Recently, the data query website Dune announced the update of its LLM roadmap, the first on-line query explanations (Query Explanations) capabilities supported by GPT-4, and will gradually add more features, such as natural language querying (Natural Language Querying), SQL translation (Query Translations) and search optimization.

Unlike other paid query commercialization routes for data analytics sites, Dune is always open to ordinary users after its launch, so the DeFi Summer settles up in the round, and the LLM's accession is expected to turn its settled data query statements, signs into actual killers, and guides the ordinary user to join the creator's ranks, commercially outperforming a crowd of competitors.

Thanks to the openness and transparency of blockchain data, anyone can access blockchain data directly, but raw data (Raw Data) is often difficult to identify, and it is difficult for non-professional programmers to understand its meaning, but the data contain huge economic value, so all kinds of data analysis tools are created, providing tools for indirect access and in-depth analysis for various analysts, researchers and ordinary retailers.

Dune is the most notable of these because it offers truly free and powerful analytics tools where anyone can query, share and display data through SQL statements, and even some projects directly choose Dune as the official information display platform.

But the surface of Dune's SQL query function is UGC model, equally assigned to each user, but in reality Dune SQL model has two problems, one is the threshold too high, SQL is (structured query language) Structured Query Language abbreviation, once the logic of the query becomes complicated, its code volume may increase dramatically, non-professional programmers are difficult to write, which causes a large number of users can only be viewers; the second is the SQL standard supported between the Dune V1 and V2 versions is not unified, and the PostgreSQL and Spark SQL used separately in V1, the subsequent plan by DuneSQL to complete unification.

Prior to this LLM upgrade, Dune was already preparing for the unified query engine, and plans to move it all to Dune SQL after July to ensure product logic consistency. The updated Dune SQL is based on the implementation of the open source query engine Trino, which Dune optimized to suit Dune's own needs, with no substantial differences from the popular Sqprk SQL, more in specific functions and syntax improvements.

But the problem of the SQL query threshold is still not resolved by changing the SQL paradigm, which means that most users and programmers must adapt to the new syntax format, and the need for automation tools is not only realistic for ordinary users, but also for programmers who are not familiar with the new Syntax.

In fact, in the migration Dune SQL Roadmap, Dune has been trying to experiment with automated tools that can unify different syntax formats into Dune’s SQL statements, and after GPT-4 has used LLM to show the power of human-computer interaction, it has also successfully launched its own LLM program.

The typical process for Dune is to deconstruct data on the chain, where professional users perform queries through SQL, and then share the dashboard with users in need. The key in this process is to execute queries, whereas most users are unable to use query functions due to lack of code knowledge.

After the introduction of query interpretation function, the above process occurs some fine-tuning, in the professional user-written SQL query interface, there will be an interpretation page, in a natural language format intuitively tell us to see the specific role of the user code, equivalent to adding an explanation patch to theSQL query, and will not change the current workflow, which is also the lesson of the team in the consolidation of SQL sentences, that is, to reduce the interference to the user both in the habit, but as much as possible in the existing process optimization experience.

After LLM joined, the skill gap between professional users and ordinary users was deleted to some extent. Using GPT-4's ability to understand code, it allows ordinary users to intuitively understand the role of query statements without the need to master SQL knowledge. Before this feature is introduced, users can only passively read the chart on the dashboard page, and after the introduction of the query interpretation function, ordinary users can also understand the specific meaning of SQL code to play.

Essentially, the function of query interpretation is equivalent to a translation and translation between code and humans, then in turn, it can also be translated into SQL statements, with the LLM added, thanks to Dune's stored massive amount of questioning statements data, which is not difficult to.

Natural Language Querying is an important direction for Dune’s subsequent LLM improvement. Natural language queries can allow users to execute chart generation tasks in the way of communicating instructions, which is more in line with the common people's way of thinking than SQL statements, dragging generations and other patterns, avoiding attention to the realization of details. Users can issue instructions in English, Dune translates with LLM, and eventually decommand the command system through SQL to generate diagrams that meet the instructions requirements.

Moreover, natural language queries are not an alternative to professional users, such as analyst groups, but a reinforcement, the existing Dune has 700,000 charts, a significant part of the analytical tasks are overlapping and conflicting, and the use of natural-language queries also helps the system to understand the relationships between the charts and thus further improve the efficiency of the entire analytical work.

With the LLM joining, existing charts, SpellBooks, and document data will also be re-integrated to imitate OpenAI’s chatbots, and Dune will also develop conversation bots to help users understand and leverage existing knowledge systems in a simpler way without being disturbed by irrelevant information.

Everyone is a data analyst.

Dune’s goal is not to build a mere chain-based data analytics platform, but rather a data pipeline that allows information to flow freely, allowing users to grasp, convert, manage, query, visualize and use data to take action.

The prerequisite for data flow is to be modular, arbitrarily combined and configured to ultimately build a community-shared dataset, rather than concentrating within a complex SQL statement or a paid API, ultimately reaching an equal picture where everyone can interact with the data.

In summary, Dune’s LLM program is “translated” and assistant, with the goal of allowing the average user to read everything that data represents, not just the presentation of the end result, but deeper into the process of generating, and ultimately everyone can conduct data analysis on a chain.