Simian Words

The generation vs verification delta explains why LLM's are useful

Ever heard that you still need to verify what an LLM says so it implies that LLMs are as good as useless? I always felt that it was a lazy argument. I gave this argument some thought and I came out with an explanation that goes beyond just LLMs.

I was recently looking for a word in English - I knew I had this in the tip of my tongue but was not able to find it. I asked ChatGPT to help me.

This was my question:

Me: synonym for "provides" that also means "places" on you

This was the answer it gave

ChatGPT: 

The most precise options:

imposes — strongest fit; clearly means placing something on someone (often a burden, duty, condition)

assigns — more neutral, structured; often used for roles or duties

places (on you) — exactly what you described, but more literal than elegant

confers — formal; gives something onto someone (often positive, like status or rights)

bestows — similar to confers, slightly more ceremonial

The word I was looking for was "confers". Now I don't have to explain why I don't need to verify what the LLM provided. That would be a stupid exercise. It should be clear to anyone that the LLM has genuinely helped me and there is close to zero chance of it being incorrect. I just know I was looking for "confers".

I think this same process extends to anything the LLM helps me with - it generates and I verify. The complexity or effort needed in my verification is much lower than the complexity or effort needed for the LLM to generate it.

Another example to drive the point across: if I want to create a nice looking Logo for my brand, I can ask someone to help me do it and just show me the end product. I don't need to know their process, I don't need to know the options they tried and discarded. I just need to know the final product to make the judgement.

Another example: if I ask advice from my friends on a certain topic, how do I know that they are correct? Its a very stupid question but you would be surprised to know that this is fairly contentious. Ever heard the acronym IANAL? Coming back to the point, the reason advice still works is that every person has some power by which they can "fit" certain ideas in the world model in their head and intuitively understand whether it works or not. You don't need to have the power to come up with the idea, but just enough power to know whether it "fits". This is the important delta I want to emphasise.

Just like when I find the single missing puzzle piece, I know for sure that this is the one. I don't need to then take all the other pieces and verify by elimination.

Can we use this concept for LLM's?

Ok you might be thinking: can the LLM itself be the verifier? YES. That's exactly why reasoning models work - it literally looks at its exact output each time and changes its outputs. It becomes its own verifier.

I asked myself, what separates the three tiers of GPT?

The thinking version verifies at each token produced. The pro version verifies against many chain of thought chunks. Can we go higher? I don't know how the "max" version of GPT might work.

Does this concept transfer more generally?

This gap between generation and verification is how most of society runs and mostly how knowledge creation works.

The free market allows companies to do the "generation" and hide all the complexity related to how they created the product and just return the product itself. The consumers are the "verifiers" who can enjoy and criticise the product by taking part in the market and voting with their wallet.

As the CEO of a company, you may delegate large portions of your work to your subordinates. You give them certain creative freedom but as the final accountable person - you have enough judgement to know what works and what doesn't. The CEO doesn't need to know the intricate problems a developer faced while debugging an issue - they just need the end product. Although this is a very important caveat: a CEO who can hold more context is a much better CEO. The best CEO or a boss is one who can empathise deeply with their subordinates. But a human has certain physical constraints!

The society works in this way because a single human can not hold all this complexity in their head. They hold the necessary context and delegate the rest.