The website SemiAnalysis has published an internal letter from a senior Google developer titled “We Have No Moat, And Neither Does OpenAI”, posted by an anonymous person on a public Discord server was shared. The developer believes that the increasing success of open-source projects in the field of artificial intelligence and natural language processing is putting both Google and OpenAI under pressure. He cites the impressive speed at which free language models such as Metas LLaMA are evolving as the reason for this. LLaMA was released shortly after the announcement as a torrent and is now the basis of numerous other models, such as Alpaca (Stanford University), Vicuna, LLaVA and Koala (Berkeley University).
After just one month, the open source community has already achieved variants with instruction tuning, reinforcement learning from human feedback (RLHF), quantization and various other quality improvements. While only large companies could previously afford the training of large language models, key innovations in open source projects such as Low Rank Adaptation (LoRA) the entry barrier for training and experiments has been significantly lowered, the developer writes. At the same time, these techniques enabled fast iterations, high quality, and lower costs. As a result, open source models have intensified competition with Google and OpenAI and pose a serious threat.
Adapt to “open source reality”.
According to the developer it’s critical for Google (and OpenAI) to adapt to this open-source reality and find ways to collaborate with the community rather than remain in control of their models.
Google should consider making smaller versions of its models available to the general public and bridging the gap between commercial and open source offerings. This also includes working with researchers and developers in order to strengthen our own innovative strength. Because the free projects have already solved problems that Google is still struggling with. Therefore, collaborating with the open source community and adapting to the new conditions is crucial to create long-term value and maintain the influence of both companies in the AI industry.
also read
(v.a.z.a.)
