Exponential Curves: Supercomputer to Smartphone to Singularity | Turn the Lens Ep39

Or listen on:

Episode Description

Ray Kurzweil recently released his new book, *The Singularity is Nearer: When We Merge with AI*, a follow-up to his 2006 blockbuster, *The Singularity is Near: When Humans Transcend Biology*. I had the chance to see him speak at our local bookstore as part of the launch event.

A central theme of Ray’s talks is the exponential growth of computing power—how we’re constantly achieving more computations per second at a constant cost. While exponential curves are easy to discuss and graph, understanding these transformations on a visceral level is challenging. How can we fully grasp the scale of change that exponential growth reflects?

This episode tackles that question, showing the exponential advances in computing power at a fraction of the cost, using examples and everyday objects that everyone can relate to, regardless of technical background.

We’ve all heard that the smartphones in our pockets are far more powerful than the supercomputers of the past. But what does that really mean? How can we quantify, visualize, and bring this leap in technology to life in terms that feel real and relatable? We delve into comparisons—RAM, storage, and power usage—to illustrate the mind-blowing scale of progress and explore what these advances mean for the future of computing.

More importantly, we examine the impact of this rapid growth as we move forward along an ever-steepening exponential curve. Join me as I use everyday items to help us better grasp this extraordinary rate of change, connect with our technological past, and imagine the future as exponential progress continues to shape our lives in profound ways.

Exponential Curves: Supercomputer to Smartphone to Singularity | Turn the Lens podcast with Jeff Frick

#AI #GenAI #ExponentialGrowth #ComputingPower #Future #RayKurzweil #TheSingularity #MooresLaw #AIandFuture #SupercomputerToSmartphone #Change #RateOfChange #FutureOfTech #Rice #Chessboard #Curves #Understand #Visualize #TechEvolution #Exponential #5G #T1 #FLOPS #Bandwidth #Storage #RAM #Visualization #TechForEveryone #SuperComputer #SmartPhone #Singularity #ExponentialCurves #SmartphoneVsSupercomputer #Podcast #TurnTheLens

‍

Episode Links and References

Links and References

—--

Ray Kurzweil

Mentioned in the Episode

2024-Sept-04
‍‘GPT-5 to be 100x more powerful
‍By Evolving AI
‍https://evolvingai.io/p/gpt5-100x-powerful

2024-June-25
‍Ray Kurzweil with Amy Kurzweil (in person event)
‍Kepler’s Literary Foundation, Menlo Park, CA
‍https://www.keplers.org/upcoming-events-internal/ray-kurzweil
‍https://drive.google.com/file/d/1U3LbM-zocqpbFgKlXyChPgnGbqR6mpeq/view?usp=sharing

‍The Singularity is Nearer: When we Merge with AI
‍Ray Kurzweil
2024, Viking Publishing
‍https://www.thesingularityisnearer.com/
‍https://youtu.be/Dpgy0sVzKjw?si=_mkIdPcChEcYTKxk
‍https://www.amazon.com/Singularity-Nearer-Ray-Kurzweil/dp/0399562761/

‍The Singularity is Nearer featuring Ray Kurzweil | SXSW 2024 Session
‍2024-Mar-10, Austin Texas
‍https://youtu.be/xh2v5oC5Lx4?si=dCqBuBDXoEGasVkl
‍https://youtu.be/i412Ev33Dys?si=Fv3GbjMNqynIyl2Z

‍Jack Nilles: Telecommuting, Trade Offs, Resistance, Incentives | Work 20XX podcast with Jeff Frick
2024-May-07
‍https://www.work20xx.com/episode/jack-nilles-telecommuting-tradeoffs-resistance-incentives-work-20xx-podcast-with-jeff-frick-ep25

‍The Singularity is Near: When Humans Transcend Biology
Ray Kurzweil
2005, Penguin Books
‍https://www.amazon.com/Singularity-Near-Humans-Transcend-Biology/dp/0670033847/

‍Longevity Escape Velocity
‍https://en.wikipedia.org/wiki/Longevity_escape_velocity

‍Rice or Wheat and the Chessboard parable / math problem
‍https://en.wikipedia.org/wiki/Wheat_and_chessboard_problem
‍https://www.youtube.com/watch?v=t3d0Y-JpRRg&ab_channel=JillBritton
‍https://www.youtube.com/watch?v=vfH8lQwBHfQ&t=4s&ab_channel=SylviaGoodman

‍T1
‍https://en.wikipedia.org/wiki/T-carrier
‍https://www.techopedia.com/definition/9618/t1-line

‍5G
‍https://en.wikipedia.org/wiki/5G
‍https://www.statista.com/topics/3447/5g/

—---

‍The Cray-1 Supercomputer
‍https://en.wikipedia.org/wiki/Cray-1
‍https://www.computerhistory.org/revolution/supercomputers/10/7
‍https://www.britannica.com/topic/Cray-1

‍The Cray-1 Supercomputer
‍Release Date: 1975
1977 Price - $7.9M (equivalent to $39.7M in 2023)
$1M for the hard drives.
Weight - 5.5 tons
FLOPS - 160 MFLOPS
Power - 115 kW @ 208 V 400 Hz
CPU - 64 bit 80 JHz
Memory - 8 Megs
Storage - Attached - 80 Megs to 250 Megs

‍Google Pixel 9 Pro Smartphone
‍Release Date - 2024
Price - $1,499
Weight - 7 oz
CPU - Octa-core Octa-core (1x3.1 GHz Cortex-X4 & 3x2.6 GHz Cortex-A720 & 4x1.92 GHz Cortex-A520)
GPU - Mali-G715 MC7
Memory - 16 Gigs
Storage - 1 TB

—-------------

Subset of my analysis conversation with ChatGPT

—---------------

‍FLOPS

Q - If we use 'credit card purchase transactions' as a proxy for 'floating point operations' does that make sense when doing a compare?

A - Using "credit card purchase transactions" as a proxy for "floating point operations" (FLOPs) in a comparison can work conceptually, but it depends on the context of the comparison. Here are a few key considerations:

1. Scale and Throughput:
‍   - Floating Point Operations (FLOPs): These are the fundamental operations in computing, such as addition, subtraction, multiplication, or division with floating point numbers. Computers can execute trillions of FLOPs per second (teraflops or petaflops in modern systems).
   - Credit Card Transactions: These are discrete, slower, and far fewer in volume compared to FLOPs. Credit card companies handle millions of transactions per day, but the scale of individual transactions is much smaller compared to computing FLOPs.
   Comparison: FLOPs happen at a much larger scale and speed than credit card transactions. If you're comparing them, it’s crucial to normalize for the difference in magnitude (FLOPs are at the microsecond level, whereas transactions are much slower).

2. Latency vs. Throughput:
‍   - FLOPs: High throughput and low latency. Computers can process many floating point operations with negligible delay.
   - Transactions: Higher latency (waiting for approval, network communication) and lower throughput.
   Comparison: A FLOP may occur in nanoseconds, while a credit card transaction might take seconds to minutes. You’d need to account for this difference in time and the number of operations handled in parallel.

3. Analogous Use in Comparison:
‍ - As a Proxy: Using credit card transactions as a proxy could make sense if you're trying to compare the handling of discrete units—like the ability to process many operations. For example, if you're comparing how efficiently a financial system handles many transactions with how efficiently a supercomputer handles many FLOPs, you can make a conceptual analogy about their performance and scalability.

4. Use Case and Goal
‍ - Conceptual Comparison: If the goal is to compare system capabilities or illustrate the scale of operations (e.g., "How many transactions can a bank handle compared to how many FLOPs a computer can process?"), this can work as an illustrative analogy.
- Technical Comparison:If you're aiming for a precise technical comparison, this might be less accurate. Credit card transactions involve many steps (validation, communication, etc.) that don't directly align with the simplicity and speed of FLOPs in computing.

Summary:
Using credit card transactions as a conceptual proxy for floating point operations can be helpful in illustrating large-scale processing, but it's important to clarify the differences in scale, speed, and complexity. If the comparison focuses on throughput and scalability, this proxy works. For technical precision, you'd need to adjust for the large differences in speed and scale between the two.

Q - I want to compare growth and scale in computing 'horsepower' from a Cray to a modern smart phone, and seems like flops are a good measure

A - Using FLOPs is indeed a great metric for comparing the growth and scale of computing power between a Cray supercomputer and a modern smartphone. FLOPs provide a direct measure of a system’s raw processing power, especially in tasks requiring mathematical calculations (like scientific computing, machine learning, etc.).

Key Points for Comparison:

1. Cray Supercomputers (Historical Context)
‍ - The Cray-1 (released in 1976), one of the most iconic early supercomputers, could achieve around 160 megaflops (million floating point operations per second).
- Later Cray models, like the Cray-2 (released in 1985), could perform up to 1.9 gigaflops (billion FLOPs).

2. Modern Smartphones (Today)
‍ - Modern flagship smartphones, like an iPhone 15 or a high-end Android device have processors capable of achieving trillions of FLOPs per second (teraflops).
- For example, Apple's A17 Pro chip is capable of over 5 teraflops.

Growth and Scale:
- From Megaflops to Teraflops: The jump from megaflops (Cray-1) to teraflops (modern smartphones) represents an increase by a factor of millions (1 teraflop = 1 million megaflops).
- Efficiency & Portability: What's most striking is that a supercomputer once filling entire rooms has been miniaturized into a device that fits in your pocket, with thousands or even millions of times the computational power.

By comparing FLOPs, you effectively capture the extraordinary growth in raw computing power, as well as the massive advances in efficiency and scale over the past few decades.

Does this approach work for the analysis you're planning?

Q - I need a proxy measure (like credit card transactions per day per unit geography) because people have a hard time 'visualizing' a trillion anything

A - I see what you're aiming for—a more relatable proxy for comparing computational power growth in a way that's easier to visualize. Using something people encounter daily, like credit card transactions, can indeed help make the concept of trillions of FLOPs more comprehensible.

Here's an approach to craft this comparison:

‍Proxy Measure: Credit Card Transactions

We can use "credit card transactions" as a familiar, high-volume activity to help people grasp the immense scale of computational power growth.
1. Average Credit Card Transactions:
- Globally, credit card networks process about 500 million transactions per day. For simplicity, you could break this down by region or per capita.
- In the U.S., there are roughly 100 million transactions per day

2. Cray vs. Smartphone:
- Cray-1 (1976) could process 160 million FLOPs per second. Let's say for context, this means in a whole day, the Cray-1 could handle the equivalent computational load of 1 day's worth of credit card transactions in the U.S. (a rough analogy).
- Modern Smartphone (e.g., iPhone 15, 5 teraflops) can handle around 5 trillion FLOPs per second. To give a daily proxy, the smartphone could process in 1 second what would take 10,000 days of U.S. credit card transactions, or roughly 27 years of transactions, at the same speed.

3. Visualizing Growth:
‍ - To make this comparison relatable, you could say something like:
- "A Cray-1 in the 1970s could handle the computational equivalent of all U.S. credit card transactions in one day. Today’s smartphone, in just 1 second, handles more computational power than 27 years worth of U.S. credit card transactions."

4. Other Proxy Ideas:
- Google searches per day: Roughly 8.5 billion Google searches are processed every day. This could also serve as a comparison point.
- Tweets per day: About 500 million tweets are sent per day, providing another familiar activity that happens in large numbers.

This approach transforms the concept of FLOPs into something more relatable, giving people a tangible sense of just how powerful modern smartphones are compared to older supercomputers.

Does this proxy work for your needs?

Q - 5 teraflops is X times greater than 160 megaflops?

A - 5 teraflops is 31,250 times greater than 160 megaflops.

Q - Let's convert this to global (from US) - Yes, a Cray-1 supercomputer could process 147 million credit card transactions in approximately 15.3 minutes, assuming each transaction requires around 1,000 FLOPs. The actual time could vary depending on the complexity of each transaction. - redo for daily global credit card transactions

A - Let's adjust the original calculation to reflect global credit card transactions rather than U.S.-based transactions.

Global Credit Card Transactions:

Globally, there are around 1.98 billion credit card transactions per day as of 2023
Capital One Shopping

Assumptions:

We continue assuming each transaction requires 1,000 FLOPs (as we did for the U.S. case).
The Cray-1 supercomputer can process 160 million FLOPs per second.

Step 1: FLOPs Required for Global Transactions

For 1.98 billion transactions, the total number of FLOPs required is:
1.98×10^9×1,000 = 1.98×10^12 FLOPs (1.98 trillion FLOPs).

Step 2: Time for Cray-1 to Process 1.98 Billion Transactions

The Cray-1 can process 160 million FLOPs per second, so the time required to process 1.98 trillion FLOPs is:
(1.98×10^12) / (160×10^6) = 12,375 seconds.
‍
Converting 12,375 seconds to minutes:
12,375/60 ≈ 206.25 minutes.

Conclusion:

A Cray-1 supercomputer could process 1.98 billion global credit card transactions in approximately 206.25 minutes (about 3 hours and 26 minutes), assuming each transaction requires around 1,000 FLOPs. The actual time could vary depending on transaction complexity.

Q - Compare that to a modern smartphone with 5 teraflops capability

A - A modern smartphone with 5 teraflops of processing power could process 1.98 billion global credit card transactions in approximately 0.396 seconds.
This highlights the vast difference in processing power between the Cray-1 supercomputer, which would take over 3 hours, and a modern smartphone, which can handle the same task in less than half a second!

—------------

Bandwidth

Q - ok, let's compare the throughput of a t1 to a modern 5g smartphone like the Samsung s24 or iPhone 16pro

A - Let's compare the throughput of a T1 line to that of a modern 5G smartphone like the Samsung S24 or iPhone 16 Pro.

T1 Line:
- A T1 line has a throughput of 1.544 megabits per second (Mbps).
- It was a standard for high-speed internet in the past, primarily used by businesses for dedicated connections.

Modern 5G Smartphone:
- 5G networks can provide dramatically faster speeds, depending on the specific implementation (low-band, mid-band, or mmWave).
- Low-band 5G: Around 50–250 Mbps.
- Mid-band 5G: Around 300–1,000 Mbps (1 Gbps).
- mmWave 5G: Up to 4,000–10,000 Mbps (4–10 Gbps).
- The Samsung S24 and iPhone 16 Pro are capable of supporting mmWave 5G, meaning they can reach speeds up to 10 Gbps in ideal conditions.

Throughput Comparison:
- T1 line (1.544 Mbps) vs. 5G smartphone (up to 10,000 Mbps).
- The modern 5G smartphone can potentially be 6,477 times faster than a T1 line in ideal conditions.

Summary:
- A T1 line provides 1.544 Mbps, while a modern 5G smartphone can reach up to 10 Gbps (10,000 Mbps) on the fastest 5G networks.
- This means a modern 5G smartphone could offer speeds that are 6,477 times faster than a T1 line under optimal 5G conditions.

Would you like a more specific comparison for a particular use case or additional details?

Q - if a t1 cost $1,500 per month in 1976 dollars, what's the equivalent in today's dollars ?

A - The equivalent of $1,500 in 1976 dollars would be approximately $7,820 in today's dollars, accounting for an average inflation rate of 3.5% over the period from 1976 to 2024.

Q - they cost $1,500 in 1976 or 1973? When were T1s released for broad adoption?

A - T1 lines were originally introduced in 1962 by Bell Labs, primarily for use in telephone systems to carry multiple voice signals over copper wire. However, their use for data communication and broader adoption by businesses didn’t occur until the late 1970s to early 1980s. By this time, T1 lines became a standard choice for businesses needing higher-speed, dedicated connections to carry both voice and data.

Key timeline points:
- 1962: T1 technology introduced.
- Late 1970s: Businesses began adopting T1 lines for data communications.
- 1980s: T1 lines became widely used for data networking, especially for internet and business communications.

If the $1,500 cost you mentioned was referring to the mid-1970s, it would likely be from around 1976 rather than 1973, as that's when T1 lines started to be commercially used for data transmission, though they were still somewhat specialized.

Would you like to explore any specific details further?
‍

Q - if a typical residential garden hose represented at t1 capacity, how many fire hoses to represent 10 Gbps 5g?

A - Let’s compare the throughput of a residential garden hose representing a T1 line with a fire hose representing 10 Gbps 5G speeds.

Step 1: Throughput of a T1 Line
- A T1 line has a throughput of 1.544 Mbps.

Step 2: Throughput of 5G (10 Gbps)
- A modern 5G smartphone with 10 Gbps (10,000 Mbps) capability.

Step 3: Scaling Factor (Garden Hose to Fire Hose)
We need to calculate how many times 10 Gbps is greater than 1.544 Mbps:
10,000 Mbps / 1.544 Mbps = 6,478 times

Conclusion:
If a residential garden hose represents the capacity of a T1 line, you would need approximately 6,478 fire hoses to represent the throughput of 10 Gbps 5G. This illustrates the massive difference in data capacity between a T1 line and modern 5G technology.

Q - but don't 2.5 inch fire hoses, at greater pressure than a garden hose change the math? Need to include PSI in the calculation

A - You're absolutely right! To provide a more accurate comparison between a garden hose and a fire hose, we need to account for both the diameter of the hoses and the pressure (PSI) at which the water flows, since both factors affect the overall throughput (similar to how both bandwidth and signal strength affect data throughput).

Let's break this down:

Step 1: Throughput of a Garden Hose (T1 Line Equivalent)
- A standard garden hose has a diameter of around 5/8 inch and operates at about 40 PSI.
- At this pressure, a garden hose typically delivers around 17 gallons per minute (GPM).

Step 2: Throughput of a Fire Hose (5G Equivalent)
- A 2.5-inch fire hose typically operates at a pressure of around 100 PSI.
- At 100 PSI, a 2.5-inch fire hose can deliver approximately 250 GPM.

Step 3: Comparing Throughput
Let’s calculate how many fire hoses would be needed to match the flow rate difference between a garden hose and fire hoses, corresponding to the throughput difference between a T1 line and 10 Gbps 5G.

1. Garden Hose (T1 Line) Flow Rate: 17 GPM at 40 PSI.
2. Fire Hose (5G 10 Gbps) Flow Rate: 250 GPM at 100 PSI.

Step 4: Calculate the Fire Hose Multiplier
We already calculated the data throughput difference as 6,478 times greater for 5G vs. T1.

Now, let’s find out how many fire hoses would match that multiplier if 1 garden hose represents 1 T1 line:
((6,478 (throughput difference) / ((250 GPM (fire hose) / 17 GPM (garden hose))) = 440.12 fire hoses

Conclusion:
If a garden hose at 40 PSI represents a T1 line, you would need approximately 440 fire hoses operating at 100 PSI to represent the throughput of 10 Gbps 5G. This accounts for both the pressure and the flow difference between a garden hose and a fire hose, much like accounting for both bandwidth and speed in data transmission comparisons.

—-----------

‍RAM

Q - recalculate with the desk surface being 72" x 30" - So if 4 meg was represented by a the area of the top of standard desk, how many desks to represent the area of 16 gigs? - How many square feet is the total? equivalent to what familiar surface area?

A - If 4 megabytes of RAM is represented by the surface area of a standard desk measuring 72" x 30", you would need approximately 3,999 desks to represent the area of 16 gigabytes of RAM.

Total Square Footage:
- The total area covered by these desks would be approximately 59,985 square feet.

Familiar Surface Area Comparison:
- This is roughly equivalent to the size of a football field, which typically covers about 57,600 square feet (including the end zones). So, it’s just a bit larger than a full football field!

Q - 72" x 30" is how many square feet?

A - A desk with dimensions of 72 inches by 30 inches has a surface area of 15 square feet.

—--------------

Storage

Q - 1 TB is X time larger than 250 megs?

A - 1 TB is X time larger than 250 megs?

Q - if we switch to documents, how many pages of documents fit in 250 mg

A - To estimate how many pages of documents fit into 250 MB, we first need to approximate the size of a typical document page.

Assumptions:
- A typical text document (e.g., a Word or PDF file) with standard formatting and no images or complex formatting is about 50 kilobytes (KB) per page.
- This could vary slightly depending on fonts, formatting, and whether images are embedded.

Step 1: Convert 250 MB to KB:
- 1 MB = 1,024 KB.
- Therefore, 250 MB = 250 × 1,024 = 256,000 KB.

Step 2: Calculate Pages:
- If each document page is 50 KB, then the number of pages that can fit into 250 MB is:
256,000 KB / 50 KB per page = 5,120 pages

Conclusion:
A 250 MB drive could hold approximately 5,120 pages of standard text documents.

Q - how many pages can an old school 2 drawer file cabinet hold?

A - An old-school 2-drawer file cabinet typically holds about 2,500 to 3,000 pages per drawer, depending on the thickness of the paper and how tightly packed the documents are. For both drawers combined, it would hold around:
2 drawers X 2,500 to 3,000 pages = 5,000 to 6,000 pages

Conclusion:
A standard 2-drawer file cabinet can hold 5,000 to 6,000 pages of documents.

Q - so how many file cabinets required to store 250 megs of info compared to 1TB?

A - Let’s break it down step by step to determine how many 2-drawer file cabinets would be needed to store 250 MB and 1 TB of information in terms of document pages.

Step 1: Pages for 250 MB of Information
From the previous calculation, 250 MB can store approximately 5,120 pages of documents.

Step 2: Pages for 1 TB of Information
- 1 TB = 1,024 GB = 1,048,576 MB.
- Since 250 MB stores 5,120 pages, the total number of pages for 1 TB would be:
(1,048,576 MB / 250 MB) x 5,120 pages = 21,457,664 pages

Step 3: File Cabinets for 250 MB
Since a 2-drawer file cabinet holds about 5,000 to 6,000 pages, for 250 MB (which holds 5,120 pages), you would need 1 file cabinet

Step 4: File Cabinets for 1 TB
For 1 TB of data (which holds 21,457,664 pages), you would need:
21,457,664 pages / 5,500 pages per file cabinet (average) ≈ approx 3,901 file cabinets.

Conclusion:
- To store 250 MB of information, you’d need 1 file cabinet.
- To store 1 TB of information, you’d need approximately 3,901 file cabinets.

This comparison highlights the incredible storage density difference between the two eras!

Q - side by side, how wide would 3,900 file cabinets be?

A - To calculate the width of 3,900 file cabinets placed side by side, we need to know the typical width of a 2-drawer file cabinet.

Assumption:
- A standard 2-drawer file cabinet is typically about 18 inches wide.

Step 1: Convert Width to Feet
- 18 inches = 1.5 feet per file cabinet.

Step 2: Calculate Total Width for 3,900 File Cabinets
If each cabinet is 1.5 feet wide, then the total width for 3,900 file cabinets would be:
3,900 X 1.5 = 5,850 feet

Step 3: Convert to Miles
- 1 mile = 5,280 feet
‍- Therefore, the total width in miles is:
5,850 / 5,280 ≈ 1.11 miles.

Conclusion:
If placed side by side, 3,900 file cabinets would stretch approximately 1.11 miles wide!

—--------------

Other Ray Kurzweil Books

‍Danielle: Chronicles of a Superheroine
‍Ray Kurzweil
2019, Wordfire Press
‍https://www.amazon.com/Danielle-Chronicles-Superheroine-Ray-Kurzweil/dp/1614756392/

‍How to Create a Mind: The Secret of Human Thought Revealed
‍Ray Kurzweil
2012, Viking
‍https://www.amazon.com/How-Create-Mind-Thought-Revealed/dp/0670025291/

‍Transcend: Nine Steps to Living Well Forever
‍Ray Kurzweil & Terry Grossman
2010, Rodale Books
‍https://www.amazon.com/Transcend-Nine-Steps-Living-Forever/dp/1605299561/

‍Fantastic Voyage: Live Long Enough to Live Forever
‍Ray Kurzweil & Terry Grossman
2004, Rodale Books
‍https://www.amazon.com/Fantastic-Voyage-Live-Enough-Forever/dp/1579549543/

‍The Age of Spiritual Machines: When Computers Exceed Human Intelligence
‍Ray Kurzweil
2000, Penguin Books
‍https://www.amazon.com/Age-Spiritual-Machines-Computers-Intelligence/dp/0670882178/

‍The Age of Intelligent Machines
‍Ray Kurzweil
1992, MIT Press
‍https://www.amazon.com/age-intelligent-machines-Ray-Kurzweil/dp/0262111217/

‍Ray Kurzweil Wikipedia
‍https://en.wikipedia.org/wiki/Ray_Kurzweil

‍Ray Kurzweil Amazon Author Page
‍https://www.amazon.com/stores/Ray-Kurzweil/author/B001ILHHDS

—

Other Ray Kurzweil appearances

—-

2024-Oct-24
‍The Man Who Predicted AGI Decades Ago w/ Ray Kurzweil | Ep #125
‍Peter H. Diamandis YouTube Channel
‍https://www.youtube.com/watch?v=xqS5PDYbTsE&t=0s&ab_channel=PeterH.Diamandis

2024-Jun-27
‍The Last 6 Decades of AI — and What Comes Next | Ray Kurzweil | TED
‍TED YouTube Channel
‍https://www.youtube.com/watch?v=uEztHu4NHrs&ab_channel=TED

2024-Apr-09
‍Ray Kurzweil: Does AI Make Immortality Possible?
‍Center for Natural Artificial Intelligence YouTube Channel
‍https://www.youtube.com/watch?v=cipDXOhx9X4&ab_channel=CenterforNaturalandArtificialIntelligence

2022-Dec-01
‍Ray Kurzweil Wants To Put Nanobots in Our Bedroom
‍StarTalk YouTube Channel
‍https://www.youtube.com/watch?v=u8yC2BtcAoQ&ab_channel=StarTalk

2022-Nov-22
‍What The 2030s Will Look Like with Ray Kurzweil
‍StarTalk YouTube Channel
‍https://www.youtube.com/watch?v=ZJF6GoE-R8s&ab_channel=StarTalk

2022-Sep-17
‍Ray Kurzweil: Singularity, Superintelligence, and Immortality | Lex Fridman Podcast #321
Lex Fridman YouTube Channel
‍https://www.youtube.com/watch?v=ykY69lSpDdo&ab_channel=LexFridman

2018-Dec
‍Ray Kurzweil on what the future holds next | The TED Interview by Chris Anderson
TED
‍https://www.ted.com/talks/the_ted_interview_ray_kurzweil_on_what_the_future_holds_next?subtitle=en

2014-Jun-02
‍Ray Kurzweil: Get ready for hybrid thinking
‍TED YouTube Channel
‍https://www.youtube.com/watch?v=PVXQUItNEDQ&ab_channel=TED

2009-Feb
‍A university for the coming singularity
‍TED
‍https://www.ted.com/talks/ray_kurzweil_a_university_for_the_coming_singularity?subtitle=en

2007-Jan-12
‍The accelerating power of technology | Ray Kurzweil
‍TED YouTube Channel
‍https://www.youtube.com/watch?v=IfbOyw3CT6A&ab_channel=TED

—------------

‍‍Disclaimer and Disclosure

All products, product names, companies, logos, names, brands, service names, trademarks, registered trademarks, and registered trademarks (collectively, *identifiers) are the property of their respective owners. All *identifiers used are for identification purposes only. Use of these *identifiers does not imply endorsement. Other trademarks are trade names that may be used in this document to refer to either the entities claiming the marks and/or names of their products and are the property of their respective owners.

We disclaim proprietary interest in the marks and names of others.
No representation is made or warranty given as to their content.
The user assumes all risks of use.

Episode Transcript

English Transcript

Exponential Curves: Supercomputer to Smartphone to Singularity | Turn the Lens podcast with Jeff Frick Ep39

---

Jeff Frick:
Hey, Jeff Frick here coming to you from the future.

You know, I'm finalizing the edits on this latest episode on exponential curves and the fact that we have a supercomputer in our pocket and trying to visualize or create some comparisons of things that are easier to grok, so that we can really understand what's happening here.

And as I'm going through it, there's a lot of numbers, there's a lot of math. But ultimately, I just wanted to get to some things that we can kind of grasp or visualize to try to quantify better how much technology has changed.

And what's important is really the concept of how much it's going to change in the future. So you'll hear a lot of numbers. Try not to get too wrapped up in the numbers. I'll put all the math in the notes so you can go and run your own numbers.

But the concepts to really focus on, you know, we've gone from a Cray supercomputer in 1973 to our mobile phones, and our mobile phones are thousands, if not tens of thousands of times more powerful for fractions of the cost, fractions of the weight, fractions of the energy consumption.

So if you think about where we were in 1973 and where we are today, and now you start to think about where are we going to be two years from now, three years from now, five years from now, ten years from now, on these exponential curves, it really begs the question, what will we be carrying around in our pockets or in our brains or in our heads, or however we're going to be connected to this technology down the road?

You know, ChatGPT came onto the scene two Novembers ago. They're already talking about ChatGPT. I think it's [ChatGPT] five that’s coming out soon and is a hundred times more powerful. So it's two years, 100 times. So 10x twice, to get here.

So the most important part of this episode is trying to visualize how fast this stuff is moving and try to think about where it's going to be going.

We didn't even get into some of the interesting things that Ray Kurzweil talks about, like mortality escape velocity, digital twins when applied to biology and people. I'll put a bunch of his links in the notes below as well, so you can watch some of his interviews and videos, because it’s a really interesting concept, and I think it's important that we always kind of think of where we are today as kind of where it's going to end, and it's not at all.

You look back in any technology, and just a few years back, you know, what we were doing seems almost juvenile sometimes. And then you think about the implications, implications in something like medical technology where, you know, artificial knees now are an outpatient procedure. Artificial hips are practically an outpatient procedure.

Stents and pacemakers and all these things that we kind of take for granted were amazing technological developments in the not-too-distant past. So what does the future hold? So keep that in mind as we go through some of the numbers and some of the illustrations, don't get hung up on the math. Run the math yourself. Try to think of some interesting ways to compare what the power of computing was 50 years ago compared to what the power of computing is today, and try to think about what it’s going to be five years from now, 10 years from now, 20 years from now, even 50 years from now. It’s a very different future.

So with that, I’ll send you back to the past, where we recorded this episode, which I’ve been trying to get out for so, so long. So I’m just going to hit publish a tribute to my buddy Keith Townsend, and get it out there. Hopefully, you enjoy it. Don't get too hung up on the numbers and the math. Really try to focus on the illustrations, and if you've got better ones, I'd love to hear it.

Thanks for listening. Thanks for watching.

Cold Open:
[clap, clap, clap]
OK, in three, two, one
‍

Jeff Frick:
Hey, welcome back, everybody. Jeff Frick here coming to you from the home studio for another episode of *Turn the Lens*. Going solo today, but I’ve got Ray Kurzweil here in kind of a virtual spirit, a virtual sense. Unfortunately, he's not here in person. But he recently came out with his latest book called *The Singularity Is Nearer* (2024, Viking), and it’s pretty wild. He’s out doing the book tour, and I had the good fortune of seeing him here at one of our local bookstores.

If you get the chance, I would strongly suggest you do. There’s another great interview where he talks about a lot of the concepts. If you don’t have the chance to see him, I’ll put a link to that in the description below.

This whole concept of exponential curves is really hard for us as humans to grok. I know I’ve mentioned it before, but I thought, you know, let’s go through some specifics and really see if we can give you some tools to see, to make it more tangible, digestible, and recognizable of what exactly is going on here.

So Ray always shows this graph showing the scale and the power of computation over time, from 1935 to the present day. It’s up and to the right, and you say, “Wow, looks like a pretty steady curve.” The reason it doesn’t clearly express the exponential-ness is because it's performance over price.

This growth continues at a rapid rate. If you look to the left, you’ll see the scale goes from 0.0000001, it’s like five zeros and a one as a fraction, to the top, which isn’t thousands, millions, or billions, but trillions. You go from a decimal place with six places to a trillion at the top. So how do you even start to understand that?

What I wanted to do is just make a real simple comparison against a couple of devices that most of us have probably heard of, if we're not very familiar with them. And that is the Cray-1 supercomputer, which came out in the late 70s, and a modern smartphone, like my Galaxy S24. So we'll go through some of the basic computing measurement parameters, try to convert it into something that's easier to grasp, and really use that to try to illustrate how far things have gone. Or, excuse me, how far things have come since the late 70s, which seems like a long time ago, 50 some odd years ago, thereabouts. But it's really not that long.

And what's important is that the time going forward, we're only on a steeper part of the curve. So let's go through some of these things.

So first off, let's talk about computing horsepower. And that is measured in something called "flops," which stands for floating point operations per second, which is doing the math, the computer's doing the math.

A Cray-1 in the late 70s, and I'll put all these numbers in the notes below, could do 160 megaflops. A modern smartphone does 5 teraflops. And a teraflop is a million megaflops. So you go from 160 mega to 5 teraflops. So how much more is that? And it's more, it's 31,000 times more.

31,000 times more between a Cray-1 supercomputer and your phone. So again, it's just hard to even understand what that means. So what I decided to do is let's look at that in terms of how long it would take to compute all the credit card transactions that happen in the world today [in a single day].

Right now, according to ChatGPT, there's about 1.98, close to 2 billion credit card transactions that happen around the globe every day. So let’s just call it 2 billion in round numbers.

At 160 megaflops, it would take approximately 3.5 hours to process all those transactions. So basically, think old-school batch processing. All the credit cards run all day long. You want to run it through the machines that you can, put all the money where it's supposed to go, that would take about 3 hours and 26 minutes with the state-of-the-art machine in the late 70s.

Using your modern smartphone at 5 teraflops, that same calculation would take, how long do you think? Again, the comp is 3 hours and 26 minutes. Modern smartphone? It would take 0.39 seconds, less than half a second to do the equivalent work, 3.5 hours to less than half a second. That's how much smarter your smartphone is than the state-of-the-art computing not that long ago. So that's flops as a proxy for computing power.

Let's look at working area, or RAM. And RAM, I always like to think of as kind of the tabletop of your computer, how much stuff can you have open at one time? How many things can it be working on?

And the old Cray-1 had 4 MB of RAM, 4 megs. Your modern smartphone, again, these are top-end smartphones, can have 12 GB to 16 GB, so let's just go with 16 gigs.

So how much bigger is 16 GB than 4 MB? Well, it's 4,000 times. 4,000 times.

So if we think about that in terms of workspace, let's just say your modern desk. And I just looked up a catalog for a modern standing desk. It's 72 inches by 30 inches, which is about 15 square feet of working area, a pretty good-sized desk.

So if that represented 4 MB, how much desk space would you need to represent 16 GB? Compared to one desk on a Cray-1, how much desk space is that equivalent to in your modern smartphone at 16 GB?

And it's basically a football field. It's like 6,000 square feet compared to 15 square feet, or, excuse me, 60,000 square feet compared to 15 square feet. So it's basically comparing a desk working area to a football field, including the end zones, which come in at about 60,000 square feet.

Huge increase, 4,000 times.

Okay, let's talk about storage. It turns out the Cray-1 didn’t actually have internal storage. They used attached storage. But according to Wikipedia, it generally would be between 80 MB and 250 MB. So we're going to use 250 MB as the high end and compare that to your modern smartphone, which has 1 TB.

So how do we think about that in terms of something that we can visualize? Because the raw number, again, is 4,000 times, which is hard to grok, something 4,000 times bigger.

So I thought, okay, well, let's think of it in terms of documents and storage space. It turns out that one two-drawer file cabinet, which is, I still have lots in my garage, and we used to have them at the office all the time, so just a standard 2-drawer file cabinet at a calculation of 50 kilobits per page in a PDF, a two-drawer file cabinet can hold about 5,000 pages. So let’s just call it 5,000 pages for quick and dirty math. And that's about a foot of shelf space. Times that by 4,000.

Okay, that's about 4,400 feet of bookshelf space or 0.85 miles, almost a mile. So instead of a foot of bookshelf space, you now have a mile of bookshelf space. That's the difference between how much can be stored on a 250 MB drive compared to a 1 TB drive.

Okay, so stay with me. Let's talk about power consumption.

The Cray-1 supercomputer used 115 kW, 115 kW. As a comparison, the average home in the U.S. today uses about 30 kW per day. 30 kW per day. So it's four times more [per day] than you use in your home [in a month].

A modern smartphone, according to ChatGPT, when it's doing typical tasks, uses between 2 watts and 4 watts, not 115 kW, just 2 to 4 watts. When it's working hard, it could spike to 8 to 12 watts.

So it's literally 0.000008 times more efficient to do, remember, 4,000 times more computing power. These are just crazy numbers and really hard to kind of get your head around.

But then here's the kicker. On top of it all, the Cray-1 in 1977 cost $7.9 million. So that's approximately $40 million in today’s money, $40 million.

What do you pay for your smartphone? For a nice top-end Samsung or Apple? $1,000, $1,200, maybe $1,500? It's crazy for doing 4,000 times more work than what would cost $40 million based on where the Cray-1 was years ago.

So $1,000 versus $40 million for 4,000 times the computational power than what you had back then.

And then just the last little piece that we measure in computing is bandwidth.

So back in the day, in the late 70s, state of the art was what was called a T1, which was about 1.4 Mbps was the best performance on its best day compared to today. Most of the phones today are all 5G, and 5G has a range, but most of the top-end ones, and again, we've been talking about the top-end phones, can do 10 Gbps.

So how do you compare the bandwidth, the throughput of 1.4 Mbps to 10 Gbps in 5G?

And I really struggled with ChatGPT to find a good comp. So I finally got to the point where I said, if a garden hose at 5/8 inches and 40 psi (pounds per square inch) at a typical house puts out about 17 gallons per minute, so if that represented a T1, state of the art back in the late 70s, what type of hoses would you need to represent the throughput of 5G?

I tried to find like a single big giant hose, like a construction pipe, but struggled with lots of things. So finally we got to, okay, what if we compare it to fire hoses?

So a fire hose is about 2.5 inches in diameter and those go at about 100 psi. And so a fire hose at 2.5 inches and 100 psi puts out 250 gallons per minute.

So we're comparing, that’s 250 gallons per minute. The garden hose has 17 gallons per minute. To get it up to the size of 5G, it's 6,000 times more, 6,000 times more.

So it would take 440 fire hoses to be 5G compared to one garden hose for T1. That is so much more throughput.

And then again, here's the kicker, it's all divided by price. Back in the 70s, it would cost you about $1,500 a month. Today, if you account for inflation, that $1,500 a month turns into about $7,800 a month.

So imagine if your phone bill, just for your connectivity, was $7,800 a month for only 1/6,000 (0.0001667) of the throughput.

So again, I had kind of ChatGPT help me write a little summary document. It’s crazy.

So comparing your modern smartphone to a Cray-1 supercomputer, you get about 4,000 times the working area in terms of the RAM. You get 31,000 times more horsepower in terms of the flops. You get 4,000 times more storage, using 0.00001 of the power, and it costs you, for the unit, $1,000 compared to $40 million. And for your monthly service fee, call it $100 compared to close to $8,000 for the Cray-1.

So I don't know if that helps illustrate what's happening. And again, as I've said many, many times, today’s the slowest day of technological change for the rest of your life.

I think the ChatGPT event two years ago, two Novembers ago, is really kind of exponential curves thrown in our face, that we can now, via our phone and 5G, talk to a supercomputer that lives in the cloud, that understands natural language better than ever, and will talk back to you, answer in a human voice, and give you the information.

And it's only getting faster and faster every day.

So I hope this was helpful. I don’t know if it’s helpful, I struggle with this. I think humans' ability to grok exponential curves is a big part of the challenge that we have in technology.

As I've talked about in other videos, there’s human biological development and evolution, which is pretty slow. There's also civilization's evolution over the last many thousands of years, which has also been relatively slow. It wasn’t that long ago that you probably presumed in your career that you'd have the same career that your folks had, which was the same that their folks had, which was the same that their folks had, which was farming for most people, not that many generations ago. And you would presume your kids would do the same.

But even civilization’s change speed is nothing compared to the exponential speed that we’re seeing in computing.

And again, another really illustrative story is the tale of the kid and the chessboard and the grain of rice [or wheat]. It's an old fable where some kid beat the king in chess, and the king said, “How could I pay you back?” And he said, “I just want one grain of rice on the first day on the first square of the chessboard.”

The second day, I want you to double it, so two grains of rice. Third day, double it, four grains of rice. Fourth day, double it, eight grains of rice. And so on for the 64 days, for the 64 squares on the chessboard.

And if you do the math, and there are a lot of great videos out there talking about it, by the time you get to the 64th day, you basically would be covering the entire country of India deep in grains of rice. There’s not that much rice. So exponential curves grow fast. And that’s just doubling, that’s not multiplying it by ten.

It’s a tough thing to grok, and it’s a good thing to think about. Hopefully, this is interesting. Maybe it’s not interesting. I'll throw some visuals up, even the visuals, trying to get ChatGPT to help me do the visuals, to do the comparison between one garden hose flow versus however many fire hoses I said, it can’t do it. To have a mile's worth of file cabinets or bookshelf space, these are really difficult concepts to grasp, and we're not really built for it, which is why there are so many advances in technology on so many fronts. And they just sometimes are surprising how fast they move.

And as I've always said, you only know about the technological advances in the industries in which you play, but it's happening across all industries concurrently, and you just don't see it all the time because it’s not stuff that you're that aware of. So, interesting times ahead.

Today’s the slowest day of technological change for the rest of our lives. It’s only going to get faster.

Hope this was helpful. Hope it was entertaining. I've been wanting to do this for a long time. As Keith Townsend said one time when we were together, he had a post that he struggled with, trying to get it down. I've struggled with this. I've struggled with trying to get the concepts together that are illustrative, but I don’t want to miss the window and not get the post up, which is something Keith and I talked about. That’s the worst of all worlds.

So hopefully you enjoyed it. Hopefully it helped. I think it's interesting.

Thanks again for watching. Thanks for tuning in on YouTube. Thanks for listening in on the podcast. Appreciate, appreciate you being there. And reach out, say “Hi.”

Signing off from the home studio, Jeff Frick. See you next time. Take care.

---

Disclaimer and Disclosure

Jeff Frick

Entrepreneur & Podcaster

Jeff Frick has helped tens of thousands of executives share their story.

Disclaimer and Disclosure

‍Disclaimer and Disclosure

We disclaim proprietary interest in the marks and names of others.
No representation is made or warranty given as to their content.
The user assumes all risks of use.

‍

Latest episodes

Browse Episodes

Google NotebookLM: AI, Ingest, Create a Podcast | Turn the Lens Ep38

As a longtime media producer, podcast host, and guest, I was really curious to hear how the ‘create a podcast’ would convert this stack of data into a conversational podcast. - Jeff Frick



Listen Now