The Cost of Using LLMs

Can't read or see images? View this email in a browser

This is one of the most common thoughts CXOs have when using LLMs today. The answer, however, depends on several factors, including the number of tokens processed, the specific API used, and prices, among other things. AIM noted that GPT-4 is 50 times more expensive than Llama 2, specifically for summarisation of the Wikipedia text into half its size. Presently, there are a plethora of options to choose from. While OpenAI’s models are highly rated, we have observed that open-source models like Llama 2, Falcon 180-B, and Mistral 7B are catching up to GPT-4 in terms of performance and are slowly gaining traction. Many industry leaders are advocating for domain-specific LLMs. However, selecting an optimal LLM amidst this vast array of open-source and proprietary model demands becomes challenging, where they have to meticulously evaluate, alongside balancing between cost and performance.  Wikipedia, for instance, has six million articles, each around 750 words long. That equals 1,000 tokens because three-fourths of a word is equal to one token, so it translates to 6 billion tokens in all. When we reduce the size of Wikipedia by half, we will be left with 3 billion tokens as output.  Here’s a quick glimpse of the cost variations for summarisation among different models revealing significant differences in pricing structures. Similarly, the cost for RAG, fine-tuning and API calling also changes for different models. Read to find out more.  Read the LLM Economics – A Guide to Generative AI Implementation Cost report from AIM Research for some in-depth analysis and cost estimation to leverage LLMs for enterprise use cases.  Check out the LLM Calculator from MachineHack here.  The OpenAI-G42 Partnership  While AI companies are busy competing with each other, OpenAI and G42 have formed a first-of-its-kind strategic alliance to work together to build cutting-edge products for the Middle East market. This new development comes in the backdrop of G42 launching an Arabic language AI model Jais, which contains 13 billion parameters and combines Arabic and English data. The model was built in collaboration with academicians and engineers, partly from the scarcity of bilingual language models. Interestingly, Jais was built on supercomputers produced by Cerebras Systems. The partnership with OpenAI will help the company achieve its multilingual model goals.  Read the full story here.  Indian IT & Freshers Amidst rampant tech layoffs at top firms in the world, Indian IT majors are showing signs of disparity in hiring freshers as they are mostly focusing on training and upskilling their existing employees, alongside scaling work automation and digital initiatives.  Here’s a quick glimpse:  Read the full story here. TAUSIF ALAM & AMIT RAJA NAIK Friday, Oct 20, 2023 | Was this email forwarded to you? Sign up here DOWNLOAD OUR MOBILE APP Stay Connected info@analyticsindiamag.com © 2023 Analytics India Magazine

This email was sent by support@analyticsindiamag.com to alexvarboffin.abbb@blogger.com Not interested? Unsubscribe | Manage Preference | Update profile Analytics India Magazine | 280, 2nd floor, 5th Main, 15 A cross, Sector 6, HSR layout Bengaluru, Karnataka 560102