I am a software engineering student who is very interested in and passionate about machine learning and AI. I first became interested in Machine learning when I was in my second year in 2020. During those years, I tried a lot of things in machine learning, most importantly computer vision and NLP. I tried different things such as face recognition, Image classification, Image generation from text, etc.
When GPT3 came out in 2021, I was excited and wanted to try it and implement it in my application because I was also familiar with language models like Bert from Google, when I saw the video marketing of GPT3 I was just wow.
After seeing a lot of demo videos and applications from OpenAI, I decided to send an email to OpenAI for beta testing of GPT3 because using a powerful model like this is not as easy as nowadays.
But after a few years of GPT3 release, there are some changes that I have noticed
The cost of Implementing a Generative model in software applications rapidly decreases
When I first started using the first release of GPT3, the price of davinci (which was the most powerful model at that time) was 0.06$/ 1k tokens but now the cost of GPT4-o(the most powerful model at the time I write this) 0.0025$ /1k tokens which are 95.83% cheaper, Moreover because there are many open source models came out every week I think this price will keep decreasing overtime and most of the application will be implementing LLM.
I was overwhelmed with the new LLM model released every week
In 2022 when chatGPT and GPT3 were released I felt like it was a golden age of AI seem there many big companies they started working on their LLM model and releasing new models (such as Google), When I opened LinkedIn I saw a new release of a new model every day and I felt a bit overwhelming with the information and the things I need to catch up with the current trend.
sometimes I just want to say ‘Can you guys just relax’.
Many companies just claim that they are AI companies by just calling OpenAI API
I never thought that becoming an AI-based company would be easy like this time, but with OpenAI API many companies claim that they are building and training AI model as a selling point of their product. With these claims, I feel like it is a scam.
There is a prompt engineer role
This was surprising for me at first when people created a new role just to write instructions for the LLM model, but I think this is a very important skill for people who work with LLM since most of the tasks in LLM are achievable by using prompting without fine-tune or training.
Sometimes people prefer to use the LLM model on most tasks over a statistics model
I think this is not just me, since we can apply LLM to other problems like classification, clustering, etc, by using prompting without requiring additional resources and skills to train a statistics-based model, when we use an LLM model we tend to set a very high expected accuracy of the model that why we tend not to use statistic model because statistic model may not much accurate as LLM, but in using LLM in this task may not optimize the resource since we need to pay money for LLM and if even we run it in our server we still need a very large resource to run the models.
Conclusion
As an early user and learner of Data Science and machine learning, I feel like this is the golden age of AI since there are a lot of releases new AI models every week and a lot of people know and use them in daily life, if we compare to the time before ChatGPT release there are limited group of people who understand and use AI in daily life, I think in the future there will be a lot more AI model release and most of the software product will be implement AI in their system especially LLM because there will be much easier, more accurate, and much cheaper to implement those technology into software product.
