Here's an example prompt that you could use your for your first iteration, and you don't yet know very much about the domain: https://t.co/9HpugontJ8
The important bit is the 2nd and 3rd paragraphs, which target Deep Research at the surrounding links and blog posts that mention The Math Academy Way link (I don't think Deep Research can read PDFs yet, but I may be mistaken).
In my case, after doing the learning system, I noticed a few areas of focus, and came up with an Iteration 2 prompt like this one--where I mention the "state machine" and New Masters to focus its research: https://t.co/nEqBDloSnG
Compare that to the control prompt (No Math Academy mentioned) that missed the skill hierarchy and practical exercises: https://t.co/P44DJ0a1XF
The control prompt basically just summarizes the syllabus of many online classes, whereas the 1st and 2nd Iteration prompts cause it to focus on practical exercises and skill hierarchy, i.e. the actual things you need to make progress.
The output of these example prompts is very similar to the the real-world learning system that I've been using for about the last half year.
1/ I’ve spent most of the last few weeks since the Google, Caltech papers to think about tradable implications around quantum computing and crypto
specifically what happens to the market around q-day
“Getting the harness right is not a prompt engineering problem. It is a systems engineering problem. And it is the most important engineering problem in applied AI right now.”
One of my favorite lessons I’ve learnt from working with smart people:
Action produces information. If you’re unsure of what to do, just do anything, even if it’s the wrong thing. This will give you information about what you should actually be doing.
Sounds simple on the surface - the hard part is making it part of your every day working process.
I disagree with @gregisenberg's take on AI-enhanced agencies (and their margins)
his argument:
AI agents make agencies 10x more efficient -> margins go up -> way smaller teams can capture much more value -> agencies will find this new format that's closer to software/automation than a traditional service
I heavily disagree with his "margins just go up" part
this will only be true short term
it's a play on transient market inefficiency, not on long-term MOAT
here's why:
If every agency can use AI to cut costs by 80%, what will happen next?
they compete on price. hard.
it's econ 101 - when your costs drop and barriers to entry are low, you can't just pocket the difference forever
long-term, savings will be passed on to the customer and won't be retained on the agency side (at least a huge part of it)
once clients understand that services cost less to deliver, they demand prices to be lower
we are already seeing this in ghostwriting services:
since AI use is becoming mainstream, prices are going DOWN
he's right, agencies will be more like software companies
This means that the pricing will also mimic software pricing (=much lower than most agency retainers right now).
If now 2 people can do the work of 50 people agencies, the barrier to entry is MUCH lower:
competition is higher
there's a simple rule in service businesses:
efficiency gains get competed away in competitive markets
I don't completely disagree with Greg, I think that AI is exciting news for agency businesses
there's a lot of land to grab right now if you are early in the automation game of services
you can absolutely make shitload of money using this playbook
being early and fast is key here
the market inefficiency and slow pace of big players to adapt is your opportunity
but it's not the case that margins to go up and prices to stay the same
there's a 0% long-term for that to happen
5/
Be careful with “Under-Promise & Over-Deliver”
Under-Promise & Over-Deliver is a good policy for setting external expectations e.g. with customers, investors
It is a terrible practice for setting personal goals because it builds the habit of aiming lower than our potential.
Seeing three kinds of "AI companies" emerge:
1. Builds foundational models, and attempts to monetize them (eg ChatGPT, Claude) or not (Lllama, Mistral etc)
2. Offers LLM-enhanced functionality to customers, and customers can choose model(s) they use (eg Sourcegraph, Perplexity)
Less than 2 weeks since OpenAI started rolling out GPT-4o.
And people have been busy executing god-like tasks.
50 truly mind-boggling examples:
(29th is my favorite)