Artificial intelligence still has a hallucinogen

For the best possible result of Amnesty International, institutions need the best possible data.

The answer that many organizations were forced to overcome this challenge A generation for retrieval (rag). With a rag, the results are established in the data from a database. As it turned out, though, not every piece is the same, and improving the database to get the best possible results may be difficult.

Database Mongodb It is not strange for the world of artificial intelligence or rag. A database bearing the name of the company has already been used for RAG, and Mongodb was launched Development of artificial intelligence applications Initiatives. While the company and its users – such a medical giant Novo Nordsk – He achieved success with Gen AI, there is still more.

In particular, hallucinations and accuracy are still a problem that restores some organizations to the introduction of Gen AI. To this end, Mongodb announced today the acquisition of a private ownership Artificial Intelligence JourneyWhich develops advanced inclusion and recovery models. The journey that was raised 20 million dollars in financing in October 2024 in a round backed by a giant snowflake for cloud data. The Voyage AI experience will bring in the inclusion of generation and re-obstetrics-critical ingredients for research and retrieval that works with artificial intelligence materials-directly on the Mongodb database platform.

“Over the past year, especially with organizations trying to think about how to build applications supported by artificial intelligence, it has become increasingly clear that the quality of the applications they build, or their absence, has become one of the special barriers. AI on critical use cases. “

What are the challenges of hallucinations? Do you not get a rag?

The main idea behind the Rag is that instead of just relying on the knowledge base of trained data, the Gen AI engine can get data from a database.

The creation of RAT is very accurate very complex, and there is still a potential risk of hallucinations – a challenge facing Mongodb and its users. While AZAM refused to provide any specific example or accident as Gen Ai Rag failed in a user, he noticed that accuracy is always a source of concern.

Improving accuracy and reducing hallucinations involves multiple steps. The first is to improve the quality of retrieval (“R” in a rag).

“In many cases, the recovery quality is not good enough,” said Tengyu MA, founder and CEO of Voyage AI, for Venturebeat. “In the step of retrieval, if they do not recover relevant information, the retrieval is not very useful, and hallucinations of the big language model (LLM) because it should be guessed for some context.”

AI Voyage models now help Mongodb improve RAC in some of the main ways:

• Models for the field and re -rank: These are trained in large quantities of non -structured data from the specified voters, allowing them to better understand the terms and connotations of these areas.
• Specialization and control: Users can adjust the retrieval mechanism for unique data groups and cases of use.

Mongeodb competition

Mongodb is not the first or the only seller who realizes the need and the value of the presence of the technique of inclusion and the recipient significantly. After all, this is one of the reasons that have invested Snowflake in Voyage AI and uses company models.

It is important to note that, even after obtaining it by Mongodb, Voyage AI models will remain available for snow and other users in AI. The big difference is that Voyage AI will now be increasingly combined into Mongodb database platforms.

Merging advanced inclusion models directly into the database is an approach followed by other competitive database sellers as well. Once again in June 2024, its Datastax declared her Ragstac Technology That combines advanced inclusion and retrieval models.

Amzel argued that Mongodb is a little different, though. For one of them, it is a operating database, instead of an analytical database. Also, unlike just providing ideas and analysis, Mongodb helps energy transactions and real world processes. Mongodb is also what is known as the “document model database”, which has a different structure from the traditional relationship database. This structure does not depend on columns and tables, which are not especially good in representing information about non -structured data (a decisive component of artificial intelligence applications).

“We are the only database technology that combines descriptive data management about customer information, operations and transactions, which is the heartbeat of what is happening at work, as well as the basis for retrieving – all of this with one system,” said Azzam.

Why Voyage AI is important for Agency Ai Work

The need for models of inclusion and very accurate retrieval is accelerated by Aulctic AI.

“The artificial intelligence factor still needs recovery methods, because the agent cannot make decisions outside the context,” said what. “Sometimes, multiple recovery ingredients are actually used even in one decision.”

MA noticed that Voyage AI is currently working on specific models for AIC AIC. He explained that artificial intelligence agent can use different types of queries that can still benefit from more improvement.

Since Gen AI is increasingly transmitting to operational use, the need to remove hallucinations is clearly important. While Mongodb has been successful with Gen AI, AZAM expects Voyage AI to open the new critical use of the task.

“If we can now say,” hey, we can provide you well north of 90 % accuracy for your applications that are not permissible today, in some cases, to reach 30 or 60 % of the accuracy of the results, “the hole expands from where it is said.” Types of opportunities that individuals can apply to their software applications. “