Bandwidth, Signal, Noise
LLMs and other AI technologies can solve some of the fundamental blockers for progress and productivity.
I'm excited about the advancements we're seeing in foundation models, agent development, and other AI related areas. They're getting better and cheaper by the day.
Putting aside the hype and the hand-wringing, I want to discuss what's possible today with currently available models and related technology.
LLMs have the potential to solve some of the biggest blockers to progress and productivity that we're facing:
Bandwidth: We're limited by our ability to parse and process information. We have finite cognitive bandwidth.
Noise: information grows exponentially, and we can't keep up because of point 1. More emails, messages, documents, meetings, reports etc. The signal to noise ratio get's weaker every day.
The combination of 1 and 2 means that human potential is being wasted. Research is slower, productivity is wasted. More of our limited time is spent extracting signal from noise.
We're about to enter an age where for the first time we'll be able to increase the amount of intelligence in the world at will, limited only by energy scarcity or abundance. The impact of this is almost beyond comprehension.
The Bandwidth Problem
A person can only parse so many bits of information in a given period. We're limited on input (reading, listening, watching), and output (writing, speaking, etc.). Hard measures of bandwidth aren't fully applicable but the analogy serves and the conclusion is the same. We have finite cognitive resources.
Part of the problem is that we currently have to ingest a lot of "noise" (irrelevant information) to discern what's valuable and relevant ("signal"). Our meat-compute is expended on triage. Noise clogs up our pipes. That noise continuously increases, but our bandwidth does not. Every day there's exponentially more content, tweets, emails, Slack messages, documents, meetings – but we don't have extra hours or calories to spare.
In contrast, the bandwidth of an LLM is many orders of magnitude higher than that of a person. AI can be used as a noise-reduction solution, a signal booster. A person paired with a competent and contextualized LLM can meaningfully increase their productivity by having AI augment and automate sorting signal from noise.
The Productivity Paradox
If we look at the development of technology over the last 20 years we see that for every potential productivity gain, it also creates noise, which creates further work, decreasing productivity.
There's no free lunch. A good example is Slack. Real-time, multi-player conversations, which are useful. But a large portion of those messages and updates are irrelevant for any one individual. There's no way to know that until you've read all the messages.
Companies have more software in their stack than ever before, but is it helpful? Ask an average knowledge worker to show you their browser tabs. Or look at your own. Ask your colleagues whether they feel they can keep up with all the information they need to.
Knowledge workers spent a significant amount of their time paying back information-debt in the form of admin work, meetings, reading and responding to emails/messages, and searching for information.
The AI Solution
If AI and LLMs only make one significant impact on the world it will be this: to flip the balance from debt to credit.
Consider the impact of increased efficiency in areas such as medical research and energy development.
It's also easy to see how the benefits of AI will accrue at the corporate level. Workers probably won't work less by default, or be paid 10x more for being 10x more productive. But in the areas where it counts, life saving research, cross-pollination of ideas in the sciences — the benefits will be for everyone.
That's if we choose to utilize AI this way. It's no good if all bandwidth is being used by consuming hyper-personalized, hyper-addictive media. But that's a different problem.