The Hype Cycle: Rideshare Dreams and AI Realities
The allure of instant creativity and wealth fueled by AI is undeniable, but are the current models sustainable? This post dives into the hidden costs of large generative AI models, from soaring hosting prices to immense energy consumption. It advocates for a more sensible approach, championing smaller, purpose-built AI tools that offer better performance and cost-effectiveness.
Rideshare promised instant freedom, a side-hustle goldmine, and traffic miracles with a simple app tap — a tech utopia that felt real for a fleeting moment, all fueled by VC cash.
My local Toyota dealership fired all their courtesy drivers, car prices were in free fall, and self-driving cars were just around the corner.
Fast-forward to present time: the generative AI wave is crashing in, promising instant creativity and fast riches with simple AI prompts. Startups are gobbling up VC money, with talk of AI agents replacing workers everywhere. We hear of models rivaling doctorate-level expertise and coding better and faster than most developers.
But just like the last time, real economics will catch up, and the crashing wave will leave many surprised.
Can we see warning signs? Of course: for the first time in my 30-plus years of tech career, hosting prices are going up instead of down. US AI datacenters consume more energy than California and Texas combined. AI is expensive! And to make it worse, we use the most expensive generative models for trivial tasks.
The unsung heroes of this AI revolution are the small, non-generative, purpose-built LLMs. They can keep external knowledge in the embeddings instead of locked in model weights that require months of training and can only be updated once a year or so. Even better, they are fast, a few orders of magnitude less expensive, and more accurate.
Another killer trait: they are precise. No hallucinations, no made-up data points, no duplicates.
We are powering all aspects of our business domain directory using them: text translation, business identification, name and address extraction, generating descriptions, keywords, and industry categories. We look at 350 million domains each month, more often if the website is updated, touching 5–6 pages. Imagine using ChatGPT or similar for this? Our AI bill would be $4-5M a year — totally unsustainable for a small startup. And with lesser accuracy and precision.
AI VC funding will dry up, and just like Uber rides don’t cost $7 anymore, your AI expenses will force you to make changes.
The same way as my local dealership that rehired all the drivers I was used to over many years.
Contact us so we can reduce your cost for anything related to company directory, search, site text, and more.
Rideshare promised instant freedom, a side-hustle goldmine, and traffic miracles with a simple app tap — a tech utopia that felt real for a fleeting moment, all fueled by VC cash.
My local Toyota dealership fired all their courtesy drivers, car prices were in free fall, and self-driving cars were just around the corner.
Fast-forward to present time: the generative AI wave is crashing in, promising instant creativity and fast riches with simple AI prompts. Startups are gobbling up VC money, with talk of AI agents replacing workers everywhere. We hear of models rivaling doctorate-level expertise and coding better and faster than most developers.
But just like the last time, real economics will catch up, and the crashing wave will leave many surprised.
Can we see warning signs? Of course: for the first time in my 30-plus years of tech career, hosting prices are going up instead of down. US AI datacenters consume more energy than California and Texas combined. AI is expensive! And to make it worse, we use the most expensive generative models for trivial tasks.
The unsung heroes of this AI revolution are the small, non-generative, purpose-built LLMs. They can keep external knowledge in the embeddings instead of locked in model weights that require months of training and can only be updated once a year or so. Even better, they are fast, a few orders of magnitude less expensive, and more accurate.
Another killer trait: they are precise. No hallucinations, no made-up data points, no duplicates.
We are powering all aspects of our business domain directory using them: text translation, business identification, name and address extraction, generating descriptions, keywords, and industry categories. We look at 350 million domains each month, more often if the website is updated, touching 5–6 pages. Imagine using ChatGPT or similar for this? Our AI bill would be $4-5M a year — totally unsustainable for a small startup. And with lesser accuracy and precision.
AI VC funding will dry up, and just like Uber rides don’t cost $7 anymore, your AI expenses will force you to make changes.
The same way as my local dealership that rehired all the drivers I was used to over many years.
Contact us so we can reduce your cost for anything related to company directory, search, site text, and more.
