Is there a worthwhile point of comparison to the 3 D's of robotics which categorizes the tasks that robots are designed to take over from humans as Dull, Dirty, or Dangerous? There are equivalencies between what robotics and AI (LLM's) can offload from humans, albeit more in the 'cognitive manual work' area for AI.
Yep, absolutely. I'm no expert in what's possible in robotics, but I think transfer between tasks is going to be way harder — maybe in the same way that playing football and basketball aren't the same skills? Whereas transfer between adjacent functions like technical support and customer success will be easier in AI.
As a result, I think the LLM versions are going to be more focused on the dull side of things because you can automate away lots of busy work in adjacent functions with a single tool.
This is a fantastic because it is very much focused on the practical. Coming from a GenAI company to the larger org that I work at now, that first prediction was especially what I noticed an ACSM with Data and AI skills.
The only area I would push back on is one of your sub points within barbell distribution point. Ever since I became privy to the advent of AI as a growing economy of scale, my immediate prediction was that any company working in cloud infrastructure, data, or both has the ability to pivot toward a "data enabler" function within the AI ecosystem.
Extensive experience in these areas (assuming they have the appropriate talent to support it) will create an intermediary enterprise resource between AI products and enterprise orgs who bought the hype and quickly came to the sobering realization that their data and tech architectures aren't good enough to support an effective enterprise AI solution.
I'm picturing this manifesting as an offering within a larger suite of services by firms like McK, ZS, or ACN, while concurrently becoming its own value proposition for new startups or value stories for existing organizations specializing in data and cloud technologies.
I'm new to the space (and to Substack), so please feel free to correct me if I'm off base!
Thanks for the thoughtful comment, Adefoluke! I totally agree that companies that already are operating at scale in infrastructure in particular will have a lasting advantage. I'm a little more skeptical about the moat that data will offer over time, as the internet gets saturated with more and more AI-generated content (and LLM inference becomes cheaper). I think interaction data that newer AI platforms collect will be higher-value than generic text.
To clarify, the point we made about the barbell distribution was about AI companies specifically. In other words, value will accrue to the bottom of the AI stack (foundation model companies and hardware providers) and the top (application providers) but be relatively thin in the middle.
Love a post Sapiens books dunk
I've been disappointed one too many times!
Is there a worthwhile point of comparison to the 3 D's of robotics which categorizes the tasks that robots are designed to take over from humans as Dull, Dirty, or Dangerous? There are equivalencies between what robotics and AI (LLM's) can offload from humans, albeit more in the 'cognitive manual work' area for AI.
Yep, absolutely. I'm no expert in what's possible in robotics, but I think transfer between tasks is going to be way harder — maybe in the same way that playing football and basketball aren't the same skills? Whereas transfer between adjacent functions like technical support and customer success will be easier in AI.
As a result, I think the LLM versions are going to be more focused on the dull side of things because you can automate away lots of busy work in adjacent functions with a single tool.
This is a fantastic because it is very much focused on the practical. Coming from a GenAI company to the larger org that I work at now, that first prediction was especially what I noticed an ACSM with Data and AI skills.
The only area I would push back on is one of your sub points within barbell distribution point. Ever since I became privy to the advent of AI as a growing economy of scale, my immediate prediction was that any company working in cloud infrastructure, data, or both has the ability to pivot toward a "data enabler" function within the AI ecosystem.
Extensive experience in these areas (assuming they have the appropriate talent to support it) will create an intermediary enterprise resource between AI products and enterprise orgs who bought the hype and quickly came to the sobering realization that their data and tech architectures aren't good enough to support an effective enterprise AI solution.
I'm picturing this manifesting as an offering within a larger suite of services by firms like McK, ZS, or ACN, while concurrently becoming its own value proposition for new startups or value stories for existing organizations specializing in data and cloud technologies.
I'm new to the space (and to Substack), so please feel free to correct me if I'm off base!
Thanks for the thoughtful comment, Adefoluke! I totally agree that companies that already are operating at scale in infrastructure in particular will have a lasting advantage. I'm a little more skeptical about the moat that data will offer over time, as the internet gets saturated with more and more AI-generated content (and LLM inference becomes cheaper). I think interaction data that newer AI platforms collect will be higher-value than generic text.
To clarify, the point we made about the barbell distribution was about AI companies specifically. In other words, value will accrue to the bottom of the AI stack (foundation model companies and hardware providers) and the top (application providers) but be relatively thin in the middle.