Great post as always. I see a lot of cross disciplinary interest in ‘creating using ai’ and the democratisation of app building in my area across engineering and non engineering communities, latter more so interestingly. An offshoot of this could be that theorists/non techies with lower barrier of entry to practice, help elevate ideas and products to a next level by ability to prototype and create complex software independently.
Thanks for an inspiring post. I'm coming from a design background and want to merge it with AI. I don't know if AI Engineer is the title I'm looking for rather AI Designer. The title doesn't really matter, I still want to be able to build things not just design the surface.
Todd, I'd love to know what you've been finding challenging in your journey. Sometimes it feels like I'm exploring these ideas in a vacuum, so I'm super curious about what my readers are looking into.
This week, the challenges are around educating end users about how to integrate LLMs into their workflows without jeopardizing privacy and security requirements. Keeping up with the firehose of new features and capabilities in all the LLMs that are available. Creating a deep personal understanding of how LLMs actually work and what technical limitations are built into their architectures. Helping managers sort through the marketing spin to find the solutions that really address their needs (even if it means that AI is not an option).
For some reason, I haven't subscribed to this Substack - but I'm a subscriber now!
I like your focus on AI application development instead LLM development. The number of various LLMs is already too big for anyone to follow in a detailed fashion and when the efforts mature most of LLMs will disappear. Since most of the AI apps use LLMs via LLM API, LLMs can be viewed as runtime libraries
I think it's clear that we're going to see a consolidation of proprietary LLMs, probably 2-3 main players, with maybe a handful of industry-specific options. The open source community will likely continue to grow, but it's probably not the starting point for most devs at this point.
Instead of building in public, I have been learning in public. Starting from first principals that you would learn in undergrad like Bayesian inference, modelling using Gaussian distribution and building my way up to deep learning, while writing about what I learned with minimum math jargons on substack. I found learning these “historical” ML methods both help me understand where today’s AI comes from and also keep me grounded from all the hype. If anyone is interested, feel free to check it out.
Great post as always. I see a lot of cross disciplinary interest in ‘creating using ai’ and the democratisation of app building in my area across engineering and non engineering communities, latter more so interestingly. An offshoot of this could be that theorists/non techies with lower barrier of entry to practice, help elevate ideas and products to a next level by ability to prototype and create complex software independently.
Fantastic read, thanks for sharing. For some reason I especially enjoy how you did the footnotes
Thanks! Not sure what you mean about the footnotes - I kind of assumed that's how everybody used them.
I just don’t see footnotes that often on Substack :)
Great blog. Might want to credit Gartner with the Hypecycle image. It's their IP.
Thanks for an inspiring post. I'm coming from a design background and want to merge it with AI. I don't know if AI Engineer is the title I'm looking for rather AI Designer. The title doesn't really matter, I still want to be able to build things not just design the surface.
Fantastic article. I’m also on the “AI Engineer” journey of discovery and I’ve been following your Substack to get inspiration. Thanks for sharing.
Todd, I'd love to know what you've been finding challenging in your journey. Sometimes it feels like I'm exploring these ideas in a vacuum, so I'm super curious about what my readers are looking into.
This week, the challenges are around educating end users about how to integrate LLMs into their workflows without jeopardizing privacy and security requirements. Keeping up with the firehose of new features and capabilities in all the LLMs that are available. Creating a deep personal understanding of how LLMs actually work and what technical limitations are built into their architectures. Helping managers sort through the marketing spin to find the solutions that really address their needs (even if it means that AI is not an option).
For some reason, I haven't subscribed to this Substack - but I'm a subscriber now!
I like your focus on AI application development instead LLM development. The number of various LLMs is already too big for anyone to follow in a detailed fashion and when the efforts mature most of LLMs will disappear. Since most of the AI apps use LLMs via LLM API, LLMs can be viewed as runtime libraries
I think it's clear that we're going to see a consolidation of proprietary LLMs, probably 2-3 main players, with maybe a handful of industry-specific options. The open source community will likely continue to grow, but it's probably not the starting point for most devs at this point.
Instead of building in public, I have been learning in public. Starting from first principals that you would learn in undergrad like Bayesian inference, modelling using Gaussian distribution and building my way up to deep learning, while writing about what I learned with minimum math jargons on substack. I found learning these “historical” ML methods both help me understand where today’s AI comes from and also keep me grounded from all the hype. If anyone is interested, feel free to check it out.