The Conundrum of Generative AI
The reality lies somewhere between fear and excitement
Day 3 of the #100daygenai series
We covered the broad landscape of Generative AI from one of the leading voices in AI — Steve Nouri on Day 1 and started to take baby steps toward what led to the development of such colossal models on Day 2.
Now, on Day 3, it is time to get a reality check of understanding whether it's hype or a bubble that will meet its own fate with the passing of time when the furore comes to an end.
Or, are we missing out on the critical once-in-a-lifetime moment to make the most of this “electricity equivalent” revolution?
We are data scientists and can best resort to data to get a reality check.
This brings me to share a renowned report — the 2023 AI Index Annual Report which is “an independent initiative at the Stanford Institute for Human-Centered Artificial Intelligence (HAI)”.
IEEE finds it as one of the most comprehensive capturing state of affairs in a detailed 302-page report.
Let us discuss some of the key highlights of the report —
- The “largeness” of large language models is expensive on the pocket — a 1.5 billion parameters model cost a whopping $50K in training. And, PaLM with 540 billion parameters → $8M in cost
- AI reached to masses with the release of DALL-E 2, Stable Diffusion, and ChatGPT
- Evolution from “narrow-tasked” traditional models to “multi-task” models, that are capable of performing multiple tasks
- While the world is busy developing applications to solve one business problem, apparently at the cost of creating more problems in terms of emitting carbon leading to serious environmental harm. Now, this hurts badly!
- Ever heard of “diamond cuts diamond”? We have seen the developments that suggest AI will improve itself — a preview of the self-supervised AI era reflected when Google used PaLM to improve the same model.
- A sign of good progression — Toxicity and biased responses from previous models have been mitigated with instruction-tuned newly trained models.
- Deepfakes have emerged as one of the big concerns among others, giving rise to interest in AI Ethics.
While there is more to the report, I will cover them in future posts by combining those data points with similar and advanced developments that happened since then.
Till then, let us revise our findings through this concise graphic:
Stay tuned on the following platforms for further updates:
