AI and Blockchain, a new model and its expression in the distributed artificial intelligence

Blockchain and trust

Blockchain noise has been listed by artificial intelligence noise in the past two years. Each of these technologies is relatively new. Artificial intelligence has longer proportions, and return to a concept to Julim And human effects. Blockchain can be considered starting with retailer and distributed programming functions. Leslie Lambport Working on sewing systems together in time and confidence is necessary to resolve decentralized confidence and thus Blockchain. So at least 40 years plus for Blockchain and 80 plus for current forms of artificial intelligence.

Distributed computing to solve problems requires a temporal matter as well as a way to create a version of the truth from a group of computers, some of which can be wrong or harmful. Distributed computing and storage is the necessary condition for decentralization. Independent governance of distributed machines gives us decentralization. Decentralization depends on the nature and the spread of entities that control the infrastructure distributed for the calculation and storage. Through these measures, Bitcoin currency cannot be considered unanimous as 5 mining gatherings and a group of large institutions, including exchanges control the slopes to and from the Bitcoin ecosystem. Whales carry 93 % of bitcoin.

Amnesty International Challenges

The well -known problems in artificial intelligence include special data leakage, useless energy use, continuous training to reuse its own product, and the availability of private and private data to target detailed solutions and get a fee for private data used in training forms. Some of these problems can be solved by combining Blockchains into artificial intelligence. General challenges are drawn in the first section.

The article continues to describe the startup, Modelx.AI. Most quotations on how to perform it is an interview with Jamiel Sheikh, CEO of Modelx.ai. Jamiel states that its starting start is in the distributed camp, as the solutions in the unified artificial intelligence area.

Artificial intelligence is usually controlled by one entity. By artificial intelligence, we mean large language models based on deep learning (LLMS) similar to ChatgPT, generating images embodied in post -proliferation v1.5, sound to text and vice versa (text to sound), and Ultimate: video generation like Sora or Cinematic milk because it combines image, sound, etc. in the future, artificial intelligence has the ability to be “The country of geniuses in the data center”. The current training methods of Amnesty International require a huge amount of data. This data includes almost everything that is promoted and accessible. Wide amounts of data prevent excessive suitability. Excessive filling of the forces The model to be specialized in lowering the amount of data, and therefore cannot be accurately predicted. Open source models have broken this narration.

There are some problems with this heavy data approach. First, if the data that the model previously consumes is created by artificial intelligence, the tone and content of the data does not contain a slight difference and original content. Artificial intelligence begins to eat its own product and can decompose to bias and ineffective. A simple word derived from Greek, “automatic devastation” describes this phenomenon. Such development is notFantasy future A scenario and seen in the wilderness, where the amount of content resulting from artificial intelligence exploded. The second is the privacy of data. All data used to train artificial intelligence in the public field or available to the public even if the data was protected by copyrights. This includes discounting data from the gates that were not supposed to be used in this way, such as Reams of Youtube Videos or all the New York Times content.

Deepseek: AI Model Open

Developments like Dibsic A similar performance has shown without large quantities of data, the latest chips or time you spend in training. Time and calculation for reasoning (actual use) using Deepseek rises. Dibsic is also an open model.

The open model means that all the source code of the model is open. Moreover, this means that the typical weights are visible. Any person can modify, re -train or improve using their own data. this Definition by OSI He was to reply Since the training data should not be shared until the form is open source. According to For critics training data is the source symbol For artificial intelligence. Without sharing training data, it cannot be said that the model is open source. OSI defended their definition.

Modelx.AI

Modelx product depends on maintaining special form data, but the model and its weights are open. Treating privacy puzzle. It is how it is training the best Amnesty International that targets a specific field using the data available to the public only and still continues to improve the results, when some data is for law. In some areas such as health care, hospitals are prevented from sharing private data due to HIPAA rules. If you take one type of data, which is X -rays; Taking public artificial intelligence and training it further using the hospital X -rays in the hospital will improve the model. However, one of the best ways to use X -rays from a group of hospitals to improve the model is much more than if you are using only one hospital. Modelx.ai invented a way to share this data in a federal setting, without losing privacy.

The form is taken and then trained in Al -Ittihad hospitals. In other words, Hospital 1 takes a mature and coach open model and trains his own data. Hospital 1, then its trainee’s model is launched to the Federation and Hospital 2 trains the same artificial intelligence for their own data. This continues until all hospitals in the Federation train them. This refined artificial intelligence through the union’s special data is only available in the Union. Blockchain part if it is supposed to prove improvements and obtain salaries as shareholders as well as to maintain special data.

Typical weights are divided and placed on the Federal Professor’s book in each training step. Each hospital gets some symbols on the basis of the work they do. The quality of the model is also measured after each improvement, as you get a quality classification and symbolic quantities. Later when using models, each hospital is pushed into distinctive symbols based on use. When I heard about this in October 2024, the open source models were a family and did not get good results. With Dibsic arrived in February 2025, these criticisms lost their bite.

Another argument that training data can be extracted or extracted from the model using certain techniques that need additional safety precautions. This includes cleaning the special data for information that can be identified as well as other protection against Deanonymization technologies that are a lot of distortion. European Union Law, Amnesty International.