GenAI — Act 3
· Originally published on Medium
An In-depth, Detailed Analysis of — What’s Next in AI?
> Orchestrating AI Agents to Solve Billion $ Problems <
Productivity, Automation & Personalization
This is what Generative AI is all about TODAY and this is where most of the value lies
This will change soon as the LLM landscape is Evolving Fast!
Enterprises have already implemented/implemented Code Review Agents, Unified Workplace search, GPTs to boost employee productivity and whatnot.
All this while the unit cost of intelligence is trending towards zero $🧠 -> 0
Soon intelligence will be too cheep to meter
(Hardware is getting insanely faster too. Groq, Nvidia, Etched)
And if you think all of this is hype, just take a look at this:
Just a Few days ago, AlphaGeometry by Deepmind got silver with their maths models, with just 1 point away from gold!
Note that most of the Gold, Silver or Bronze medelist go on to become millionaires or billionaires later on their life and
And don’t you forget, we aren't even done with scaling LLMs
LLMs are getting faster, cheaper and more intelligent in general
It’s not even a year and a half, but now it’s 12 times cheaper to make a call to GPT-4o than the GPT-4 model and also 6 times faster in terms of time to First token response.
If GPT-3 was a Shark 🦈
and GPT-4 was an Orca 🐬
Then GPT-5 will be a Whale 🐳
We are not even close to getting diminishing returns from compute!
Oh wait, there’s more, we are also making major progress in understanding how these LLMs work and ways to steer them with a very high degree of control
All this points out towards a future (happening right now, will be seen everywhere in few months) where a large part of efforts will be put into orchestrating AI models to solve important problems.
Frameworks like DSPy, LangGraph and more are getting insane popularity to build the AI workflows and Agents.
Projects like OpenDeven have really shown how capeble the AI models can be with just a little orchestration, and they have just begun.
Is AI just affecting knowlege work? NO!
Accelerated progress is being made in Robotics too
‘Everything is Going to Be Robotic’ — Nvidia
We can’t design a Chip anymore without AI
We can’t write Software without AI
We can’t file Bugs without AI
Robots are building Robots
AI is improving AI
We want every organization, every individual to use AI very aggressively
Says Nvidia (ofcourse, they sell shovels for AI, but its true)
AI Agents, Multimodal LLMs, Robotics, Neuroscience, all are advancing very rapidly. All productivity is being hugely aided by AI (see below chart)
Traditional industries like Manufacturing, Banking, Finance and the rest can’t help but modernize to avail never before seen levels of productivity
This brings me to the one of the interesting problems that has been plaguing the software and consequently all the associated industries from its inception.
Technical Debt, A trillion dollar problem.
And the long human dream of having a single interface to rule them all.
All of this requires strong orchestration systems for LLMs that get the desired work done (while the latter one might be significantly easier)
So the bottom line again, with the Act 3 with Gen AI is
Orchestrating AI Agents to Solve Billion $ Problems
References:
https://www.anthropic.com/news/mapping-mind-language-model
https://arxiv.org/abs/2402.01622
https://github.com/princeton-nlp/SWE-agent