Original - Odaily
Author - Nan Zhi
In early November,Web3 data analytics infrastructure Dune announces DuneAI, users can use its natural language engine to ask questions in any language and gain encrypted data insights without knowing SQL. Odaily will test Dune AI through multiple problems ranging from simple to complex to verify its usability and reliability.
Case 1 - Specific protocol transaction volume query
Question: Find the Blur volume in the for last 7 days, by day?
Compare the answers given by Dune AI with those on DuneMost Liked Blur query(Bottom right picture), the data comparison is as follows, there is a difference of nearly 1%.
Case 2 - Specific network transaction volume query
Question: How many transactions were processed today in arbitrum?
The picture on the left shows the answer given by Dune AI, and the picture on the right shows the query edited by Odaily. The results of the two are basically the same (the difference comes from the difference in query execution time).
Case 3 - Ambiguous query
Question: Total volume on Uniswap last week by day?
Here, three questions arise:
Dune AI limits the query scope to Ethereum, while Uniswap is deployed on multiple chains, and the original question does not have this requirement.
andUniswap on DefiLlamaComparing the data, the differences are relatively large. Taking the 24th to the 26th as an example, the differences were 6%, 9% and 11% respectively.
The original query here is the data from Monday to Sunday of the previous week, and the last week here is interpreted as the past seven days. For further verification, the query statement is modified to Uniswap daily trading volume of the previous complete week (The total volume on Uniswap for the previous full week by day), the results are as shown in the figure below, and the data still queried is the past seven days.
Case 4 - Complex query with multiple conditions
Question: The daily amount bridged between Ethereum and Arbitrum using Stargate over the past week?
The query result is empty. After Odaily checked the query, there were no problems with the overall query content and syntax. The reason for the query error wasWrong dataset selectedstargate_arbitrum.ProxyOFT_evt_SendToChain, this is a data set with only 12 pieces of data.
Case 5 - Fuzzy query
Question: Uniswap’s revenue, Unibot’s revenue?
Dune AI requires further clarification of the problem and will not independently provide different queries such as total transaction volume/daily transaction volume.
To further clarify the question daily revenue of Uniswap, Dune AI started to query, but ultimately gave a wrong answer and calculated the daily transaction volume as Uniswaps revenue.
Case 6 - New concept query
Question: What are the details of the inscription process of pols?
Dune AI does not understand the concept of inscription, although someone on the site has written a query related to pols.
summary
Dune AIs basic SQL skills are solid. As long as the corresponding data exists, the corresponding query can be written correctly. However,There are also problems such as insufficient natural semantic processing capabilities, insufficient accuracy, and lack of abstract understanding capabilities.. At present, except for relatively simple questions, Dune AI can only be used as a query writing assistant. Whether the answer is reliable requires manual verification of the SQL content.
