An excessive amount of duplication
A sure degree of competitors and parallel growth is wholesome for innovation, however the present scenario appears more and more wasteful. A number of organizations are constructing related capabilities, every contributes to an enormous carbon footprint. This redundancy turns into notably questionable when many fashions carry out in the same method at commonplace reference factors and actual world duties.
The variations within the capacities between the LLMs are sometimes delicate; most excellent in related duties akin to Language technologyAbstract and coding. Though some fashions, akin to GPT-4 or Claude, can barely overcome others on the reference factors, the hole is usually incremental as an alternative of revolutionary.
Most LLMs are skilled in superimposed knowledge units, together with the web content material accessible in public (Wikipedia, frequent monitoring, books, boards, information, and many others.). This shared base results in similarities in data and capacities because the fashions soak up the identical details, linguistic patterns and bias. Variations come up from adjusted patented knowledge units or mild architectural changes, however central common data stays extremely redundant in all fashions.
