Suggestions
Lakshmana Padmanabhan
Experienced US Healthcare payer leader
Lakshmana Padmanabhan is an experienced IT business strategist, program and project manager, and consultant. With over 30 years of experience in the field, Padmanabhan has worked in every level, starting as a programmer on IBM mainframes, transitioning to team leader, senior systems analyst, software specialist, EDP manager, head of application development, principal consultant, consultant, project manager, and director at various organizations around the world.
Padmanabhan specializes in developing and implementing IT strategies and roadmaps, conducting feasibility studies and audits, developing policies, and reviewing IT security. He is particularly experienced in ERP solution implementation and customization and has a keen eye for business opportunities, successfully developing business associations.
Padmanabhan holds certifications in Project Management (PMP) and Information Systems Auditing (CISA). He also holds a Bachelor's degree in Mechanical Engineering from the National Institute of Technology Tiruchirappalli and an MBA in Finance from the University of Madras.
Padmanabhan is currently an investor at NostraGamus-Gambit Sports and has previously worked as a consultant at Microsoft Gulf, principal consultant at BDO Bahrain, head of application development at the Ministry of Health-Bahrain, and software specialist at Emitac, among other positions.
Highlights
OpenAI has announced their GPT-4o-mini model yesterday. They claim that 4o-mini has all the multimodal capabilities of their GPT-4o, but smaller and cheaper to use. Drastic difference in their charges between the two models for 1 million input and output tokens doesn’t make much sense to me as someone who is experimenting with my personal projects and testing various LLMs and their capabilities and learning from my experiments. So I decided to use the GPT-4o-mini on my earlier project using audio and video features and do exactly the same number of repetitive tests and compare the costs in a context relevant to me. The reduction is cost has been significant. What cost me $1.47 on 9th July using GPT-4o cost just $0.02 today using GPT-4o-mini. Gives a good incentive to indulge more in testing experimenting with powerful models instead of using opensource models such as Mistral and Llama3 running locally on my laptop for free! Here is a screenshot showing my OpenAI dashboard showing the usage
I wanted to compare the just released Claude Sonnet 3.5 with ChatGPT-4o for their multi modal capabilities such as vision and audio. I used a python program which listens for an audio input while getting a visual input(through the integrated webcam in my laptop). When the audio is detected, it converts it into text using OpenAI’s whisper model functions and sends it to the LLM as a prompt along with encoded webcam frame. As a comparison demo, I showed an object (a computer mouse) to my webcam and asked the LLM to identify the object. I used the same code and asked the same question using the same object. The only thing that was changed was a single line of code, switching the LLM (see the screenshots) Sonnet 3.5 answered as below: You’re holding a computer mouse. It appears to be a wireless optical mouse, black in color
ChatGPT-4o answered as below: You are holding a wireless computer mouse with the battery compartment open, showing a battery inside
As can be seen, ChatGPT 4o’s response is more detailed than Sonnet 3.5. This doesn’t mean much as this is not any benchmark test and just a personal test, where it is impossible to maintain exactly same set conditions between the two tests. It helped me understand how I can develop applications using image and audio inputs. This opens out a vast number of use cases, which can be explored. Interesting times ahead!
