Israeli artificial intelligence developer AI21Founded by Professor Aminon Chachua, Professor Yuof Shaheham and Uri Joshin, who works in the busy field in developing basic language models, presented MASTRO – the planning and coordination system of Amnesty International designed to connect artificial intelligence worthy of organizations.
Maestro is a new product and a new business model, similar to the software company as a SAAS service more than just the models maker. MASTRO is an artificial intelligence software system that manages all the organization's artificial intelligence and “organizing”, as the company says, in a manner that works harmoniously and reduces the rate of errors by re -analyzing data and operations. In other words, in addition to the developed JAMBA model, AI21 will try to improve the way in which the institutions work with the models created by competing companies.
AI21 says that MASTRO improves the accuracy of systems such as GPT-4O and Claude Sonnet 3.5 by up to 50 % on complex tasks. “The problem with organizations is that the current language and inference models are far from accurately that only 6 % of the projects that companies have established in this field see in the end daylight, and the rest is suspended due to hallucinations or contradictions.” Organizations are trying to treat hallucinations by improving the formulation of information or dividing the process into small sub -parts that can be searched at different points, but the result, according to Goshen, is still far from consistent result.
“In a world you can not trust in the directing you get from artificial intelligence, MASTRO offers you a basket of language models, artificial intelligence tools, data sources and application programming interfaces (APIS), which will bring it on the basis, and the system will attend the various models to produce an answer and say what is the level of reliability.”
Instead of using two or two models, companies can manage all the development of artificial intelligence applications from one place, a product that reflects the target customers that AI21 aims – and not the final consumer or financial manager in the institution, but artificial intelligence managers in organizations, known as artificial intelligence builders, software developers, or automatic learning experts or experts in information systems all bear the transactions AI, like it, as an automatic echo, such as ASEAN exercises, such as the cards, like them. Models.
Goshen explains, “We do not yet know any business model that will dominate the industrial intelligence industry, but our belief is that” artificial intelligence builders in the institution will play an important role there. Instead of each section raises its automated agent who will reach all relevant information with other computing resources, there will be the authority of individual information systems that will manage this resource, such as “Human Resources Management”. While in the age of agents, these expenses can grow to tens of the total expenses of the organization. “
Related articles
Israeli -Israeli AI Co21 laboratories are raising $ 53 million
Last week, AI21 JAMBA 1.6 presented the most advanced open model for the deployment of private institutions. The new model is outperforming the driving of the open models of Mistral, Meta and COHERE through multiple standards while maintaining superior speed, accuracy and security. The model enhances data classification by 26 percentage points on its predecessor, JAMBA 1.5, and ensures more than 90 % of consistency in answering the long context questions. Through flexible publishing options, including VPC and local tramp (self -hosted), JAMBA 1.6 offers full control of its data. By combining the strength of the logic of transformers and the efficiency of the status area models (SSMS), JAMBA 1.6 improves long -context treatment, providing advanced performance of Amnesty International for Institutions.
It was published by Globes, Israel Business News – En.globes.co.il – on March 11, 2025.
© Copy Publish Publisher Itonut (1983) Ltd. , 2025.