Alex RaineyAlex is an experienced CTO and founder who largely focuses on all the technical areas of My AskAI, from AI Engineering, Technical Product Management and overall Platform Development.
Created time
Jul 15, 2024 09:30 AM
Author
Tags
Publish date
Jul 15, 2024
Slug
why-goldman-sachs-ai-investment-report-misses-the-mark
Featured
Featured
Type
Article
Ready to Publish
Ready to Publish
Silly Headline from Goldman Sachs AI Report
There are obviously some very smart people at Goldman Sachs. But this is a very silly headline for their recent AI report.
Maybe it's meant to be stupid to lure us in? 🤔
But if you read the report, they're being serious. Or at least the people they interviewed are.
Here's why I think this is silly, and more importantly, wrong.
- Comparing Forecast Investment in AI vs. Present-Day Economic Upside
- Obviously, these are different and can't be compared. How can you expect upside from an investment not yet made?
- Poor ROI on Investment to Date
- Consider where this money is being spent, e.g., chips (for training/hosting models), data centers, infra upgrades, etc. — these are things that need to be built/used over many years.
- You would never expect a fast ROI for a data center. The same is true for chips, e.g., GPUs. These chips may be used to train new AI models or run AI models in the cloud for many years to come.
- Most of this spend is in long-term investments.
- We're not going to see a good ROI if we look at the present-day benefits. Strange that an investment firm like GS would fail to see this.
- Premature Point on Recent Investments
- Assuming it's a broader point, and they're only looking at investments made recently that should already be generating a return, but aren't. This still seems like a strange & premature point to make. We're still at day 0 for this technology & its adoption within enterprises.
- We're currently looking at the AOLs of the AI era, the dial-up-modem phase of the internet. Things aren't as fast or as cheap as needed. And the capabilities don't align with expectations.
- Enterprises are testing and trialing things out. Many of the leading SaaS companies are re-inventing their products to be AI-first. But this has only really happened over the last 6, possibly 12, months.
- How can we have a credible perspective on ROI this early?
- Low Productivity Gains from AI and Long Timeframe for Complex Tasks
- Some of the interviewees say productivity gains from AI will be low, and it may take 10+ years for AI models to do complex tasks.
- This is the perspective of an academic, likely out of touch with usage on the ground at real companies.
- Consider the trajectory of the top large language models — like ChatGPT — over the last 12 months. What we can see is a totally unpredictable gain in intelligence and a lowering of costs (at the same time). Every 3 months, we have models significantly better than the last set.
- It is completely impossible (unless you work at OpenAI) to know what the model's capability will be in 12 months. Forget in 3-5 years. No one in the industry can confidently say what the models will be able to do in 6-12 months.
- The reason? The pace of change has been mind-blowing. And anyone in AI worth their salt isn't making these kinds of predictions.
- They're certainly not underestimating the potential on a longer timeframe like 5-10 years.
Historical Perspective
In 1998, Paul Krugman predicted the internet's effect on the economy would be no greater than the fax machine's.
Written by
Alex Rainey
Alex is an experienced CTO and founder who largely focuses on all the technical areas of My AskAI, from AI Engineering, Technical Product Management and overall Platform Development.
