Alternative is reimagining financial services through lending

Alternative

Why should I use your product? I’m a web app developer and I love doing what I do. I code in a language that seems highly frowned upon, yet is one of the most powerful and used languages on the Internet. It is so powerful that it makes up about 80% of the backend code on the Internet. I have even made a few hundred dollars creating small apps as a SaaS (software as a service). However, when it comes to being the competition or the alternative, the giants will always win. By giants, I mean services like Google, Dropbox, Github, and even Medium.com. Nothing I have created hasn’t been done before — and maybe that is my fault, but oftentimes I find myself trying to solve my own pain points, using other applications to do it, only to find that something within those apps is missing — that or the fact that I’m too cheap and would just prefer to create an alternative version of the product so I can use it for free — and then charge others a cheaper price — becoming the alternative and the competition. Every time I release a new app, however, it often receives several different types of responses. They go something like this: Crickets Chirping If the crickets aren’t out, than I usually get a few people signing up and trying the service. I have had a few successful launches in the past, including , , and which have acquired a few individual users and some businesses. It has been great to have a few consistent signups paying me every month, but it has not been enough to quit my day job, so I have continued my projects anyway, creating new useful tools. MyPost Call Me Private Text Privately Recently, I created a WordPress Managed Hosting solution called . This is a hosting solution ahead of its time and is unique in the fact that it does what no other host does: it has a list of features which were born out of the complaints and suggestions of so many web developers who were hosting their websites on other WordPress Managed hosts. It automates the entire process of being a web developer and webmaster, meaning you can setup your clients on and never have to worry about updating their websites again, as it does it automatically. WP Bubble WP Bubble The launch wasn’t as successful as I hoped, but I managed to acquire an investor in the product, who is hosting their website with and they pay me slightly more than what I charge for the product in their own hopes to continue supporting a good product — and a few customers — two who churned within a day or two — trials of the product — and a third who left and returned a month later, and is actually using the product to create a few websites of his own. WP Bubble The most recent product I released is called , which is an image and file hosting solution with a very fast CDN. The most prominent features include an easy comment system, tracking of geometrics on clicks, and the ability to set a password before anyone can view the file. The CDN is also enhanced with Cloudflare to keep my bandwidth costs very low — security high— and the speed of the downloading of the file very fast. The launch wasn’t big at all, as I only released it in one or two areas, and the only response I received was: BinCDN What makes this product different from Google Drive or Dropbox? This question is what prompted me to write this article. I had to ponder: everything my product does, Dropbox and Google Drive already do. So what makes my products different? I could certainly come up with many different technical reasons — and Google and Dropbox will match every single one of them. What makes me different or what makes my product different? I’m not a million or billion dollar company, but I do pride myself on the privacy and security of my products. I do pride myself on the fact that I personally took the time to research a design for a product, spent hours upon hours writing out a plan for my product, and had complete personal say in how my product looks, what it does, how it functions, and the fact that it is personally supported by me. Best of all, I pride myself on the fact that I actually wrote the code for my product and don’t need to seek approval from anyone else about my product or go through a long review process, only to have my changes or ideas rejected by someone “higher up” telling me that my code isn’t good enough. Most of my products released are not free. I used to release them for free and keep them free for as long as I could support them — what would end up happening, I’d lose interest, no longer support them, and they would die out. Now that I have monetized them, they have a longer chance for survival and I care enough about them to continue supporting them when I know others will support them too. What happens when Google decides to discontinue a product that a few people find useful and are using? What if Dropbox decided one day that it was going start charging more for everyone, as Photobucket once did? What if Facebook or Twitter decided to filter and ban people for their freedom of speech, as they often do? They have every right — they are offering it for free — and you really have no say, especially when it comes to your data. What happens when you are paying for and supporting my small products? My small business? I’m more likely to continue supporting it and supporting you because I really do care about you, as an indivudal. I care about your small business. I care about your large business. I care that you are using my products. I can develop them to the standards of my users and my product is small enough that I care enough about making those changes and catering to the demands that my customers need my products to do for them. Personally and professionally, I’m not about you. The less I have — the better. Despite the fact that both Dropbox and Google give away most of their products for free for a time, they still sell you on the large plans — and then sell your data on top of that. Double whammy, don’t you think? collecting or keeping and selling any data to third-parties And no one seems to care what happens with their files or their data, but it became a fact of life — that your information — habits, lifestyle, likes, dislikes, loves, hates, friends data, family data — is literally available and being sold to the highest bidders, all for money and the advertising industry. Sure, your data is sitting in some top notch secured warehouse, but so are my projects as well, as I spin the servers up using as my primary hosting solution for all projects I start. Vultr I have had a few of my customers ask about the security and protection of their data — before I even began building products, I learned about SHA-256 and Hash encryption methods and a lot about securing data and keeping it protected and private — so that not even I could pry through it. While smaller companies might get hacked, larger companies, such as and , are under constant attack and often fail to report that they were ever hacked — losing millions of customers’ data to hackers every year. Google Facebook While they claim to care about your data — and they sure do — and even have programs that pay people to report their bugs to them — the fact is: they are too large to really care about you personally. And sure, they do care about you — as long as you don’t try to sue them because they have so many lawyers protecting them, you’ll have years of court to go through just to get anywhere. With my own business practices and the size of my company and customers: I cannot actually afford . Let me repeat that: I care enough about you to try and answer every email, every live chat that is opened, and respond to you personally to take care of your problem within a timely manner. I may not get it right every time, but I’m personally there to help. I’ve had a product not cancel a subscription and charge a user for an extra month — and I’ve happily refunded my former customer without incident, without question, and without issue. I have not hired other people because I am too small to afford anyone else, so I personally am the web developer, the app developer, the tech support, the salesperson, the marketer, and everything in the company. NOT TO CARE I CANNOT afford NOT TO CARE about you and your business. There is another thing: I’m not Walmart, I’m not Amazon — great big companies that I buy things from — and I’m not a big business. I even barter with other small businesses to use their products and allow them to use mine, all so we can save ourselves from having to pay cash or credit to each other. I’m still a small business and I support other small businesses. Tell me all about the large businesses that are supporting you — sure, you can get products for cheap and in bulk — and that is a great way to support your business, but as a small business owner, who is buying your products and how can you even compete with the likeness of Amazon or Walmart brands? — Or any of the other brands out there? Believe it or not: these are pretty much the top companies that own everything you and everyone else will ever buy — to them, you’re not really competition, and you barely exist, and they are truly after putting you out of business and wouldn’t hesitate to do so. The biggest corporations are notorious for buying smaller companies and oblierating their operations so they can remain the only ones without competition. Don’t hate them too much — it’s the way of the world, and they aren’t going anywhere. They will all be here long after our great grandchildren pass away, and this is mainly just the companies that deal with food. I don’t make enough from my products to quit my day job. It is my dream and the dream of many to no longer have to work at their day jobs, but rather focus on their passions, but it is still not — and is a long ways away from my reality. I’m a “local online business” that is trying to make a living from my products. I have yet to resort to selling my body for sex or science. I’m pretty sure science will accept me. Indie Hackers I did slave labor for a very long time in my teenage years and for almost half of my twenties, and for a short time in my thirties. In fact, I’m still doing slave labor, but it is more mental than physical nowadays. I go to work and I code for nearly 8 hours a day. I come home and despite being mentally exhausted, I work on my products, coding for at least another 4 hours. I have bills to pay, rent that is due every month, and food that my body requires. I have taxes to pay. I don’t live a luxurious lifestyle at all, though sometimes I do dream about waking up and being on a wonderful cruise. If you build it — they won’t come. If you advertise it — they might come. If you keep pushing it — they might use it. If you find it valuable — they might see the value in it. If you charge for it — they might pay for it. If you talk about it or blog about it, eventually, SEO might help you get discovered, and you might get a few interested people who will sign up, and maybe you will get lucky enough to get a few paying customers who like your products and don’t mind paying you their own hard earned money to solve their own pains and problems. Regardless of being the competition or used as an alternative, I have never given up on my talent, I don’t consider anything a failure — because I built my products for me first and foremost. I use most of my products every single day. As long as they are useful to me, how can they actually be a failure? I then monetize them for others to use. I try to spread the word with my story telling skills or the . little funding I have to try and hire others on Fiverr to help me So hopefully I’ve answered the question “ I’m just me, a human being, an individual trying to make it in this world. I enjoy programming. I enjoy creating products. I enjoy when others use my products and actually pay the little money I need to keep funding them and my future endeavors. I don’t have to answer to any stockholders or sell your data to companies to become the most profitable company in the world. What makes this different than Google Drive or Dropbox?” Yes, I would love to be the next Google, Facebook, Apple, or even Dropbox, but I already know it’s not going to happen, and I’m okay with that. I don’t have the intelligence of a hundred programmers. I have the intelligence of just one. What I do know is that I continue programming and solving my own pains and sharing my products with the world in hopes that maybe somewhere out there, one of you will have the same exact pain as me — love my product, use my product, and even pay me for my product. So all of the giants may be my competition, but my products are still an alternative. Image Sources: Google

Competing Against The Giants As An Alternative

The need for decentralized cloud infrastructure has become increasingly evident, not only to compete with centralized infrastructure but also to offer greater security and reliability to users. But those of us in the Web3 world know that it's difficult to make clients understand why decentralization is better in this area. In this article, I'll explain my perspective and experience in both worlds (for example, as a Microsoft MVP and Alibaba Cloud MVP) on how we can begin to mainstream decentralized clouds to both corporate and personal users. In this article, I'll explain my perspective and experience in both worlds (for example, as a Microsoft MVP and Alibaba Cloud MVP) on how we can begin to mainstream decentralized clouds to both corporate and personal users. What are the similarities between centralized and decentralized clouds? Before looking at the benefits of decentralization, we need to look at the similarities between centralized and decentralized services. This will help us understand at what point there isn't a single factor that makes a customer switch to a decentralized provider, and how we can differentiate ourselves later. These similarities can be broken down as follows: User experience: Although centralized services have more features (because they have been on the market longer), the user experience is similar when using interfaces from decentralized providers, so the learning curve is minimal in this area.
Infrastructure as a Service: Decentralized providers have focused on simplifying access to infrastructure for their users, specifically as IaaS (Infrastructure as a Service) where there is no difference (in terms of functionality) between the two types of providers.
Scalability: Both centralized and decentralized services are designed to scale with user needs, allowing for the management of larger volumes of data or users. User experience: Although centralized services have more features (because they have been on the market longer), the user experience is similar when using interfaces from decentralized providers, so the learning curve is minimal in this area. User experience Infrastructure as a Service: Decentralized providers have focused on simplifying access to infrastructure for their users, specifically as IaaS (Infrastructure as a Service) where there is no difference (in terms of functionality) between the two types of providers. Infrastructure as a Service Scalability: Both centralized and decentralized services are designed to scale with user needs, allowing for the management of larger volumes of data or users. Scalability There are many more similarities, and taking them into account is especially relevant when we want to sell decentralized solutions, where we need to look at other features that make a difference so our customers can decide on the options we offer, and that's what we'll look at now. There are many more similarities, and taking them into account is especially relevant when we want to sell decentralized solutions, where we need to look at other features that make a difference so our customers can decide on the options we offer, and that's what we'll look at now. The technical advantages of a decentralized Cloud In technical terms, we know that decentralization is the main factor because it gives us unique characteristics, such as: Broad participation Any person or institution can contribute to the ecosystem with resources, applications, etc., making the ability to build, deploy, and use services completely open and more adaptable to any business model. Incentives Participants can be rewarded for their contributions, attracting more stakeholders and creating a richer, longer-term ecosystem. Greater resistance to failure By distributing data and resources across multiple nodes rather than relying on a centralized hub, a single point of failure is less likely to impact system availability. Reduced latency In a well-established decentralized network with many participants, latency can be much lower than in its centralized counterparts because nodes are more likely to be closer to the end user. And there are other very valuable technical features, such as flexible and unlimited scalability, seamless operation, open innovation, and more. But these won't be features that will be useful to sell to your customers since they are looking for certain attributes that are not technical, are not found in centralized services, and can make decentralized services start to scale to millions of customers. And there are other very valuable technical features, such as flexible and unlimited scalability, seamless operation, open innovation, and more. But these won't be features that will be useful to sell to your customers since they are looking for certain attributes that are not technical, are not found in centralized services, and can make decentralized services start to scale to millions of customers. The benefits for end users and customers But decentralization isn't a special factor for a customer when deciding to use a service (if you think about it, there are still people who believe that the money in their bank is their money and can't be restricted). Decentralization is a bridge to real benefits for end users, and some of those benefits are: But decentralization isn't a special factor for a customer when deciding to use a service High availability at a lower cost In my experience working with various clients and implementing cloud solutions, they often seek to reduce costs and do so by compromising on high availability, confident that there will be no problems or at least that the impact will be very low. The reality is that they later realize that high availability is key, but they no longer have the budget to improve their implementations. This is where decentralized services make a difference. In addition to the lower cost of use by not depending on a single entity, they also provide greater confidence that high availability can be achieved by distributing the workload among many nodes distributed globally. Greater security in data processing If we look at the latest news about security breaches in many applications and services, we see that all (or almost all) of them reside on centralized servers, because it only takes one error or attack (not always as sophisticated) for user and customer data to be exposed, including their behavior and privacy. This not only reduces the levels of trust in some platforms but can also help us increase them in the decentralized Cloud. Decentralization, supported by infrastructure like blockchain networks, can provide greater guarantees regarding data processing, such as data sovereignty because they are based on proven encryption methods with decades of research and use. That's what we have to sell: secure infrastructure and encryption methods that empower our customers over their information. That's what we have to sell: secure infrastructure and encryption methods that empower our customers over their information. Greater customization One of the advantages of a decentralized ecosystem is that the level of development and services that can be built are completely open; the only limit is your imagination (and resources, of course). It's in this realm that you can sell something unique, something truly different. Centralized services, while they allow for some customizations, are always associated with their own products and you can hardly make adjustments and connections. Many times you will have to do some additional development (which can complicate your defined architecture), which is an effort and resources to consider, while in a decentralized Cloud, you have greater control over the use of your services. Greater portability Something I do frequently when designing Cloud architectures is do a portability cost analysis, that is, how easy or difficult it will be to migrate from the selected provider to another, and generally centralized services are not very good in this matter, because the incentive is to retain customers at any cost, and the strongest is with their services that are not similar in other providers, which forces you to stay tied to that provider (think for example of Power BI Cloud, if you want to migrate your reports you will have to do them manually in the other provider). Decentralized clouds are built on open-source technologies, so the cost of portability is greatly reduced (it is not zero, but it is a lower impact effort than doing it in Azure or AWS, for example). Decentralized clouds are built on open-source technologies, so the cost of portability is greatly reduced (it is not zero, but it is a lower impact effort than doing it in Azure or AWS, for example). Final words In this article, the idea was to explain, based on experience and in a simple way, how we can make a difference when we want to sell decentralized platforms in the cloud services space, where small and large companies require their use for their own needs, and how we should focus on less technical areas. One example that is making a difference in this space is Aleph Cloud, where you can deploy your applications using decentralized storage and computing, and even host your applications in a better and more secure way. Aleph Cloud Aleph Cloud Links Aleph Cloud: https://aleph.im/
About myself: https://www.linkedin.com/in/nescampos/ Aleph Cloud: https://aleph.im/ https://aleph.im/ About myself: https://www.linkedin.com/in/nescampos/ https://www.linkedin.com/in/nescampos/

Amazon and Microsoft Don’t Want You to Know About This Cloud Alternative

Alternative Canonicalizations Expand Chemical Similarity Search Capabilities

Are you looking for a free alternative to ChatGPT? If so, you’re in the right place! In this article, I will introduce you to HuggingChat, an open-source AI chatbot from Hugging Face. Unlike other AI models like ChatGPT or Gemini, HuggingChat runs on various open-source models that you can explore, modify, and even fine-tune according to your needs. Let’s dive in! What is HuggingChat? HuggingChat is an AI chatbot similar to ChatGPT but built on open-source models. It is designed to provide conversational AI without the limitations of proprietary models like OpenAI’s ChatGPT or Google’s Gemini. Unlike ChatGPT, which is based on OpenAI’s proprietary model, HuggingChat uses multiple AI models from different providers, such as: Llama


Mistral


DeepSeek


Mixtral


Etc Since these models are open-source, you can see their code, modify them, and even integrate them into your own projects. How to Use HuggingChat Using HuggingChat is simple and doesn’t require setup. Step 1. Create an Account First, go to HuggingChat and create a free account. Once logged in, you will be taken to the chat interface. Step 2. Select a Model Unlike ChatGPT, where you only have access to a fixed model, HuggingChat lets you choose different AI models for different tasks. Here’s how: Click the Settings icon or the Models link in the sidebar. Browse the available models.
Select the one that best suits your needs. Each model has different strengths, such as better reasoning, faster responses, or the ability to generate images. Step 3. Start Chatting The interface is very similar to ChatGPT, so if you’ve used ChatGPT before, you’ll feel right at home. Simply enter a prompt and the model will generate a response. Features of HuggingChat 1. Search-Enabled Responses One of the most useful features of HuggingChat is the search function. If an AI model doesn’t have up-to-date information, you can enable search, which allows it to look up the latest facts online. For example, when I asked about the latest iPhone model, the default AI model didn’t know the answer. However, by enabling the search function, I was able to get the correct information. 2. Image Generation HuggingChat supports image generation! However, not all models can generate images. To use this feature: Go to Settings.
Choose a model that supports image generation (e.g., Llama 3).
Click on the image generation icon.
Enter your prompt (e.g., “Generate an image of a futuristic city”). 3. Multiple AI Models HuggingChat allows you to experiment with different AI models to see which works best for your task. Some models are better at reasoning, some are faster, and some support image generation. 4. Custom Assistants Similar to OpenAI’s Custom GPTs, HuggingChat has Assistants, which allow users to create custom AI configurations for specific tasks. For example, you can create an Assistant specialized in: Coding (Python, JavaScript, etc.)
Writing and summarizing text
Translating languages Or you can select Assistants from the community. 5. File and Image Uploads Certain AI models in HuggingChat allow you to upload files or images for processing. This is useful if you want to analyze data, read documents, or perform image-recognition tasks. 6. Free and Open-Source As I mentioned before, unlike ChatGPT or Gemini, HuggingChat is completely free and based on open-source AI models. Video Review of HuggingChat Watch on YouTube: ChatGPT Alternative — HuggingChat Conclusion If you want a free ChatGPT alternative that is customizable and privacy-friendly, HuggingChat is worth trying. Give it a try and let me know your thoughts! Do you think it can replace ChatGPT? Let me know in a comment below! Cheers! ;) Are you looking for a free alternative to ChatGPT? If so, you’re in the right place! In this article, I will introduce you to HuggingChat, an open-source AI chatbot from Hugging Face. Unlike other AI models like ChatGPT or Gemini, HuggingChat runs on various open-source models that you can explore, modify, and even fine-tune according to your needs. Let’s dive in! What is HuggingChat? HuggingChat is an AI chatbot similar to ChatGPT but built on open-source models. It is designed to provide conversational AI without the limitations of proprietary models like OpenAI’s ChatGPT or Google’s Gemini. Unlike ChatGPT, which is based on OpenAI’s proprietary model, HuggingChat uses multiple AI models from different providers, such as: Llama Mistral DeepSeek Mixtral Etc Llama Llama Mistral Mistral DeepSeek DeepSeek Mixtral Mixtral Etc Etc Since these models are open-source, you can see their code, modify them, and even integrate them into your own projects. How to Use HuggingChat Using HuggingChat is simple and doesn’t require setup. Step 1. Create an Account First, go to HuggingChat and create a free account. Once logged in, you will be taken to the chat interface. HuggingChat Step 2. Select a Model Unlike ChatGPT, where you only have access to a fixed model, HuggingChat lets you choose different AI models for different tasks. Here’s how: Click the Settings icon or the Models link in the sidebar. Click the Settings icon or the Models link in the sidebar. Browse the available models. Select the one that best suits your needs. Browse the available models. Select the one that best suits your needs. Each model has different strengths, such as better reasoning, faster responses, or the ability to generate images. Step 3. Start Chatting The interface is very similar to ChatGPT , so if you’ve used ChatGPT before, you’ll feel right at home. Simply enter a prompt and the model will generate a response. very similar to ChatGPT Features of HuggingChat 1. Search-Enabled Responses One of the most useful features of HuggingChat is the search function . If an AI model doesn’t have up-to-date information, you can enable search, which allows it to look up the latest facts online. search function For example, when I asked about the latest iPhone model , the default AI model didn’t know the answer. However, by enabling the search function, I was able to get the correct information. latest iPhone model 2. Image Generation HuggingChat supports image generation! However, not all models can generate images. To use this feature: Go to Settings. Choose a model that supports image generation (e.g., Llama 3). Click on the image generation icon. Enter your prompt (e.g., “Generate an image of a futuristic city”). Go to Settings . Settings Choose a model that supports image generation (e.g., Llama 3). Click on the image generation icon . image generation icon Enter your prompt (e.g., “Generate an image of a futuristic city”). 3. Multiple AI Models HuggingChat allows you to experiment with different AI models to see which works best for your task. Some models are better at reasoning, some are faster, and some support image generation. different AI models 4. Custom Assistants Similar to OpenAI’s Custom GPTs , HuggingChat has Assistants , which allow users to create custom AI configurations for specific tasks. Custom GPTs Assistants For example, you can create an Assistant specialized in: Coding (Python, JavaScript, etc.) Writing and summarizing text Translating languages Coding (Python, JavaScript, etc.) Writing and summarizing text Translating languages Or you can select Assistants from the community. 5. File and Image Uploads Certain AI models in HuggingChat allow you to upload files or images for processing. This is useful if you want to analyze data, read documents, or perform image-recognition tasks. 6. Free and Open-Source As I mentioned before, unlike ChatGPT or Gemini, HuggingChat is completely free and based on open-source AI models. Video Review of HuggingChat Watch on YouTube: ChatGPT Alternative — HuggingChat Watch on YouTube: ChatGPT Alternative — HuggingChat ChatGPT Alternative — HuggingChat Conclusion If you want a free ChatGPT alternative that is customizable and privacy-friendly, HuggingChat is worth trying. Give it a try and let me know your thoughts! Do you think it can replace ChatGPT? Let me know in a comment below! Cheers! ;)

Meet HuggingChat: A FREE and Open-Source Alternative to ChatGPT!

Traditional hashing techniques like perceptual hashing (pHash) and locality-sensitive hashing (LSH) are widely used for image similarity detection. However, they often fail in real-world scenarios where images undergo transformations such as cropping, noise addition or color changes. Graph Neural Networks: A Smarter Approach to Image Similarity Graph Neural Networks (GNNs) provide a more robust alternative by modeling image relationships as a graph and propagating similarity information through message passing. Instead of treating images as isolated entities, Graph Neural Networks (GNNs) represent images as nodes in a graph where edges define their relationships. This allows context aware learning making similarity detection far more robust. Let’s talk about a Real world Needs Consider an AI driven content moderation system that should detect inappropriate images across billions of uploads daily. If a flagged image is slightly altered/cropped, filtered or resized traditional hashing fails allowing harmful content to evade detection. Now imagine an e-commerce search engine that helps customers find similar products. Two shirts with different backgrounds but identical designs might have very different pixel values but should still be recognized as the same product. In both cases a GNN-based approach would outperform traditional hashing by understanding relationships beyond pixels using graph structures to connect similar items dynamically. Building an Image Similarity System with Graph Neural Networks Let’s walk through a tutorial to build GNNs on any image set. Extract Image Features Using CLIP or ResNet:
Before constructing a graph, we extract feature embeddings from images using a pre-trained model import torch
import torchvision.transforms as transforms
from PIL import Image
from torchvision import models

#Load a pre-trained model
model = models.resnet50(pretrained=True)
model.fc = torch.nn.Identity()
model.eval()

#Image transformation
transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor()
])

#Load and preprocess an image
def extract_features(image_path):
    image = Image.open(image_path).convert('RGB')
    image = transform(image).unsqueeze(0)  #Add batch dimension
    with torch.no_grad():
        features = model(image)
    return features.numpy().flatten()

feature_vector = extract_features("my_image.jpg") Construct a Graph from Image Embeddings
We can now construct a KNN (K Nearest Neighbor) graph where each node represents an image and edges indicate similarity. import networkx as nx
import numpy as np
from sklearn.neighbors import NearestNeighbors

#Simulated feature vectors for 100 images (assuming 512D embeddings)
image_embeddings = np.random.rand(100, 512)

#Build a KNN graph
k = 5
nbrs = NearestNeighbors(n_neighbors=k, metric="cosine").fit(image_embeddings)
distances, indices = nbrs.kneighbors(image_embeddings)

#Construct the graph
G = nx.Graph()
for i in range(len(image_embeddings)):
    for j in range(1, k):  # Skip self-loop
        G.add_edge(i, indices[i, j], weight=1 - distances[i, j])

#Visualize graph structure
print("Graph nodes:", G.number_of_nodes())
print("Graph edges:", G.number_of_edges()) Train a Graph Neural Network (GNN) for Similarity Learning We now use PyTorch Geometric to train a GraphSAGE model on this graph. import torch
import torch.nn.functional as F
from torch_geometric.nn import SAGEConv
from torch_geometric.data import Data

#Convert the NetworkX graph to PyTorch Geometric format
edge_index = torch.tensor(list(G.edges)).t().contiguous()
x = torch.tensor(image_embeddings, dtype=torch.float)

#Define a GraphSAGE model
class GraphSAGE(torch.nn.Module):
    def __init__(self, in_channels, hidden_channels, out_channels):
        super().__init__()
        self.conv1 = SAGEConv(in_channels, hidden_channels)
        self.conv2 = SAGEConv(hidden_channels, out_channels)
    def forward(self, x, edge_index):
        x = self.conv1(x, edge_index)
        x = F.relu(x)
        x = self.conv2(x, edge_index)
        return x

#Initialize and train model
model = GraphSAGE(512, 256, 128)  #Reduce embedding size to 128D
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)

for epoch in range(50):
    optimizer.zero_grad()
    out = model(x, edge_index)
    loss = F.mse_loss(out, x)  #Self-supervised learning
    loss.backward()
    optimizer.step()
    print(f"Epoch {epoch+1}, Loss: {loss.item()}") Querying Similar Images Using the GNN Embeddings Now that training is complete, we can retrieve the top k most similar images for a given query. from sklearn.metrics.pairwise import cosine_similarity

#Extract GNN embeddings
with torch.no_grad():
    learned_embeddings = model(x, edge_index).numpy()

#Compute cosine similarity
query_idx = 0  # Example query image
similarities = cosine_similarity([learned_embeddings[query_idx]], learned_embeddings)
top_k = np.argsort(-similarities[0])[:5]

print("Most similar images:", top_k) Final Thoughts GNNs create a smarter and more adaptive approach to image similarity. Unlike hashing, which relies on rigid pixel comparisons, GNNs learn meaningful relationships between images, making them far more robust to transformations, occlusions, and noise. In today’s world where image datasets are exploding in size from social media platforms to medical imaging archives and the rise of synthetic data, scalability and accuracy are critical. GNNs provide an efficient way to structure and retrieve similar images improving applications like visual search, content moderation, recommendation engines and AI driven media management. The future of image search isn’t just about finding duplicates, it’s about understanding the deeper connections between images, and GNNs take us one step forward to that a reality. Traditional hashing techniques like perceptual hashing (pHash) and locality-sensitive hashing (LSH) are widely used for image similarity detection. However, they often fail in real-world scenarios where images undergo transformations such as cropping, noise addition or color changes. Graph Neural Networks: A Smarter Approach to Image Similarity Graph Neural Networks: A Smarter Approach to Image Similarity Graph Neural Networks (GNNs) provide a more robust alternative by modeling image relationships as a graph and propagating similarity information through message passing. Instead of treating images as isolated entities , Graph Neural Networks (GNNs) represent images as nodes in a graph where edges define their relationships. This allows context aware learning making similarity detection far more robust . isolated entities Graph Neural Networks (GNNs) nodes in a graph context aware learning far more robust Let’s talk about a Real world Needs Consider an AI driven content moderation system that should detect inappropriate images across billions of uploads daily. If a flagged image is slightly altered /cropped, filtered or resized traditional hashing fails allowing harmful content to evade detection . content moderation system slightly altered traditional hashing fails harmful content to evade detection Now imagine an e-commerce search engine that helps customers find similar products. Two shirts with different backgrounds but identical designs might have very different pixel values but should still be recognized as the same product. e-commerce search engine identical designs very different pixel values In both cases a GNN-based approach would outperform traditional hashing by understanding relationships beyond pixels using graph structures to connect similar items dynamically . GNN-based approach understanding relationships beyond pixels connect similar items dynamically Building an Image Similarity System with Graph Neural Networks Let’s walk through a tutorial to build GNNs on any image set. Extract Image Features Using CLIP or ResNet:
Before constructing a graph, we extract feature embeddings from images using a pre-trained model Extract Image Features Using CLIP or ResNet: Before constructing a graph, we extract feature embeddings from images using a pre-trained model Extract Image Features Using CLIP or ResNet: Extract Image Features Using CLIP or ResNet: Before constructing a graph, we extract feature embeddings from images using a pre-trained model import torch
import torchvision.transforms as transforms
from PIL import Image
from torchvision import models

#Load a pre-trained model
model = models.resnet50(pretrained=True)
model.fc = torch.nn.Identity()
model.eval()

#Image transformation
transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor()
])

#Load and preprocess an image
def extract_features(image_path):
    image = Image.open(image_path).convert('RGB')
    image = transform(image).unsqueeze(0)  #Add batch dimension
    with torch.no_grad():
        features = model(image)
    return features.numpy().flatten()

feature_vector = extract_features("my_image.jpg") import torch
import torchvision.transforms as transforms
from PIL import Image
from torchvision import models

#Load a pre-trained model
model = models.resnet50(pretrained=True)
model.fc = torch.nn.Identity()
model.eval()

#Image transformation
transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor()
])

#Load and preprocess an image
def extract_features(image_path):
    image = Image.open(image_path).convert('RGB')
    image = transform(image).unsqueeze(0)  #Add batch dimension
    with torch.no_grad():
        features = model(image)
    return features.numpy().flatten()

feature_vector = extract_features("my_image.jpg") Construct a Graph from Image Embeddings
We can now construct a KNN (K Nearest Neighbor) graph where each node represents an image and edges indicate similarity. Construct a Graph from Image Embeddings We can now construct a KNN (K Nearest Neighbor) graph where each node represents an image and edges indicate similarity. Construct a Graph from Image Embeddings Construct a Graph from Image Embeddings We can now construct a KNN (K Nearest Neighbor) graph where each node represents an image and edges indicate similarity. import networkx as nx
import numpy as np
from sklearn.neighbors import NearestNeighbors

#Simulated feature vectors for 100 images (assuming 512D embeddings)
image_embeddings = np.random.rand(100, 512)

#Build a KNN graph
k = 5
nbrs = NearestNeighbors(n_neighbors=k, metric="cosine").fit(image_embeddings)
distances, indices = nbrs.kneighbors(image_embeddings)

#Construct the graph
G = nx.Graph()
for i in range(len(image_embeddings)):
    for j in range(1, k):  # Skip self-loop
        G.add_edge(i, indices[i, j], weight=1 - distances[i, j])

#Visualize graph structure
print("Graph nodes:", G.number_of_nodes())
print("Graph edges:", G.number_of_edges()) import networkx as nx
import numpy as np
from sklearn.neighbors import NearestNeighbors

#Simulated feature vectors for 100 images (assuming 512D embeddings)
image_embeddings = np.random.rand(100, 512)

#Build a KNN graph
k = 5
nbrs = NearestNeighbors(n_neighbors=k, metric="cosine").fit(image_embeddings)
distances, indices = nbrs.kneighbors(image_embeddings)

#Construct the graph
G = nx.Graph()
for i in range(len(image_embeddings)):
    for j in range(1, k):  # Skip self-loop
        G.add_edge(i, indices[i, j], weight=1 - distances[i, j])

#Visualize graph structure
print("Graph nodes:", G.number_of_nodes())
print("Graph edges:", G.number_of_edges()) Train a Graph Neural Network (GNN) for Similarity Learning Train a Graph Neural Network (GNN) for Similarity Learning Train a Graph Neural Network (GNN) for Similarity Learning We now use PyTorch Geometric to train a GraphSAGE model on this graph. import torch
import torch.nn.functional as F
from torch_geometric.nn import SAGEConv
from torch_geometric.data import Data

#Convert the NetworkX graph to PyTorch Geometric format
edge_index = torch.tensor(list(G.edges)).t().contiguous()
x = torch.tensor(image_embeddings, dtype=torch.float)

#Define a GraphSAGE model
class GraphSAGE(torch.nn.Module):
    def __init__(self, in_channels, hidden_channels, out_channels):
        super().__init__()
        self.conv1 = SAGEConv(in_channels, hidden_channels)
        self.conv2 = SAGEConv(hidden_channels, out_channels)
    def forward(self, x, edge_index):
        x = self.conv1(x, edge_index)
        x = F.relu(x)
        x = self.conv2(x, edge_index)
        return x

#Initialize and train model
model = GraphSAGE(512, 256, 128)  #Reduce embedding size to 128D
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)

for epoch in range(50):
    optimizer.zero_grad()
    out = model(x, edge_index)
    loss = F.mse_loss(out, x)  #Self-supervised learning
    loss.backward()
    optimizer.step()
    print(f"Epoch {epoch+1}, Loss: {loss.item()}") import torch
import torch.nn.functional as F
from torch_geometric.nn import SAGEConv
from torch_geometric.data import Data

#Convert the NetworkX graph to PyTorch Geometric format
edge_index = torch.tensor(list(G.edges)).t().contiguous()
x = torch.tensor(image_embeddings, dtype=torch.float)

#Define a GraphSAGE model
class GraphSAGE(torch.nn.Module):
    def __init__(self, in_channels, hidden_channels, out_channels):
        super().__init__()
        self.conv1 = SAGEConv(in_channels, hidden_channels)
        self.conv2 = SAGEConv(hidden_channels, out_channels)
    def forward(self, x, edge_index):
        x = self.conv1(x, edge_index)
        x = F.relu(x)
        x = self.conv2(x, edge_index)
        return x

#Initialize and train model
model = GraphSAGE(512, 256, 128)  #Reduce embedding size to 128D
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)

for epoch in range(50):
    optimizer.zero_grad()
    out = model(x, edge_index)
    loss = F.mse_loss(out, x)  #Self-supervised learning
    loss.backward()
    optimizer.step()
    print(f"Epoch {epoch+1}, Loss: {loss.item()}") Querying Similar Images Using the GNN Embeddings Querying Similar Images Using the GNN Embeddings Querying Similar Images Using the GNN Embeddings Now that training is complete, we can retrieve the top k most similar images for a given query. from sklearn.metrics.pairwise import cosine_similarity

#Extract GNN embeddings
with torch.no_grad():
    learned_embeddings = model(x, edge_index).numpy()

#Compute cosine similarity
query_idx = 0  # Example query image
similarities = cosine_similarity([learned_embeddings[query_idx]], learned_embeddings)
top_k = np.argsort(-similarities[0])[:5]

print("Most similar images:", top_k) from sklearn.metrics.pairwise import cosine_similarity

#Extract GNN embeddings
with torch.no_grad():
    learned_embeddings = model(x, edge_index).numpy()

#Compute cosine similarity
query_idx = 0  # Example query image
similarities = cosine_similarity([learned_embeddings[query_idx]], learned_embeddings)
top_k = np.argsort(-similarities[0])[:5]

print("Most similar images:", top_k) Final Thoughts GNNs create a smarter and more adaptive approach to image similarity. Unlike hashing, which relies on rigid pixel comparisons, GNNs learn meaningful relationships between images, making them far more robust to transformations, occlusions, and noise. In today’s world where image datasets are exploding in size from social media platforms to medical imaging archives and the rise of synthetic data, scalability and accuracy are critical. GNNs provide an efficient way to structure and retrieve similar images improving applications like visual search, content moderation, recommendation engines and AI driven media management. The future of image search isn’t just about finding duplicates, it’s about understanding the deeper connections between images, and GNNs take us one step forward to that a reality.

Graph Neural Networks for Image Similarity: An Alternative to Hashing?

If you’ve been curious about OpenAI’s Operator but hesitated at the price tag of a $200 monthly fee, we’ve got great news. Meet Open-CUAK— an open-source platform that puts you in full control of automation agents, minus the high costs. Introducing Open-CUAK: Scalable Automation, Fully Local We’re thrilled to introduce Open-CUAK—a powerful, open-source platform for managing and deploying automation agents at scale. Unlike closed SaaS platforms, Open-CUAK runs locally, leveraging any vision-compatible frontier model to execute automation tasks seamlessly (frontier or open-source). ⭐ Give us a star on GitHub! ⭐ Why Open-CUAK? Automation in the real world demands reliability, scalability, and cost-efficiency. That’s why we built Open-CUAK (pronounced "quack" 🦆🗣️), the Kubernetes for Computer Use Agents (CUA)—allowing businesses to train, deploy, and manage AI-driven automation agents effortlessly. Unlike those developer-centric automation projects, Open-CUAK is designed for end-users—working out-of-the-box with minimal setup. It offers a user-friendly interface that allows non-developers to automate workflows easily, without needing deep technical expertise. What Can Open-CUAK Do? ✅ Run Operator-like workflows locally for full privacy and security. ✅ Enable vision-based automation with human-like accuracy. ✅ Turn any browser into an Operator-companion via a lightweight extension. ✅ Use a dedicated remote browser to enhance security and bypass restrictions. ✅ Integrate with any vision-compatible model, including Claude, Gemini, LLaVA, and open-source alternatives. ✅ Overcome bot detection mechanisms, unlocking powerful automation potential. 🔜 Train agents using SOP-based workflows, ensuring reliability and adaptability. 🔜 Centralize account access management, giving you granular control over agent permissions. ⏳ Monitor large-scale operations with built-in observability tools. ⏳ Deploy and scale hundreds of agents in parallel to execute real-world tasks. ⏳ Open-source an RL-trained CUA model, making automation more accessible than ever. Open-Source, Customizable & Ready to Deploy Open-CUAK makes automation open, accessible, and adaptable. You can now: Run Open-CUAK locally, eliminating expensive SaaS fees.
Customize workflows to match your industry-specific automation needs.
Experiment with different vision models, fine-tuning automation agents for maximum efficiency. Demo Videos 🎥 Agent uses Canva to design a poster An automation agent uses Canva to create a poster for President’s Day. When it encounters issues, such as whether to use Pro templates (which require a Canva paid plan), it proactively asks the user for additional instructions. https://www.youtube.com/watch?v=uMJ98ym5tvE&embedable=true 🎥 Agent searches for a flight ticket Uses the built-in remote browser running locally to search for flight tickets on Expedia. https://youtu.be/0iKFBHZySyc?si=_z10i604vhRQZxWd&embedable=true Get Started Today 🚀 “When automation is reliable, it becomes scalable. And when it becomes scalable, it becomes profitable.” 🔗 Try Open-CUAK now → GitHub Repo 👥 Join the community → Discord, X, Youtube 💬 Need help? → Connect with us on Discord & Twitter/X! Let’s build the future of AI-powered automation—together. 🦆🗣️🗣️🗣️ If you’ve been curious about OpenAI’s Operator but hesitated at the price tag of a $200 monthly fee, we’ve got great news. Meet Open-CUAK — an open-source platform that puts you in full control of automation agents, minus the high costs. Open-CUAK Introducing Open-CUAK: Scalable Automation, Fully Local We’re thrilled to introduce Open-CUAK —a powerful, open-source platform for managing and deploying automation agents at scale. Unlike closed SaaS platforms, Open-CUAK runs locally, leveraging any vision-compatible frontier model to execute automation tasks seamlessly (frontier or open-source). Open-CUAK ⭐ Give us a star on GitHub ! ⭐ GitHub Why Open-CUAK? Automation in the real world demands reliability, scalability, and cost-efficiency. That’s why we built Open-CUAK (pronounced "quack" 🦆🗣️), the Kubernetes for Computer Use Agents (CUA) —allowing businesses to train, deploy, and manage AI-driven automation agents effortlessly. Open-CUAK Kubernetes for Computer Use Agents (CUA) Unlike those developer-centric automation projects, Open-CUAK is designed for end-users —working out-of-the-box with minimal setup. It offers a user-friendly interface that allows non-developers to automate workflows easily, without needing deep technical expertise. Open-CUAK is designed for end-users user-friendly interface What Can Open-CUAK Do? ✅ Run Operator-like workflows locally for full privacy and security. Run Operator-like workflows locally ✅ Enable vision-based automation with human-like accuracy. Enable vision-based automation ✅ Turn any browser into an Operator-companion via a lightweight extension. Turn any browser into an Operator-companion ✅ Use a dedicated remote browser to enhance security and bypass restrictions. Use a dedicated remote browser ✅ Integrate with any vision-compatible model , including Claude, Gemini, LLaVA, and open-source alternatives. Integrate with any vision-compatible model ✅ Overcome bot detection mechanisms , unlocking powerful automation potential. Overcome bot detection mechanisms 🔜 Train agents using SOP-based workflows , ensuring reliability and adaptability. Train agents using SOP-based workflows 🔜 Centralize account access management , giving you granular control over agent permissions. Centralize account access management ⏳ Monitor large-scale operations with built-in observability tools. Monitor large-scale operations ⏳ Deploy and scale hundreds of agents in parallel to execute real-world tasks. Deploy and scale hundreds of agents in parallel ⏳ Open-source an RL-trained CUA model , making automation more accessible than ever. Open-source an RL-trained CUA model Open-Source, Customizable & Ready to Deploy Open-CUAK makes automation open, accessible, and adaptable . You can now: open, accessible, and adaptable Run Open-CUAK locally, eliminating expensive SaaS fees. Customize workflows to match your industry-specific automation needs. Experiment with different vision models, fine-tuning automation agents for maximum efficiency. Run Open-CUAK locally , eliminating expensive SaaS fees. Run Open-CUAK locally Customize workflows to match your industry-specific automation needs. Customize workflows Experiment with different vision models , fine-tuning automation agents for maximum efficiency. Experiment with different vision models Demo Videos 🎥 Agent uses Canva to design a poster Agent uses Canva to design a poster An automation agent uses Canva to create a poster for President’s Day. When it encounters issues, such as whether to use Pro templates (which require a Canva paid plan), it proactively asks the user for additional instructions. https://www.youtube.com/watch?v=uMJ98ym5tvE&embedable=true https://www.youtube.com/watch?v=uMJ98ym5tvE&embedable=true 🎥 Agent searches for a flight ticket Agent searches for a flight ticket Uses the built-in remote browser running locally to search for flight tickets on Expedia. https://youtu.be/0iKFBHZySyc?si=_z10i604vhRQZxWd&embedable=true https://youtu.be/0iKFBHZySyc?si=_z10i604vhRQZxWd&embedable=true Get Started Today 🚀 “When automation is reliable, it becomes scalable. And when it becomes scalable, it becomes profitable.” “When automation is reliable, it becomes scalable. And when it becomes scalable, it becomes profitable.” 🔗 Try Open-CUAK now → GitHub Repo Try Open-CUAK now GitHub Repo 👥 Join the community → Discord , X , Youtube Join the community Discord X Youtube 💬 Need help? → Connect with us on Discord & Twitter/X ! Need help? Discord Twitter/X Let’s build the future of AI-powered automation— together . 🦆🗣️🗣️🗣️ together

Open-CUAK: The Open-Source Alternative to OpenAI’s Operator

Singapore, February 19: Mira Network, a pioneer in AI verification technology, is pleased to announce the launch of Klok, a multi-LLM chat app built on its decentralised infrastructure that ensures AI outputs are verified, unbiased and uncompromised. Klok will allow users to participate in multiple AI models such as GPT-4o mini, Llama 3.3 70B Instruct, and DeepSeek-R1 for free (up to a daily limit). More AI models will be added to the app every week. Users will also be able to gain Mira Points that will be rewarded by Mira in the future, by using Klok everyday. Users can unlock ‘Klok PRO’ through referrals that will enable them to enjoy the benefits of higher rate limits on each AI model and advanced features such as search and multimodal. By earning points and engaging in AI features, Mira’s aim is to ensure AI tools are inclusive, accessible, and participatory. “We are thrilled to launch Klok at this pivotal moment in the evolution of AI technology. Powered by our infrastructure, Klok is poised to be a game-changer in the industry as it verifies outputs from multiple LLMs before they are executed. Any output that is not verified is discarded and regenerated,” said Karan Sirdesai, Co-founder and CEO of Mira Network. The inclusion of multiple LLMs within a decentralised infrastructure is vital in addressing the risk of AI hallucinations. AI hallucinations and biases result in false, incorrect or misleading AI outputs that can have significant consequences in real-world applications. For instance, if an AI agent, that relies on the analytical capabilities of AI models to perform tasks, “hallucinates” the contract address while swapping cryptocurrencies, it could lead to huge financial losses for the user. Mira is designed as a purpose-built L1 where each AI model operates as a trustless independent node. By using consensus mechanisms among multiple LLMs, Mira is establishing the foundational backbone where developers are building innovative applications and empowering users with AI models that are reliable, efficient and secure. “The latest launch presents a significant milestone for Mira and its mission to build the trust layer for all in AI and crypto. With Klok, we are setting the stage for a future where AI operates transparently for all,” said Sirdesai. Mira has raised over $9 million from prominent venture capital firms such as Framework, Accel, Mechanism and Bitkraft, as well as angels like Balaji Srinivasan and Sandeep Nailwal. The multi-LLM chat app is a part of Mira’s testnet that introduced a suite of APIs, including Generate, Verify, and Verified Generate, enabling distributed verification and access to top AI models like GPT-4o and Llama 3.1 405B. The latest launch of Klok follows the launch of Wiki Sentry, an AI agent that fact-checks Wikipedia against verified sources, built on Mira’s decentralized infrastructure. About Mira Network Mira is building the foundational verification layer that enables trustless AI systems through advanced consensus mechanisms. The network combines sophisticated claim transformation and distributed verification protocols to achieve reliable AI execution at scale. With over 500,000 active users and multiple production deployments, Mira has emerged as a pioneer in AI verification technology. By solving the fundamental challenges of error rates and complex reasoning verification, Mira is establishing new standards for AI reliability—paving the way for truly autonomous AI systems that can operate without human oversight. Follow Mira on X and Join their Discordfor real-time updates.This press release is brought to you by EAK Wire, the leading__Web3 PR Newswire__. Singapore, February 19: Mira Network, a pioneer in AI verification technology, is pleased to announce the launch of Klok, a multi-LLM chat app built on its decentralised infrastructure that ensures AI outputs are verified, unbiased and uncompromised. Singapore, February 19: Mira Network , a pioneer in AI verification technology, is pleased to announce the launch of Klok, a multi-LLM chat app built on its decentralised infrastructure that ensures AI outputs are verified, unbiased and uncompromised. Mira Network Mira Network Klok will allow users to participate in multiple AI models such as GPT-4o mini, Llama 3.3 70B Instruct, and DeepSeek-R1 for free (up to a daily limit). More AI models will be added to the app every week. Users will also be able to gain Mira Points that will be rewarded by Mira in the future, by using Klok everyday. Klok Klok Users can unlock ‘Klok PRO’ through referrals that will enable them to enjoy the benefits of higher rate limits on each AI model and advanced features such as search and multimodal. By earning points and engaging in AI features, Mira’s aim is to ensure AI tools are inclusive, accessible, and participatory. “We are thrilled to launch Klok at this pivotal moment in the evolution of AI technology. Powered by our infrastructure, Klok is poised to be a game-changer in the industry as it verifies outputs from multiple LLMs before they are executed. Any output that is not verified is discarded and regenerated,” said Karan Sirdesai, Co-founder and CEO of Mira Network. The inclusion of multiple LLMs within a decentralised infrastructure is vital in addressing the risk of AI hallucinations. AI hallucinations and biases result in false, incorrect or misleading AI outputs that can have significant consequences in real-world applications. For instance, if an AI agent, that relies on the analytical capabilities of AI models to perform tasks, “hallucinates” the contract address while swapping cryptocurrencies, it could lead to huge financial losses for the user. Mira is designed as a purpose-built L1 where each AI model operates as a trustless independent node. By using consensus mechanisms among multiple LLMs, Mira is establishing the foundational backbone where developers are building innovative applications and empowering users with AI models that are reliable, efficient and secure. “The latest launch presents a significant milestone for Mira and its mission to build the trust layer for all in AI and crypto. With Klok, we are setting the stage for a future where AI operates transparently for all,” said Sirdesai. Mira has raised over $9 million from prominent venture capital firms such as Framework, Accel, Mechanism and Bitkraft, as well as angels like Balaji Srinivasan and Sandeep Nailwal. The multi-LLM chat app is a part of Mira’s testnet that introduced a suite of APIs, including Generate, Verify, and Verified Generate, enabling distributed verification and access to top AI models like GPT-4o and Llama 3.1 405B. testnet testnet The latest launch of Klok follows the launch of Wiki Sentry , an AI agent that fact-checks Wikipedia against verified sources, built on Mira’s decentralized infrastructure. Wiki Sentry Wiki Sentry About Mira Network About Mira Network Mira is building the foundational verification layer that enables trustless AI systems through advanced consensus mechanisms. The network combines sophisticated claim transformation and distributed verification protocols to achieve reliable AI execution at scale. With over 500,000 active users and multiple production deployments, Mira has emerged as a pioneer in AI verification technology. By solving the fundamental challenges of error rates and complex reasoning verification, Mira is establishing new standards for AI reliability—paving the way for truly autonomous AI systems that can operate without human oversight. Follow Mira on X and Join their Discord for real-time updates. This press release is brought to you by EAK Wire, the leading__ Web3 PR Newswire __. X Discord Web3 PR Newswire

Mira Network Launches Klok: A ChatGPT Alternative with Multiple AI Models and Rewards

Mira Network ukax Klok ukar uñstayi: Mä ChatGPT Alternativa walja Modelos de AI ukat Premios ukanakampi

মিরা নেটওয়ার্ক চালু করেছে Klok: একাধিক AI মডেল এবং পুরষ্কার সহ একটি ChatGPT বিকল্প

Mira Network lansira Klok: ChatGPT alternativu s više AI modela i nagrada

Mira Network lancerer Klok: Et ChatGPT-alternativ med flere AI-modeller og belønninger

Mira Network lanza Klok: una alternativa a ChatGPT con múltiples modelos de inteligencia artificial y recompensas

Mira Network julkaisee Klok: ChatGPT-vaihtoehdon, jossa on useita AI-malleja ja palkintoja

Mira Network が Klok をリリース: 複数の AI モデルと報酬を備えた ChatGPT の代替手段

„Mira Network“ pristato „Klok“: „ChatGPT“ alternatyvą su keliais AI modeliais ir apdovanojimais

मीरा नेटवर्कले Klok सुरु गर्‍यो: धेरै एआई मोडेलहरू र पुरस्कारहरू सहितको च्याटजीपीटी विकल्प

Mira Llika Klok: Huk ChatGPT Alternativa achka AI Modelokunawan chaymanta Premiokunawan qallarichin

Mira Network запускает Klok: альтернативу ChatGPT с несколькими моделями ИИ и наградами

Inilunsad ng Mira Network ang Klok: Isang Alternatibong ChatGPT na may Maramihang AI Models at Rewards

Table of Links Abstract and 1 Introduction 2. Data 3. Measuring Media Slant and 3.1. Text pre-processing and featurization 3.2. Classifying transcripts by TV source 3.3. Text similarity between newspapers and TV stations and 3.4. Topic model 4. Econometric Framework 4.1. Instrumental variables specification 4.2. Instrument first stage and validity 5. Results 5.1. Main results 5.2. Robustness checks 6. Mechanisms and Heterogeneity 6.1. Local vs. national or international news content 6.2. Cable news media slant polarizes local newspapers 7. Conclusion and References Online Appendices A. Data Appendix A.1. Newspaper articles A.2. Alternative county matching of newspapers and A.3. Filtering of the article snippets A.4. Included prime-time TV shows and A.5. Summary statistics B. Methods Appendix, B.1. Text pre-processing and B.2. Bigrams most predictive for FNC or CNN/MSNBC B.3. Human validation of NLP model B.4. Distribution of Fox News similarity in newspapers and B.5. Example articles by Fox News similarity B.6. Topics from the newspaper-based LDA model C. Results Appendix C.1. First stage results and C.2. Instrument exogeneity C.3. Placebo: Content similarity in 1995/96 C.4. OLS results C.5. Reduced form results C.6. Sub-samples: Newspaper headquarters and other counties and C.7. Robustness: Alternative county matching C.8. Robustness: Historical circulation weights and C.9. Robustness: Relative circulation weights C.10. Robustness: Absolute and relative FNC viewership and C.11. Robustness: Dropping observations and clustering C.12. Mechanisms: Language features and topics C.13. Mechanisms: Descriptive Evidence on Demand Side C.14. Mechanisms: Slant contagion and polarization A.2. Alternative county matching of newspapers For robustness, we also apply an alternative matching procedure that covers more newspapers but only uses total instead of county-specific circulation. First, we obtain the main county for each newspaper outlet based on the newspaper name and geographical information provided by NewsLibrary (e.g., The Call (Woonsocket, RI) or the Albany Democrat-Herald (OR)), the U.S. Newspaper Directory, or a manual web search. For the circulation, we use more broad-based but less granular data: we assign total circulation (as of 2004) according to the Inter-University Consortium for Political and Social Research (ICPSR) to this main county. Hence, each newspaper is only assigned to one county, where its total circulation is assumed to accrue. This matching approach produces a dataset of 682 unique outlets and 24 million article snippets. As Table C.8 shows, with the alternative matching, the coefficients are significant and three to four times larger than in the main Table 2. A.3. Filtering of the article snippets Table A.1 gives an overview of the number of articles collected and how we obtain the number of articles used in our main analyses and robustness checks. This paper is available on arxiv under CC 4.0 license. Authors:
(1) Philine Widmer, ETH Zürich and philine.widmer@gess.ethz.ch;
(2) Sergio Galletta, ETH Zürich and sergio.galletta@gess.ethz.ch;
(3) Elliott Ash, ETH Zürich and ashe@ethz.ch. Table of Links Abstract and 1 Introduction 2. Data Abstract and 1 Introduction 2. Data 3. Measuring Media Slant and 3.1. Text pre-processing and featurization 3. Measuring Media Slant and 3.1. Text pre-processing and featurization 3.2. Classifying transcripts by TV source 3.2. Classifying transcripts by TV source 3.3. Text similarity between newspapers and TV stations and 3.4. Topic model 3.3. Text similarity between newspapers and TV stations and 3.4. Topic model 4. Econometric Framework 4.1. Instrumental variables specification 4.1. Instrumental variables specification 4.2. Instrument first stage and validity 4.2. Instrument first stage and validity 5. Results 5.1. Main results 5.1. Main results 5.2. Robustness checks 5.2. Robustness checks 6. Mechanisms and Heterogeneity 6.1. Local vs. national or international news content 6.1. Local vs. national or international news content 6.2. Cable news media slant polarizes local newspapers 6.2. Cable news media slant polarizes local newspapers 7. Conclusion and References 7. Conclusion and References Online Appendices Online Appendices A. Data Appendix A.1. Newspaper articles A.1. Newspaper articles A.2. Alternative county matching of newspapers and A.3. Filtering of the article snippets A.2. Alternative county matching of newspapers and A.3. Filtering of the article snippets A.4. Included prime-time TV shows and A.5. Summary statistics A.4. Included prime-time TV shows and A.5. Summary statistics B. Methods Appendix, B.1. Text pre-processing and B.2. Bigrams most predictive for FNC or CNN/MSNBC B. Methods Appendix, B.1. Text pre-processing and B.2. Bigrams most predictive for FNC or CNN/MSNBC B.3. Human validation of NLP model B.3. Human validation of NLP model B.4. Distribution of Fox News similarity in newspapers and B.5. Example articles by Fox News similarity B.4. Distribution of Fox News similarity in newspapers and B.5. Example articles by Fox News similarity B.6. Topics from the newspaper-based LDA model B.6. Topics from the newspaper-based LDA model C. Results Appendix C.1. First stage results and C.2. Instrument exogeneity C.1. First stage results and C.2. Instrument exogeneity C.3. Placebo: Content similarity in 1995/96 C.3. Placebo: Content similarity in 1995/96 C.4. OLS results C.4. OLS results C.5. Reduced form results C.5. Reduced form results C.6. Sub-samples: Newspaper headquarters and other counties and C.7. Robustness: Alternative county matching C.6. Sub-samples: Newspaper headquarters and other counties and C.7. Robustness: Alternative county matching C.8. Robustness: Historical circulation weights and C.9. Robustness: Relative circulation weights C.8. Robustness: Historical circulation weights and C.9. Robustness: Relative circulation weights C.10. Robustness: Absolute and relative FNC viewership and C.11. Robustness: Dropping observations and clustering C.10. Robustness: Absolute and relative FNC viewership and C.11. Robustness: Dropping observations and clustering C.12. Mechanisms: Language features and topics C.12. Mechanisms: Language features and topics C.13. Mechanisms: Descriptive Evidence on Demand Side C.13. Mechanisms: Descriptive Evidence on Demand Side C.14. Mechanisms: Slant contagion and polarization C.14. Mechanisms: Slant contagion and polarization A.2. Alternative county matching of newspapers For robustness, we also apply an alternative matching procedure that covers more newspapers but only uses total instead of county-specific circulation. First, we obtain the main county for each newspaper outlet based on the newspaper name and geographical information provided by NewsLibrary (e.g., The Call (Woonsocket, RI) or the Albany Democrat-Herald (OR)), the U.S. Newspaper Directory, or a manual web search. For the circulation, we use more broad-based but less granular data: we assign total circulation (as of 2004) according to the Inter-University Consortium for Political and Social Research (ICPSR) to this main county. Hence, each newspaper is only assigned to one county, where its total circulation is assumed to accrue. This matching approach produces a dataset of 682 unique outlets and 24 million article snippets. As Table C.8 shows, with the alternative matching, the coefficients are significant and three to four times larger than in the main Table 2. A.3. Filtering of the article snippets Table A.1 gives an overview of the number of articles collected and how we obtain the number of articles used in our main analyses and robustness checks. This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv Authors: (1) Philine Widmer, ETH Zürich and philine.widmer@gess.ethz.ch; (2) Sergio Galletta, ETH Zürich and sergio.galletta@gess.ethz.ch; (3) Elliott Ash, ETH Zürich and ashe@ethz.ch. Authors: Authors: (1) Philine Widmer, ETH Zürich and philine.widmer@gess.ethz.ch; (2) Sergio Galletta, ETH Zürich and sergio.galletta@gess.ethz.ch; (3) Elliott Ash, ETH Zürich and ashe@ethz.ch.

ALTERNATIVE

alternative.co

-- employees

Since n.d.

COMPANY RANKING

ALTERNATIVE

EVERGREEN INDEX #3411

Representing this company? Click here to claim and customize this page!

Alternative WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!

ALTERNATIVE

alternative.co

-- employees

Since n.d.

COMPANY RANKING

ALTERNATIVE

EVERGREEN INDEX #3411

RELATED COMPANIESRANK

STORIES ABOUT ALTERNATIVE

Representing this company? Click here to claim and customize this page!

Alternative WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!