1. Click on your PFP

2. Click the settings cog wheel

3. Click storage

4. Buy more storage

That it, that's all you need to do to increase your storage on Warpcast!

Ciniz is no different than any other project here. Let's take a look at this mint from the above post and the associated whitepaper:

Collect NFT

Imagine a world where the power to shape the future of artificial intelligence is in your hands. A world where you don't need to be a tech giant or a wealthy corporation to make a difference and own a piece of the future. That's the vision behind Onchain Gaias (OGs) – a revolutionary platform that puts the development of Artificial General Intelligence (AGI) within reach of everyday people.

Sounds great, but thats about where it ends. Let's break down what this whitepaper is trying to tell us.

The world of AI has been dominated by a select few companies with deep pockets and vast computing resources. This concentration of power has created significant barriers for average people to participate and contribute their ideas. The rapid advancements in AI are largely controlled by and siloed within these tech giants, leaving everyday individuals with little say in how this transformative technology is developed and used. As a result, the future of AI – and by extension, the future of our world – is being shaped by the narrow interests of a few powerful corporations, rather than the collective wisdom and values of humanity as a whole.

If you're only thinking about OpenAI and ChatGPT, Google's Gemini then sure, you might think that Microsoft and Google have a near monopoly on AI. However, this couldn't be further from the truth, there are thousands of startups who are working on all different things within AI. Just because you are someone who doesn't actually understand what AI is (thinking it's just ChatGPT) then you have no conceptual idea of what is happening in the world of AI. The actual only truth in the above writing is that computing resources do require a lot of money.

But what if there was a way to level the playing field? What if everyone, everywhere, could contribute to the development of AGI and benefit from its progress? This is where OGs come in. By harnessing the global appeal of gaming and gambling, OGs creates a decentralized network where anyone can train sophisticated AI models, known as Gaias (General AI Agents), through the simple act of playing games. The platform operates on blockchain technology, ensuring fair, transparent, and secure computation. As the OGs network grows, it can leverage more advanced infrastructure to support increasingly complex games and AI models.

The problem here is that the author has absolutely NO understanding of how machine learning or deep learning models are trained. Sure, you could make a game, and you could have a deep learning model that is taking in patterns from how people play the game, and train it....essentially you're looking at something like Alpha Go, or a program that you play Chess against. But what good is this actually going to provide? By simply playing a game you're not contributing anything but training that specific game...there is no world changing / groundbreaking anything here.

As the OGs platform grows, it will introduce increasingly complex games and environments, fostering the development of more advanced Gaias. By enabling the transfer of learned skills and behaviors across different domains, OGs lays the groundwork for the emergence of artificial general intelligence. However, OGs isn't just about technical progress; it's about creating a more equitable and accessible future for AI development. By aligning incentives and fostering collaboration among a diverse global community, OGs challenges the current paradigm of centralized AI development, putting the power back in the hands of the people.

Again, you're not doing anything except giving data about patterns within the specific "game" you are playing. If it's chess, checkers, go, majong...whatever it is, you can't take this "training data" and transfer it to something else.

If you built a model to predict how far Babe Ruth would hit a home run, you would take into account all sorts of variables specific to Babe Ruth. Height, weight, bat weight, bat length, bat wood material, angle of attack (swing) etc. etc. If you tried to "transfer" this model to predict Mike Trout, Sammy Sosa, Barry Bonds or any other basebal legend...it wont do anything because you've overfit the model for Babe Ruth.

As for the rest of the whitepaper...it's pure snake oil, and people fall for it because they think AI is magic, when in reality is just a man behind the curtain and not some magical wizardry.

Historically speaking, halving did = moon.

In the lead up to the first halving the price of Bitcoin was around $12. Less than one year later it was up to $229 and then after that it was over $1,100. From there we saw Bitcoin get just shy of $20,000 after the 2nd halving. And after the third, the price was back down around $8500 but launched to $40,000 then $63,000 then $67,000. Check the charts below.

As we approach the next halving, where the reward goes from 6.25 BTC down to 3.125 BTC will we see the price rocket over $100,000 or higher in 2024 ?!

Personally, I don't think so and here is why:

• Crypto Is Mainstream - In 2024 crypto is about as mainstream as it gets.

  • Hundreds if not thousands of exchanges where you can buy and sell Bitcoin and thousands of other tokens on different blockchains.

  • Multiple publicly traded companies where you can buy Bitcoin and other tokens.

  • Multiple Bitcoin ETF's

• Not All Halvings are Equal - Going from 50-25 is not the same as 6.25 to 3.215

  • The impact of a halving is less and less over time. We'll never again experience the massive drop of 25 whole Bitcoins from a halving again.

• Time - The price mooning never happened right at the same time of the halving in the past.

  • So what would change this time? Typically, it was over a year after a halving occurred before the price really started to moon.

I've seen far too many people comment on the fact they think something magical is goin to happen on a day in the near future that just sends Bitcoin to Mars. But that doesn't make any sense. Why, if everyone knows about halving, would everyone wait until it happens in the next ~20-30 days to start buying? The accumulation has already happened, there really is no catalyst.

If I knew that for certain, I absolutely wouldn't be telling anyone, but in my opinion we'll be somewhere in the $40,000 - $50,000 range and not the $100,000 + that some would lead you to believe.

This is your AI metrics cheat sheet:

• Precision: Out of all the items the model labeled as positive, how many truly are?
  - Think of it as the model’s trustworthiness when it claims something is positive.

• Recall (Sensitivity): Out of all actual positive items, how many did the model correctly identify?
  - It’s about not missing out on true positives.

• F1 Score: The balance between Precision and Recall.
  - When you want a single metric that considers both false positives and false negatives.

• AUC-ROC: The model’s ability to distinguish between classes.
  - A score of 1.0 is perfect. Above 0.9 is excellent. Below 0.5 means the model is worse than random guessing.

• Mean Absolute Error (MAE): The average difference between predicted and actual values.
  - Lower is better. If MAE is zero, be cautious: it could indicate overfitting or other issues.

• Root Mean Squared Error (RMSE): Like MAE, but punishes large errors more.
  - Critical when significant mispredictions can have major consequences.

• R-squared: How well the model’s predictions match the real data.
  - Ranges from 0 to 1. Closer to 1 means the model explains more of the variability.

• Silhouette Score: Measures how similar items are within a cluster compared to other clusters.
  - Ranges from -1 to 1. Higher values indicate better-defined clusters.

• Davies-Bouldin Index: Lower values indicate better partitioning of clusters.
  - Zero is the theoretical ideal score.

• Completeness: Percentage of non-missing data points.
  - Aim for 100%, but be wary of artificially complete data.

• Consistency: Ensuring data doesn’t contradict itself.
  - Inconsistent data can lead to misleading model results.

• Outlier Detection: Identify data points that deviate significantly.
  - Outliers can skew model training.

• Cross-Validation Score: A measure of a model’s performance on different subsets of data.
  - Ensures the model isn’t just memorizing the training data.

• Bias-Variance Tradeoff: Balance between the model’s adaptability and its ability to generalize to new data.
  - Ideally a model that captures patterns without overfitting or being overly simplistic.

• Feature Importance: Identifies the input variables that have the most influence on the model’s predictions.
  - Understanding what the model deems important.

• SHAP Values: Breaks down the contribution of each feature to specific model predictions.
  - Clarifies the reasoning behind individual predictions.

• Training vs. Validation Error: A low training error coupled with a high validation error suggests potential overfitting.

• Overfitting: Model performs well on training data but poorly on new data.

• Underfitting: Model performs poorly on both training and new data.

This guide serves as a simple starting point for how to measure the effectiveness of different models!

Deep learning is the final piece of the diagram above. The first two posts in this series can be found here:

https://paragraph.xyz

Artificial Intelligence in 4 Minutes

What is AI?Artificial Intelligence refers to machines performing tasks that would normally require human intelligence. AI breaks down into two catego...

https://paragraph.xyz

Machine Learning in 3 Minutes

Machine Learning (ML) is a subset of artificial intelligence that involves teaching machines how to learn from data and improve their performance ove...

Deep Learning is a subset of machine learning that mimics the way a human brain works. The ‘deep’ in deep learning relates to the multiple layers in these neural networks that allow for complex and abstract representations of input data. It’s possible to have hundreds or even thousands of hidden layers to create a hierarchical or ‘deep’ structure. Neural networks are computational models inspired by how the human brain functions, specifically, how neurons within the brain process and transmit information.

A Neural Network has three main layers:

1. Input layer: where raw data is input.

2. Hidden layer(s): where the magic of learning occurs.

3. Output layer: makes a decision or prediction about the input data.

The network’s main goal is to adjust these connections (referred to as weights in deep learning) to minimize the difference between its prediction and the actual target — essentially a sophisticated trial and error process to make the network’s predictions as accurate as possible.

Deep learning algorithms have been applied to medical imaging to detect cancer cells with impressive accuracy. In one study, researchers trained a deep learning model on breast cancer pathology images and achieved better performance than pathologists.

Google Translate uses deep learning for language translation services. By training on vast amounts of multilingual data, these models can provide increasingly fluent translation services.

Tesla, Waymo, Cruise, and other self-driving vehicles utilize deep learning when processing real-time data from sensors (camera, lidar, radar). The algorithms can then identify objects, recognize road signs, and predict the behavior of other vehicles to create increasingly safe roads.

Deep learning algorithms are utilized to analyze vast amounts of patient data (medical records, genetic info, lifestyle factors) to predict risks and create personalized treatment plans. These models can assist doctors in making more accurate diagnoses, selecting appropriate therapies, and improving patient outcomes.

Deep Learning extends the principles of Machine Learning to more complex and large-scale tasks.

There are three main types: supervised learning, unsupervised learning, and reinforcement learning.

In supervised learning, machines learn from labeled data. That means we feed the algorithm a dataset with input-output pairs, and the algorithm learns to map the inputs to the correct outputs. Once trained, the algorithm can take a new set of inputs and predict the corresponding outputs.

An example of this would be a system that identifies and flags emails as spam. The machine learning algorithm learns to identify patterns associated with spam emails such as certain keywords, the email’s format, the sender’s email address, etc. After training the model can analyze new, never seen emails and predict if it’s spam or not.

The input-output pairs for this would look like:

Input:

• Subject: “Claim Your Free Bitcoin”

• Body: “You have a free Bitcoin waiting for you. Click here to connect your wallet.”

• Sender: spammer@example.com

• Recipient: yourname@gmail.com

Output: 1 (spam)

Input:

• Subject: “Meeting tomorrow”

• Body: “Quick reminder that we have a team meeting tomorrow at 10 am.”

• Sender: ceo@example.com

• Recipient: yourname@example.com

Output: 0 (not spam)

Unsupervised learning involves training machines on data where the output is unknown. The algorithm’s job is to find underlying patterns or structures in the data.

Netflix recommendation system is a good example of unsupervised learning, and while Netflix system is far more sophisticated than this example, this serves as a good baseline understanding. Netflix utilizes unsupervised learning to analyze user data such as: viewing history, ratings, and engagement signals to identify patterns and similarities among users and the movies they watch to create user clusters or segments.

Input:

• User A: Watched “Stranger Things,” rated it 5 stars

• User B: Watched “Narcos” and “Breaking Bad,” rated them 4 stars

• User C: Watched “The Crown” and “Downton Abbey,” rated them 5 stars

Output: User Clusters

• Cluster 1: Users like sci-fi and thriller genres

• Cluster 2: Users like crime and drama genres

• Cluster 3: Users like historical and period dramas

Through clusters the Netflix recommendation system can predict a user’s preferences based on the movies watched and rated by other users in the same cluster. For example:

Input:

• User D: Watched “Black Mirror”

Output: Recommendation

• Recommend “Stranger Things” (based on Cluster 1’s preferences)

The more a user interacts with the platform and generated more data, the system will continually update and refine the clusters, enhancing the accuracy of its recommendations and hopefully preventing this 👇 scenario.

Reinforcement learning is a bit different. It involves an agent learning to make decisions by interacting with its environment. The agent takes actions, gets feedback, and uses that feedback to make better decisions over time.

When applied to chess, the agent (AI player) interacts with the chessboard, receives rewards or penalties based on its moves, and adjusts its strategies to improve its gameplay over time.

Input:

• Chessboard State: The current state of the chessboard, including the positions of the pieces.

• AI Player’s Turn: Whether it’s the AI player’s turn to move or the opponent’s turn.

Output:

• Selected Move: The move chosen by the AI player based on the current state and its learned policy.

Throughout the training process, the AI player uses reinforcement learning algorithms to evaluate positions, search for optimal moves, and update its policy based on rewards obtained during gameplay.

What is AI?

Artificial Intelligence refers to machines performing tasks that would normally require human intelligence. AI breaks down into two categories, narrow AI and general AI. Narrow AI, also known as “weak AI” performs a specific task, or set of closely related tasks which would typically require human intelligence to accomplish.

Artificial General Intelligence (AGI) refers to machines that can perform any intellectual task a human can, and perhaps better. When Elon Musk talks about AI wiping out the human race (science-fiction), this is referring to a potential negative consequence that could happen if the conceptual idea of AGI becomes reality.

I. Natural Language Processing (NLP): Ask Siri or Google Assistant a question and that service is able to understand and interpret the question and generate a response in human language by using NLP.

II. Knowledge Representation: Retrieving knowledge in a manner that a machine can understand. IBM’s Watson uses advanced knowledge representation techniques with NLP to understand and answer questions across a wide range of topics. Watson intakes the question through NLP, processes the information using knowledge representation techniques, and then outputs the answer using NLP.

💡 What’s the difference between NLP and Knowledge Representation? 💡

Siri takes a question asked in spoken language, interprets the intent of the question, and then transforms it into text using NLP. One person might ask “What’s the weather?” while another might ask “What’s the temp?” and the system understands these are both requests for the current weather. A more complex question, “How many teams have come back from a 3–1 deficit in the NBA Finals to win?” (the answer is one, the Cleveland Cavaliers in 2016). Siri doesn’t know that information, or have that stored in an internal database. Instead, a search engine (Google) retrieves the top result and then returns a human-like response using NLP.

While Siri has some elements of knowledge representation, it’s not the same as in the Watson example. Watson applies NLP to understand the content and structure of the question asked. Using knowledge representation, Watson maps the question to potential answers in its database. Watson understands the context, infers meaning, understands the relationship between different pieces of information, and responds to complex questions with a depth and precision that Siri can not.

Ask Watson that same question, “How many teams have come back from a 3–1 deficit in the NBA Finals to win?” and Watson will analyze an internal database, consider the historical data of NBA games, the specific teams involved, the concept of a “3–1 deficit,” and so forth. Watson is able to provide an answer along with related context, such as the year the comeback occurred, the team who achieved it, and other details.

III. Learning (Machine Learning): Computers learning from data and improving their performance over time. Machine learning algorithms train on data and learn to make predictions or decisions without being specifically programmed for that task.

Instead of programming Watson with specific answers to each potential question, Watson trained on a vast amount of information from encyclopedias, books, research papers, and other data. Watson learned over time how to better answer questions by learning from mistakes, improving performance, and becoming more effective at playing Jeopardy.

IV. Reasoning and problem-solving: Enabling machines to make decisions and find solutions based on available information and rules by analyzing a given situation, understanding the relevant factors, and determining the best course of action.

AlphaGo was a computer that defeated the world champion Go player, Lee Sedol, in 2016. Using sophisticated reasoning and problem-solving algorithms, AlphaGo explored different possible moves, evaluated potential outcomes, and selected the move with the highest probability of success.

V. Social Intelligence: Computers understanding and interpreting human emotions, social cues, contexts, and responding in a way that is socially appropriate. Siri uses a basic form of social intelligence to interpret the intent behind user commands and respond in a way that mimics human conversation by using a combination of NLP, machine learning, and pre-programmed responses.

VI. Perception: Perception involves the ability to sense and understand the environment, mimicking human senses like vision (computer vision) or hearing (speech recognition). Robotics falls into this category as well, and a great example of that is Boston Dynamics. Their robots, like Atlas and Spot, use perception systems to interact with and navigate their surroundings. Using computer vision to analyze visual data from cameras and sensors, and perceive objects, obstacles, and terrain in real-time. Self-driving vehicles fall fit here as well. Tesla, Waymo, Cruise technology all have to perceive and understand their environment to function safely and efficiently while driving on the road. Here’s how they do it:

1. Input Collection: A self-driving vehicle uses a variety of sensors, including cameras, LIDAR, and RADAR to capture information about the surroundings. This could include detecting other vehicles, pedestrians, signs, road markings, and more. Sensors serve as the vehicle’s “eyes” and “ears,” capturing raw data about the environment.

2. Perception Processing: Raw visual data from a camera might identify and categorize different objects, like cars, pedestrians, or road signs. This process can also involve elements of Machine Learning, where the AI system learns from past data to better interpret current and future data.

3. Action Planning: Based on this processed perception data and its internal knowledge representation, the vehicle’s AI system decides what actions to take, accelerate, brake, turn, change lanes, etc. This is a form of reasoning and problem-solving, as the system needs to make decisions based on its understanding of the environment.

4. Actuation: Computers needs to manipulate physical hardware to carry out the planned action. Planned actions becoming physical actions. This is where the “robotics” comes into play.

Each component of narrow AI serves a specific function and plays a key role in the overall system. While certain examples may combine multiple components of narrow AI, such as perception and reasoning, it is important to note that these individual parts operate within their own domains and do not communicate or learn from each other in the same system. This clear separation and lack of interconnectivity distinguish narrow AI from general AI.