UraniumBlazer

Original post

Original creator

This is fanart of a webcomic called “Castle Swimmer”. It’s really really good! Here’s a link.

Original post

Original creator

This is fanart of a webcomic called "Castle Swimmer". It's really really good! Here's a link.

Orbit is an LLM addon/extension for Firefox that runs on the Mistral 7B model. It can summarize a given webpage, YouTube videos and so on. You can ask it questions about stuff that's on the page. It is very privacy friendly and does not require any account to sign up.

I personally tried it, and found it to be incredibly useful! I think this is going to be one of my long term addons along with uBlock Origin, Decentraleyes and so on. I would highly recommend checking this out!

YouTube description: 58 of you watching this video right now will not be alive next week. And it’s not because of some freak accident or rare disease. It’s because of everyday actions you probably think are harmless. Let’s save your life today by looking at what is most likely to kill you next week – so you can avoid it.

The idea is simple. A worker-consumer hybrid coop that develops, maintains and hosts a lemmy-like fediverse platform that is open sourced.

There r two pricing tiers- a free and paid tier. If u pay a monthly membership fee, you become a member of the consumer body. If u r hired by the coop, u of course become part of the worker body.

The core of the coop's workings are direct democratic. Creating, filling and destroying job positions are all done direct democratically. To pass a piece of legislation, either one of the following conditions need to be met:

1. Simple passing: Both, worker and consumer bodies cast more than 50% votes each for the given bill.
2. Consumer override: If the consumer body casts more than two thirds of the votes for a bill.

Assume that the quality of the platform is as good as Lemmy is right now. Assume that the functionality is similar too.

Would you be interested in being a member? Do u think this is a good idea?

I personally find Lemmy's current donations based model to be severely lacking from a fundraising point of view. There needs to be a better form of organisation imo.

The direct democratic consumer coop element would bring in more people imo. I'm hoping that the worker coop element prevents worker exploitation.

Do you think this is an absolutely horseshit idea? Or do u kinda like it? Or do u have any suggestions? I'm seriously considering this, which is what made me ask this here. I have a Lemmy client nearing the MVP stage which I was developing with this purpose in mind. Sorry if this is the wrong community for the post.

I love this artist so much! Please do check her out.

TLDR: Google's DeepMind has developed a new open sourced AI system called AlphaProteo, which can design novel proteins that bind to target molecules. This technology has the potential to accelerate progress in various fields, including drug development, disease understanding, and diagnosis.

AlphaProteo was trained on vast amounts of protein data and has learned the intricate ways molecules bind to each other. It can generate candidate proteins that bind to target molecules at specific locations, and its designs have been validated through experiments.

The system has shown promising results, achieving higher experimental success rates and better binding affinities than existing methods. It has also been able to design successful protein binders for challenging targets, such as VEGF-A, which is associated with cancer and complications from diabetes.

However, the system is not perfect and has limitations, such as being unable to design successful binders against certain targets. To address these limitations, DeepMind is working to improve and expand AlphaProteo's capabilities.

The development of AlphaProteo raises important questions about responsible development and biosecurity. DeepMind is working with external experts to develop best practices and is committed to sharing its work in a phased approach.

Overall, AlphaProteo has the potential to revolutionize protein design and accelerate progress in various fields, but it requires careful consideration of its limitations and potential risks.

I'm using lemmy-js-client for app development. I want to render comments in the nested form (like normal people do).

The problem is, the lemmy backend spits out comment lists in a fashion that is unfriendly for nested rendering. Why? It outputs comments whose paths are like follows (for example):

0.1.2.3.4

0.1.2.1

0.1.2.3.4.5

0.1.3.1.5

0.1

Let's say the limit that I've set here is 10. Many a times, the parent comments of the comments in the page are out of the page.

For example, let's say I asked lemmy for comments for a given post. It gives me an output like above. There are many children comments here on page 1 (like 0.1.2.1, 0.1.2.3.4 and so on). Their parent (0.1.2) is NOT on this page. It is on the page that follows.

Hence, I would need to do client side bs to get the correct parent comments.

What is your approach for doing the above?

This is what I have settled for now unfortunately. I fetch all the comments under a post and then convert them in my nested form. This means, that my app doesn't paginate. This thus results in really slow loading times for posts with more comments. The more comments a post has, the slower they will load. This sucks.

I tried other ways, mind you. While implementing pagination, I simply removed orphan comments when on a given page. If the user decided to go to page 2 (simply by scrolling), suddenly, these orphan comments would not be orphans anymore due to them finding their parent comments. This in turn, fucked with my ui completely (which was obvious), thus making the list randomly scroll like crazy. This was a really shitty experience for the user.

Sooooo what have you guys done? How have you handled the situation?

Neural networks have become increasingly impressive in recent years, but there's a big catch: we don't really know what they are doing. We give them data and ways to get feedback, and somehow, they learn all kinds of tasks. It would be really useful, especially for safety purposes, to understand what they have learned and how they work after they've been trained. The ultimate goal is not only to understand in broad strokes what they're doing but to precisely reverse engineer the algorithms encoded in their parameters. This is the ambitious goal of mechanistic interpretability. As an introduction to this field, we show how researchers have been able to partly reverse-engineer how InceptionV1, a convolutional neural network, recognizes images.