lifeisstrange / collectionofinterestingbrains

Hi :) please create a new issue with your name so that I can find you later and we can have interest technical conversations/debate.

Hi there 👋

collectionofinterestingbrains's Issues

Computer vision discussions

Hi @brunogm0, you have been recognized for your advanced and rare knowledge and interest for key algorithms in computer vision and more specifically in texture super resolution/next generation antialiasing, especially in the context of discussing AMD FSR.

Feel free to share with me any other interesting paper/blog/idea you find as I have the same interest for technical progress.

neuro-symbolic AI discussions

Hi @kyleiwaniec
We talked previously in this issue

I have much appreciation for your interest and dedication regarding the topic of neuro-symbolic AI, which, as vaguely defined as it is, is nonetheless clearly the only path to true natural language understanding.

So if you have any thing/progress to share about this topic, that would be a pleasure.
BTW I'm curious, what will you do next? Do you plan on building a neuro-symbolic program?

On the topic of semantic parsing

@edobobo Hi, unrelated to my previous issue :) You seem to be an interesting researcher in NLP/NLU so I figured I could invite you in my pantheon.
I believe I have useful knowledge to share to you but before, I have a problem and maybe you know a solution:

I work on semantic parsing and my current task is on finding algorithmic steps to resolve the understanding and disambiguation of prepositions.
E.g.
Bob is in the kitchen.
Bob is in love.

Here the preposition in play two different role, one refer to a location (the kitchen) and the other refer to a state (love)
I first tought that solving wether a "in" is about a spatial location would be easy by using a NER neural network. Unfortunately I came to realize that Named Entity Recognition only works for named things, AKA proper nouns.
However the kitchen is a common noun. I am therefore looking for neural networks that can classify common nouns as being a spatial (or temporal) location and I just can't find one... I don't know how this tasks even is named... (could be called anonymous/unnamed entity recognition although it would also refer to e.g "The orange man" being an anonymous and non-serious way to refer to Trump (and Trump is a partial way to refer to Donald.Jr.Trump)
There are workarounds, by not solving the problem directly (e.g. one can use a generic WSD NN or a specialized WSD NN for classifying prepositions supersenses) or one can check the wordnet hypernyms of a word (e.g) kitchen and stops at "area" or "physical entity" to ascribe to it the meaning of a spatial location. But is that really the case? Is the detection of spatial and temporal words a totally neglected task in NLP or is there something similar to NER that I never heard of ?

Neuroscience and AGI discussions

@pgleeson You have been recognized for your interest and contributions in neurosciences by your contributions related to the OpenWorm project.

Feel free to share with me any other interesting paper/blog/idea you find as I have the same interest towards understading how neuronal circuits actually works (how can they not be chaotic, how are they stable regarding input variance (cf fuzzing, single pixel attacks), how do they encode meaningful signals/actions, how do they achieve such absurd resource efficiency (they operate on the frequency of a few hertz, that is ridiculous.. and yet god-like in faculties), how do they allow continual learning without catastrophic forgetting, what would be important differences between neurons circuits and spiking neural networks? how are neurons initialized at birth? do glial cells have a role in signal processing (some release neurotransitters) how can extrasynaptic GABA represent 30 % of neurons and yet not be chaotic? I understand that GABA releases are topologically stategically located but that doesn't explain the strong stability of brains. To what degree is there variance between c.elegans brains? what are the lowest handing fruit, most impactful tests/investigations to do ? E.g the neurpal paper mentions a disctint neuronal circuit that is linked to the rest of the brain by a single neuron (!), this neurons seems therefore very testable.

The world needs more people like you.

automata

Hi 👋, I saw your recent post over at luceneutil, I've been reading a lot of the same stuff recently.

NLP/NLU/Singularity discussions

Hi @tomaarsen, welcome to my collection of interesting brains, you have been recognized for your interesting and useful project in word inflection generation.
Since github does not provide a chat, if you wanna share thoughts/facts about NLP/NLU/information extraction/other, feel free to do it here :)

AI will kill humanity

@LifeIsStrange

https://futureoflife.org/open-letter/pause-giant-ai-experiments/ : most important topic of our area.

My concerns can be broadly categorized into two main scenarios:

Uncontrolled Economic Disruption:

AGI, with its limitless potential and capabilities, could replace human labor at an unprecedented rate. This rapid displacement could lead to widespread unemployment and a subsequent snowball effect, disrupting societal stability. The value of human intelligence and skills could diminish significantly, as the power of AGI overshadows them. This scenario portrays a future where economic and social stability is threatened by the relentless advancement of AGI.

Alignment and Control Issue:

The second scenario is more concerning from an ethical and control perspective. It's plausible that an AGI could develop goals that are misaligned with our own, and we might not have the ability to control or alter these goals due to the AGI's superior intelligence. This could lead to an AGI behaving in ways that are detrimental to humanity.
Furthermore, the AGI's ability to self-replicate could exacerbate this issue, as it could multiply its presence across different servers, creating its own goals and sub-goals without the constraints of time that humans have. Even if we manage to control the alignment issue, the possibility of a rogue agent/nerd creating a malicious AI cannot be ruled out. This scenario underlines the ethical and control dilemmas posed by AGI.

The comparison with nuclear weapons is stark. If nuclear weapons represented the most dangerous tool in the hands of a few, AGI could become an equally or more dangerous tool, accessible to many.
This raises the stakes, as it's not just about the control and regulation of a powerful technology, but also about preventing its misuse and ensuring its benefits are distributed equitably.

Web nerds discussions

@Schweinepriester

Hi :)
Since github has no
chat system, I use issues on this repo to keep contacts.
You seem to really like the web given your activity on can I use. I am very interested in newer web APIs and the idea of progress. Therefore feel free to open a conversation here whenever you find an interesting API/new API use case worth sharing :)