Google Bard and OpenAI ChatGPT at risk of losing out to open source AI models: Google senior software engineer

A leaked letter written by a Google senior software engineer discusses the rise of open source models in the field of natural language processing (NLP) and how they are outpacing the progress made by large tech companies like Google and OpenAI. The letter was shared on Discord and was later published by SemiAnalysis. According to the letter, the author suggests that the open source community has solved many of the major problems in NLP, such as scalable personal AI and responsible release, and are doing so at a faster rate than companies like Google and OpenAI. 

The letter by the Google engineer discusses the significant breakthrough in a scale called low rank adaptation, or LoRA, which reduces the cost and time of model fine-tuning. The author argues that Google and OpenAI should pay more attention to the work being done by the open-source community and prioritize enabling third-party integrations. These advancements are putting pressure on traditional AI research organizations like Google and OpenAI

The article notes several examples of accomplishments made by the open-source community, such as running foundation models on a Pixel 6 phone and creating responsible models with no restrictions. The author suggests that while Google and OpenAI may still hold a slight edge in terms of LLM quality, the gap is quickly closing, and open-source models are becoming more capable, customizable, and private.

The engineer claims that Google and OpenAI need to adapt to this new reality by collaborating with outside researchers and prioritizing third-party integrations. The author also suggests that companies need to rethink the value they offer to customers since people may not be willing to pay for restricted models when free and unrestricted alternatives are available. 

The letter argues that the best models are those that can be iterated upon quickly, and companies should focus on creating small variants that can be updated easily rather than giant models that require significant resources to train.

Also read: ChatGPT hype leading to the spread of dangerous malware: Meta researchers

Also read: ChatGPT-maker Sam Altman says era of remote work is over