Search
20 results for “rhgrouls”
-
@rhgrouls @sanchayan Day 2 of using #gitui. A lot to like - but some shortcomings.
Hope they are open to extend the current functionality. -
#gitui seems amazing. Will give it a shot.
For years now I have been using git-gui. Since tcl does not integrate great with my Desktop, pure terminal solution might work better.
Extending git-gui with their obscure language never seemed possible. With rust sky might be the limit.RT @rhgrouls
"lately, I have been using gitui, built in #rust. I set up a toggleterm + gitui keymap in #neovim, but I also like to use it outside of neovim. https://github.com/extrawurst/gitui"CC @sanchayan
-
Obviously it's a funny way to bring it. But considering #jacquesvallee ideas about absurdity, and the idea of mimicry, without claiming it is either real or false, there seems to be some overlap, right? Or is this more a case of apophenia https://en.m.wikipedia.org/wiki/Apophenia #ufo #uap
-
Interesting article
"As a culture, we’ve thus reached an impasse. On the one hand, the meagre amount of data that has been declassified or leaked isn’t enough for us to derive any firm conclusions regarding the nature of the phenomenon. On the other hand, enough has been begrudgingly but officially acknowledged that we can’t dismiss the phenomenon under prosaic accounts either."
-
Sarah Silverman sues #openai over #copyright . On the one hand, I can understand where she’s coming from. However, I ‘m not sure if copyright is exactly what is at stake here.
If I read a text, and I’m inspired to create content, I’m not infringing copyright. What’s the difference with AI?
-
@[email protected] I am looking for idea-lab mindsets overhere, not echo-chambers. I would classify your reaction as a setup for an echo-chamber. You're still invited yo the idea-lab
-
The main takeaway is that causal definitions of algorithmic fairness lead to Pareto-dominated policies.
In plain English, this means: whatever your preference (more efficiency or more equity) there will always be another strategy to is more satisfying to everyone involved.
Or, even simpler: this is bad for everyone involved.
-
How to fix this? The consequentialist framework (CF) to algorithmic fairness foregrounds the results of decisions, rather than properties of the prediction.
One starts by identifying the utility of different possible outcomes, eg efficiency and equity. Optimal decision policies can be derived with Linear Programming that uses stakeholder preferences.
This approach has advantages over static experimental designs (eg randomized trials)
-
"Learning to Be Fair: A Consequentialist Approach to Equitable Decision-Making"
While designing fair machine learning systems an approach could be to ensure parity in error rates across race, gender etc. It turns out to really matter how we define this statistically: some strategies might sound fair but ignore downstream effects and can cause unexpected harm to the very groups we try to protect.
-
A definition of consciousness:
The Integrated Information Theory (IIT) proposes that consciousness arises from the integration of information across different brain areas, and that the complexity of an organism’s consciousness is determined by the amount of integrated information it can process.
-
You can read the full paper here
-
An interesting approach towards implementiny critical thinking is to use philosophical strategies for prompt engineering. For example, the maieutic method:
"The term “maieutic” comes from the Greek word “maieutikos,” which means midwife. The method of maieutics is based on the idea that the teacher’s role is to help the student give birth to their own understanding of a subject, rather than to impart knowledge directly."
-
In general, the goal is to give the model perception, attention/awareness, emotional and ethical reasoning, critical and exploratory thinking.
All of these are realised with a submodular framework:
"CoCoMo consists of four modules: the receptor, unconsciousness, consciousness, and effector modules"
-
They propose splitting up attention and awareness in the model, which is really interesting:
"Awareness encompasses the full scope of conscious experience, while attention is a specific cognitive process that enables focus on certain stimuli or sources of information. In CoCoMo, an event that is being aware of can be placed in a low-priority task/job pool, awaiting a central scheduler to prioritize and pay attention to it. "
-
"Sigmund Freud was among the first to propose a model of the mind that incorporates both conscious and unconscious processes"
The idea of modelling multiple agents, centers of activity, areas, just as we have conscious, subconscious and unconscious processes is really appealing to me.
-
My first #MonthOfArxiv !
This paper describes a model for creating AI agents that combine knowledge with compassion.
AI is vulnerable to bias, toxicity and reliability issues; the idea is that by modelling emotional intelligence, impulse regulation and ethical reflection it should be possible to create generative AI that avoids these issues.
-
#nanoGPT reproduces #GPT-2 (124M) on OpenWebText, running on a single 8XA100 40GB node in 38 hours of training.
-
@WithDownes In terms of ecosystem, #python has a 20 year head start, but for my use cases both #julialang and #pyhton have nice ecosystems, both in terms of community and libraries.
However, what I really appreciate about Julia, is that the libraries are written in Julia. So I can always have a look and extend or modify them, without suddenly hitten a backend in another language like c/c++
-
Is there a #julialang equivalent of https://github.com/google/gin-config ?
I found using .gin files a really simple but useful way to store #hyperparameters during #deeplearning
If you do #machinelearning with #python and haven't heard of it, check it out!