Happy New Year
Yes, I know that it is only Nov 29th… but then have you forgotten that we will be entering the 3rd year of the AC era tomorrow morning. Do you even remember the BC era?
Ofcourse I am talking about the BC - Before ChatGPT and AC - After ChatGPT inflection point that occurred on Nov 30, 2022 when OpenAI released (or should I say unleashed) chatGPT onto the world.
Where are we and what have we learnt? What reasonable guesses can we make about the near future?
The world wasn’t made for us, we were made for the world. John Green, The Fault of the Stars
After the initial dropping of the jaws and the fear of missing out, I think we are now beginning to realize LLMs for what they are… a nice neat new piece of technology, that is in its initial stages of evolution and one that neither has a huge MOAT nor is particularly valuable in and of itself, especially from a business perspective.
History may yet record that the biggest contribution of the advent of the LLM era may have been not a technological advancement but a social one. An advancement where lots of “non technical” people started paying attention to technology for the impacts it can have on their daily lives.
If you choose not to decide, you have still made a choice. Neal Peart
My nieces and nephews reach out to chatGPT before they reach out to Google. No school homework in the near future is likely to get done without the assistance of some such tool. My wife uses it for work and so do many others.
In the days ahead, today's social divide between the college graduates and non graduates is going to be replaced by those who know, understand and can harness the power of this technology and those who can’t.
Make your choices wisely.
You are not designed for everyone to like you
Huggingface today has over 900K models, 200K datasets and 300K datasets. Give up any hope of ever catching up to really understanding them all. The general pace of change in this space is too fast to catch up.
Recommended by LinkedIn
Is all hope lost then? I say not. When in doubt, go back to the basics. And I classify the basics as falling into a few simple categories (a) The mathematics underlying this technology (b) The problem that you want to address (c) Looking for and finding the right partner to work with in helping address the challenge. As the cheshire cat advised Alice - “That depends a good deal on where you want to go”
Each business problem will require a specific solution. And they may not need any LLMs ;-)
Everything we hear is an opinion, not a FACT. Everything we see is a perspective, not the TRUTH. Marcus Aurelius
Built into the very fabric/design of the LLM based approach is a probabilistic generation of responses. Probabilities largely driven by the training data sets. How different is that from an opinion. To talk nothing of the images and podcasts and other outputs they produce.
So you will do well to look for models and approaches that ground their responses in FACTs and can provide you with evidence of the same. Look beyond the LLMs, there are better ways to avoid hallucinations.
HAL: I'm sorry, Dave. I'm afraid I can't do that.
Dave Bowman: What's the problem?
HAL: I think you know what the problem is just as well as I do. 2001, A Space Odyssey
The problems are only now beginning to get more widely recognized. The enormous resource footprint in building, operating and enhancing these LLMs is simply unsustainable for planet earth. The average power consumption of a human brain is 20W. That’s Einstein, Picasso, Shakespeare and every other person.
And we are already hearing about scaling laws beginning to break down, bigger models not always better, etc. etc. The story is just beginning....
Yr 3 of the AC era I predict will bring about the first large scale, secure, non GPU based, zero hallucination enterprise grade solution.
How realistic do you think that is? I look forward to hearing from you on your learnings, predictions and fears.
I help Supply Chain Leaders Automate SNOP ⚡ Reduce Manual Effort ⚡ Supply Chain Dashboards ⚡ ERP-Agnostic Solutions⚡Real-Time Insights⚡Demand and Supply Planning⚡Decision-Making Tools⚡Production Scheduling
1moNice post Srini
# Financial Crime Risk, # Financial Crime Detection, # Sanctions Screening, # KYC ,# Global Capability Centres # People and Change Manager # Agile Teams # Screening Technology
1moSrinivas. Loved reading this. A very cleverly written article interspersed with quotes that land in the right place. Hope you are well.
I'd do anything for work but I won't do that
1moI always think of the “no free-lunch theorem” when it comes to LLMs, in that no one model can correctly capture all sorts of data. There is some mention of “non-Turing machine” circumventing of this theorem on the Wikipedia page for this, but again, each user must tune their LLM/SLM application to their specific needs, anyhow.