MariaDB сангийн гүйцэтгэх захирал Каж Арнө өөрийн мэдээллийн санд вектор хайлтын функц нэмэгдсэнийг хиймэл оюун ухааны хөгжүүлэлтийн хэрэгслүүдэд хүлээн зөвшөөрүүлэхэд бэрхшээлтэй байгааг онцоллоо.
MariaDB нь 2009 онд MySQL-ээс салж үүссэн бөгөөд одоо хиймэл оюун ухааны аппликейшнүүдэд түгээмэл шаардлагатай, ижил төстэй өгөгдлийг хайх вектор хайлтын функцийг дэмждэг болсон. Гэсэн хэдий ч хөгжүүлэгчид шинэ төсөл эхлүүлэхдээ том хэлний загваруудын (LLM) санал болгосон үндсэн сонголтуудад тулгуурладаг нь MariaDB-д сөргөөр нөлөөлж байна. Тухайлбал, Claude зэрэг хиймэл оюун ухааны хэрэгслүүд вектор хайлтад ихэвчлэн PostgreSQL-ийг санал болгож, MariaDB-ийг орхигдуулдаг байна.
Каж Арнө-гийн тайлбарласнаар, энэхүү асуудал нь хиймэл оюун ухааны загваруудын сургалтын өгөгдөл болон баримт бичгүүдэд MariaDB-ийн шинэ боломжууд хангалттай тусгагдаагүйтэй холбоотой юм. Энэ нь хайлтын системийн оновчлолтой төстэй “LLM оновчлол” хэмээх шинэ сорилтыг үүсгэж байгаа бөгөөд хэрэглэгчид болон хөгжүүлэгчид MariaDB-ийг идэвхтэй ашиглаж, баримтжуулах замаар л уг байдлыг өөрчлөх боломжтой гэж үзэж байна.
Түүнчлэн MariaDB-ийн нээлттэй эхийн нийгэмлэг болон арилжааны бүтээгдэхүүн хоорондын ялгаатай холбоотой асуудлууд ч хөндөгдлөө. Тухайлбал, кластерын технологи болох Galera-ийн хувьд MariaDB plc компани нь арилжааны байгууллага болохын хувьд өөрийн бүтээгдэхүүний стратегийг бие даан тодорхойлох эрхтэй гэдгийг Каж Арнө тодотгов. Тэрээр MariaDB серверийн үндсэн кодыг хамгаалах нь нэн тэргүүний зорилт хэвээр үлдэнэ гэдгийг мэдэгдлээ.
Дэлгэрэнгүйг эх сурвалжаас харах
↓Эх сурвалжийг нээх ↓
MariaDB has built vector search into its database. Its problem is persuading AI coding tools to tell anyone about it, says the executive chairman of the MariaDB Foundation. “We’re so frustrated that the world doesn’t understand that MariaDB has vectors,” Kaj Arnö tells us. Arnö is speaking to The Register at Percona Live, where rival database vendors seem to be in an unusually collegiate mood. Not that Arnö is above throwing a little shade here and there: “I think we are in a period of calmness and sort of sailing smooth waters and seeing how AI impacts stuff, how extensibility impacts stuff… and I’m not as sure that MySQL is going through all that smooth times.” MariaDB was forked from MySQL in 2009 as Oracle’s acquisition of Sun Microsystems loomed. It now supports vector search, which lets applications find items by similarity rather than exact matches – a common requirement in AI applications. The nonprofit MariaDB Foundation supports development of the open source server and its community, while MariaDB plc develops the commercial products and provides support. MariaDB is familiar to users of many Linux distributions. Arnö’s concern is whether developers starting a new AI project will ever be pointed toward it. “People start using vector databases based on some framework or based on some default choices of the LLM,” he explains. “So if you do vibecoding and just sort of press enter when Claude Code suggests something, you will not end up using MariaDB for your vector search, and that is what we need to change.” Arnö attributes part of the problem to the information available to AI tools. Their training data may predate MariaDB’s vector features, while the documentation and examples developers encounter elsewhere may still favor established alternatives. “There are a million places where we need to update stuff so that the frameworks, the training data for the LLMs, and all of those know about the existence of MariaDB vectors,” he says. “Like, if you ask a random LLM, ‘What database should I use for vectors?’ And you do not mention MariaDB; today, based on the training that the LLM models have right now, nobody suggests MariaDB.” In one decidedly unscientific check, we asked Claude which database to use for vector search. It suggested PostgreSQL first and did not mention MariaDB. When asked to compare MariaDB with PostgreSQL for the same task, however, it had plenty to say. The tool knew about MariaDB’s capabilities; it just didn’t volunteer them. “I compare this to the old days of how to improve your Google ranking. It was sort of a mystical science. You did not really know how Google prioritized results. “Now the world has moved on [from search engine optimization] to LLM optimization.” Arnö acknowledges that vendors could try to game the recommendations, but says: “I think the only proper way of influencing it is by having it and by documenting it and by people using it.” That leaves MariaDB with a chicken-and-egg problem: without users generating examples and discussions of its vector search, AI tools have less material from which to learn about it. “The early movers have an undue advantage, and it’s like if the water starts flowing in a certain direction, it will continue to flow in that direction. And now we have to change that and make people aware of what we are doing.” MariaDB’s efforts to attract developers also raise a question about which capabilities will remain available in its community edition. Galera, the clustering technology used with MariaDB, has attracted controversy over its open source future and its place in the company’s enterprise product line. “You can have different opinions on whether that is exactly part of MariaDB, the core server, or not, but it is a product that was not part of MariaDB’s core engineering team to begin with. It was managed by an entity that was acquired by MariaDB plc, which is a commercial entity and which is not equal to MariaDB server, and they are allowed to do whatever they want with it,” Arnö says. “It’s their freedom. It’s their prerogative to decide what they want to do. What we protect is MariaDB server, and if anybody is messing with that, we would be quite upset!” ®
