific, named entities or concrete noun concepts only."

The outputs are then deduped from previous generations and any new options are displayed. Based on my result, if you spammed 'deeper' enough, the model would eventually have given you 'Rainbow Chromaticity'.

Uncooperative and non-deterministic.


Reminds me of Setup.exe


From https://www.theguardian.com/science/2015/jul/23/john-horton-...:

As he sat on the train to Cambridge, it dawned on him that since none of his classmates would be joining him at university, he would be able to transform himself into a new person: an extrovert! He wasn’t sure it would work. He worried that his introversion might be too entrenched, but he decided to try. He would be boisterous and witty, he would tell funny stories at parties, he would laugh at himself – that was key.

“Roughly speaking,” he recalled, “I was going to become the kind of person you see now. It was a free decision.”


i have fond memories of my own decision to become a tosser


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: