Artificial Intelligence and large language models
There will be no GPT-5 — Here is why!
A speculative approach to look behind the metrics of GPT models.
Let’s be honest… After last year being the year of text-to-image generation with Dall-E, Midjourney, and Stable Diffusion 2023 is definitely the year of large language models for generating texts.
OpenAI took the market by storm with GPT-3.5 and its successor GPT-4. All big tech companies strive to bite a piece out of the cake and stumble over the problems of being second place in the race.
Let's do some stats…GPT-3 has proven to be a game changer with its 175 billion parameters, 45 TB training data, 700GB RAM for full precision training, and a cost of “only” 4.6 million US$. Earlier this year GPT-4 was announced and to quote Sam Altman, the CEO of OpenAI, its costs were around 100 million US$. According to some sources the parameters are somewhat around 1.76 trillion.
Let's add some math...So, if we assume GPT-5 would comply with the growth of the previous models it will have more than 17 trillion parameters and the costs will increase by a factor 20 up to 2 billion US$. This is a huge amount of capital to invest for a business case that…
