Suggestions
Sam Morgan
Co-founder at TechMap, the #1 course for professionals to learn about tech
Sam Morgan is a co-founder at TechMap, a platform designed to help professionals learn about technology and enhance their skills in the tech industry. He has a background in tech education and product specialization, focusing on solving complex problems and adaptive systems. Prior to his role at TechMap, he gained experience at companies like x27 and Makers.23
In addition to his work at TechMap, Sam Morgan is known for his passion for continuous improvement methodologies, particularly the Toyota Kata approach. He actively engages in coaching and development, helping individuals shift from fear to confidence in their learning processes. He hosts a YouTube series titled "C.I. in 5" and is involved in a community called "The Lean Communicators" which focuses on sharing knowledge and practices in the field of continuous improvement.1
Highlights
I have a strategy question!
My co-founder Ruben Kostucki and I are debating moats for businesses in the AI world. We asked: can an proprietary LLM-powered business (especially products that are mostly wrappers around ChatGPT) have a moat, when they can be (easily) reproduced?
I think the answer is yes, but in the short term. The two main moats (beyond standard business moats like interaction model, adoption, network effects) are fine-tuning (a custom model that produces higher-quality answers a specific purpose) and knowledge indexing (proprietary datasets that the model can use for context).
But I think both of these advantages are temporary.
Fine-tuning Vendors (e.g. OpenAI, Google, Mistral) will keep releasing better and cheaper LLMs. This will erode much of the quality moat from fine-tuning.
Knowledge indexing Vendors with up-to-date knowledge (no 2021 cutoff), tools (like search engines – see Bing, Induced) and the compute for massive indexing will erode much of the moat from supplying a public (or guessable) knowledge base.
Then there's Agents: multiple interacting LLMs working with tools (such as browsers, text messaging, and phone calls) to automate sophisticated behaviour. These will be not just capable of spitting out text, but taking any action that one can perform with an internet connection. Once agents are on the scene, anything that requires a form, or clicking on the internet, can be fully automated. That includes data ingestion for knowledge bases.
Private knowledge bases (e.g. large internal datasets) may remain a viable moat. But most smaller companies don't have one of those.
So what to build? Imagine we’re living in the future (under 2 years). What could be built by that point that would have a moat in that world?
One option is an incumbent industry tool (e.g. ATS, CRM) replacement that's built from the ground up on these advanced technologies. They probably look very different from the current tools. To take on incumbents (e.g. Lever, Greenhouse), they would need to have a minimum feature set, so it's not a small project. And you need to go very fast, with a lot of cash, because that market is going to get busy.
Another option is tooling that enables the builders of these tools. For instance, fine-tuned models will remain faster and cheaper than stock proprietary ones. This may become more and more of an advantage as LLM technology sees wider adoption. Users want fast, and businesses want cheap. Tools like Humanloop are early stabs at making fine-tuning less painful.
(I'm skeptical about products that help you build a knowledge base, like Nuclia, because all this clicking and uploading will be automated by agents.)
Does anyone see other options?
Do you know anyone who can help?
I'm building a new data apprenticeship with @makersacademy. Apprentices will be combo data analyst/engineers with modern software engineering principles (e.g. TDD, clean code), and some devops in the mix. Paid for on the Apprenticeship Levy, shaped by employers, and specialised to your stack.
I'm looking for organisations to feed into the programme design. Any intros with heads of data, apprenticeship managers, or specialists would be valuable!
