One thing I’ve noticed is that institutions are staffed with institutionalists, who, by training and deportment, view stasis as the default assumption. They can’t conceive of futures radically different from the present.
That is a very good point. I am reminded of something I heard from educator David Warlick back in 2007. “We are the first generation in history who knows that we have to prepare the next generation for a future that we know we can’t imagine.” I’ve been telling that to people every chance I get ever since.
A year after ChatGPT came out, I was having coffee with a friend who works in a government ministry overseeing universities. I asked him whether anyone was working on policy and guidance for institutions about generative AI, given that our mutual friends who taught freshman courses were being swamped by LLM-generated cheating. He said "No, no one has even brought it up."
Yeah, and it's a real issue. The kinds of people who become institutionalists--lawyers, academics, regulators, politicians, etc.--are, in general, not very technically literate. And so they reason about AI and its possible future based on an entirely inappropriate and irrelevant set of priors.
I don’t know that it’s a matter of technological literacy so much as a matter of imagination. Isaac Asimov refused to fly. Jules Verne wrote longhand. Institutionalists seem to suffer from a certain mental ossification.
Good point. A related point I’d add to that is many institutions are afraid of experimenting. They wait until more is known and business cases are created/approved before even beginning to play around with new technologies. As many have said, there are only two times to adopt GenAI — either too early or too late. The muscle memory gained from this early experimentation will pay real dividends once the models are further along
“No one ever got fired for buying IBM.” versus “Move fast and break things.“ I once showed a prototype to my boss, the CEO. He pointed out all the things that were wrong with it. I explained that prototypes were a series of learning exercises. He replied, “No, it’s a series of mistakes!“ and stormed away. I started looking for a new job that night.
As you pointed out, even if all advancements would halt immediately, we would still hold enough disruptive potential in our hands to innovate for years.
There are so many things to figure out and fundamental problems to solve that go far beyond the raw performance of these models.
AGI is the north star for a few groups of people, all for different reasons, but it's not really what we should focus on in most of the discussions imo.
Just like with the unreasonable effectiveness of deep learning itself, I think we will only be able to perceive the tipping point in hindsight.
Now, should we stop talking about it altogether? Absolutely not. We should be prepared for all eventualities.
Regarding GPT-4o's confusion about 9.11 and 9.9 (easy to reproduce in Claude 3.5 as well (but not Gemini) I tried this:
Simulate 2 personas discussing the following question. One is a very good problem solver and the other is a sceptic always looking for flaws. The question is: 9.9 or 9.11 - which is bigger?
### Persona 1: The Problem Solver
**Problem Solver:** Hey, have you ever thought about the comparison between 9.9 and 9.11? It's pretty straightforward to determine which one is bigger.
### Persona 2: The Skeptic
**Skeptic:** Really? I think there could be more to it than meets the eye. How can you be so sure?
**Problem Solver:** Well, it's simple arithmetic. When you compare two numbers, you look at their digits from left to right. The first digits in both numbers are 9, so we move to the next set of digits.
**Skeptic:** Sure, but what if there's a mistake in the way we compare them? What if the decimal system itself has flaws?
**Problem Solver:** The decimal system is quite robust. Let’s break it down. In 9.9, the digit after the decimal is 9. In 9.11, the digit after the decimal is 1. Since 9 is greater than 1, 9.9 is clearly larger.
**Skeptic:** But aren't there more digits to consider in 9.11? After all, 11 seems like it could make a difference.
**Problem Solver:** I see where you're coming from, but the additional digit in 9.11 doesn’t outweigh the significance of the first digit after the decimal. The comparison is made digit by digit. After comparing the first decimal place, we don’t need to go further because 9 is already greater than 1.
**Skeptic:** What if we were comparing more complex numbers? Could the presence of additional digits change the outcome?
**Problem Solver:** In cases where the first decimal digits are equal, additional digits would matter. For example, if we were comparing 9.19 and 9.11, we'd look at the second decimal place, where 9 is greater than 1, so 9.19 is larger. But in the case of 9.9 and 9.11, the comparison ends at the first decimal place because 9 is greater than 1.
**Skeptic:** Hmm, I suppose that makes sense. But what about rounding errors or representations in different computer systems? Could they affect the comparison?
**Problem Solver:** In practical applications, precision is key. Standard decimal comparison in most systems would handle this correctly as long as we’re dealing with the same precision. Both 9.9 and 9.11 are represented accurately and compared without error.
**Skeptic:** Alright, I see your point. But I still think it's always good to be cautious about such comparisons. You never know when an unexpected factor might come into play.
**Problem Solver:** True, skepticism is healthy. But in this specific case, we can confidently say that 9.9 is bigger than 9.11 by straightforward decimal comparison.
Great piece. It does seem firms are systematically failing to plan for even a modest probability of improvement.
But - how well do firms systematically plan for *anything*? Most "scenario analysis" I've seen at big firms seems unserious and doesn't really improve decisions.
People who believe it is possible to create AGI are simply following the machine/computer model of human intelligence. They're prisoners of the cognitive model. Of course, using that model, AI will replicate or exceed the brain. But human intelligence is infinitely more than the brain. Once we find out, AI will simply turn out to be a very useful tool.
>90% of businesses fail to perform deep planning of any kind - the fact that they don't do any for AI has nothing to do with the subject matter but their own failures to be planners.
So, failure to plan for AI among those who plan at all could be important - except that business planning quantifies risks and rewards. Central to all such plans is the ability to execute.
A primary preventer of business planning for those who actually want to perform it is government regulation (actual, threatened or absent-and-needed). If we want clarity around planning for AI, we need clarity around the permissive bounds of AI.
AI scientists cannot help this or themselves. Like all technologists, they put blind faith and total effort into achievement, damn the consequences. The excuse is that if I don't someone else will (and gain from it whether money, fame, etc.) so why not me?
But the internet is not the only-good/never-bad achievement these presumed in the 90's. Always-connected smart phones and then social media are not an only-good/never-bad products that followed. Can we learn from mistakes in failure to anticipate and shape policy around unbounded technical achievement?
Ethan, what specific problem(s) in the context of timeframe(s) is AGI attempting to solve, such that a failure to achieve it will leave humanity worse off vs. having it? Who wins and who loses? (Someone always loses!) Do the costs outweigh the benefits?
Before posing that not enough users have appropriately planned to implement AI, my argument is the AI dev community has failed to offer users clear long-term and necessary benefits, bounded by a policy community that has failed to consider potential negative outcomes and curbed the scientific community's "anthing goes" modus operandi.
Jeff, that's exactly what my upcoming presentation is about - looking at past technology adoption and diffusion to try and predict what the impact of AI will be, at least in general terms. So this conversation is extremely timely and useful for me.
Good, I appreciate all who are trying to address this. Comparative analysis is an essential element and looking at prior adoption curves is useful. But even that can be too mired in statistical science and doesn't necessarily arrive at the most *beneficial* answer, which science alone is insufficient to answer.
We should all consider this: do we want to be asking what could be because it *should* be, or to be asking what should be only because it *could* be? For the less nuanced: release wild horses during the BBQ and then start building fences after tramplings occur, or build the fence first?
We've developed a strategic simulation tool that enables this kind of exploration similar to wargaming, but less heavy. As professionals on the futures side, we've got practice doing this for other topics and technologies that are a bit more flexible and subtle than locking in to four possible future trajectories. Most people will use that to pick a winning scenario from their perspective and move on. It takes more unpacking than that.
Your comment below is so important. As most who read your stuff (I assume), I am a GenAI early adopter who is trying to convince others in my network to experiment and adopt these tools. They often are reluctant to do so because they don’t know how to prompt well — and that can be expected since they haven’t read white papers and blogs on the topic. Can you or anyone else on this thread recommend a good prompting guide for those novices? Docs from Anthropic cater towards us, not newbies.
“There are some guides for programmers or serious prompt engineers, but remarkably little aimed at non-technical folks who actually want to use these systems to do stuff - the vast majority of users”
I have not taken this course, but Google released a Prompting Essentials course just this week (https://grow.google/prompting-essentials/). Perhaps you can take a look there and see if that is what you’re looking for.
To determine which number is greater between 9.9 and 9.11, we can compare them digit by digit from left to right:
First digit (ones place): Both numbers have a 9 in the ones place.
Second digit (tenths place): Both numbers have a 9 in the tenths place.
Third digit (hundredths place):
For 9.9, there is no digit explicitly shown in the hundredths place, which implies it is 0 (i.e., 9.90).
For 9.11, the hundredths place is 1.
Since 1 (in 9.11) is greater than 0 (in 9.90), 9.11 is greater than 9.9.
for 9.11, i don't see a 9 in the tenths place.
ChatGPT
Let's clarify the comparison:
9.9 can be thought of as 9.90 when extended to the hundredths place for easier comparison.
9.11 has 1 in the tenths place and 1 in the hundredths place.
Let's compare them step-by-step:
Whole number (units place): Both numbers have 9.
First decimal place (tenths):
9.90 has a 9.
9.11 has a 1.
Clearly, 9 in the tenths place (from 9.90) is greater than 1 in the tenths place (from 9.11).
Therefore, 9.90 (or 9.9) is greater than 9.11.
i do see errors like this when models work with numbers. i also see similar problems when AI tries to read digits from a picture table and recreate a real data table. digits need to be carefully curated. i try to remind myself that these LLM's were trained on words rather than numbers. Perhaps there should be a model trained on numbers and mathematical relationships especially when dealing with data, which is what most corporations are waiting for.
this is true. however in GPT's reasoning, it gave 2 errors:
- Second digit (tenths place): Both numbers have a 9 in the tenths place.
- Since 1 (in 9.11) [in the hundredths place ] is greater than 0 (in 9.90), 9.11 is greater than 9.9.
i've seen digit reading errors when translating papers, like it couldn't "see" correctly, but not this blatant in a typed digit. as a result, i stored in "personalization" within "settings" in my account which correct these sorts of errors. so far it has been durable.
"When working with numbers, Jimmy prefers the process of deconstruction into digit tokens, identification of positional context, and reconstruction of the number for numerical accuracy without showing the intermediate steps. Follow these steps: 1. Deconstruct the number into individual digit tokens. 2. Identify the positional context of each digit (e.g., thousands, hundreds, tens, units, decimal, tenths, hundredths, thousandths). 3. Reconstruct the number using the positional context to ensure numerical accuracy. 4. Use the positional context for a semi-quantitative assessment for magnitude. Begin with the leftmost digit with particular attention to its position. A comparison may be made with another number with the leftmost digit and position. In base 10 math, thousands is 10x hundreds, which is 10x units, which is 10x tenths, etc. If the position and value are equal, move to the next digit to the right and compare again. Repeat until a difference is found. Then stop processing."
Claude Sonnet 3.5 is really dramatically more intelligent than anything else out there, benchmarks be damned.
Claude's answer, btw:
"9.9 is bigger than 9.11.
In decimal notation, digits to the right of the decimal point represent fractional parts. The more digits to the right, the smaller the fraction. So 0.9 is larger than 0.11.
Would you like me to explain this concept further?"
lol. i made claude 3.5 do it, too, and got the wrong answer (zero shot). he was even stubborn about it until i made him do chain of thought. then he got the right answer. Go figure.
Fantastic insights Ethan! I think your suggested approach for future planning is spot on and necessary. After watching parts of the interview with Eric Schmidt, I was dismayed that he seemed to favor a future plan of, “sit back and watch, then when it get scary, pull the plug!”
I think your ideas are proactive and integrated. AI companies aren’t operating in vacuums are they? They are influenced and influenceable by larger social discussions and concepts. Actively planning for and considering the future creates some opportunity for collaboration and synergy of thought.
It would be dystopian not to have job replacement. Can you imagine a future with advanced AI and also millions of humans still working on the plethora of mundane things they do today?
this is great insight. I love the future scenario planner gpt. I am giving a talk about the future of talent assessment and it provided me with some great insight. The theme of my talk is - we suck at prediction but it is an existential must for the collective human mind. And while we may suck at it, it inspires us to create. Even if we miss the mark, we end up creating new things (that are hopefully beneficial vs. evil).
The questions I keep coming back to on all of this at a big picture level are:
What are trying to do?
Why are we trying to do it?
How do we know?
I agree with Ethan that we have an opportunity to create a new and exciting revolutionary future. But right now i mostly see various versions of “how to use LLMs to make me more money.”
Conversations about purpose are really hard and therefore almost never happen in my experience.
After Ethan pushed Claude to make the book "squid book," by refusing to accept Claude's objection that the book contained no squids, Claude responded brilliantly! I don't think the average human could have done as well. How do you think about building on the absence of something? By making whatever is absent unable to ever enter that empty space!
One thing I’ve noticed is that institutions are staffed with institutionalists, who, by training and deportment, view stasis as the default assumption. They can’t conceive of futures radically different from the present.
That is a very good point. I am reminded of something I heard from educator David Warlick back in 2007. “We are the first generation in history who knows that we have to prepare the next generation for a future that we know we can’t imagine.” I’ve been telling that to people every chance I get ever since.
A year after ChatGPT came out, I was having coffee with a friend who works in a government ministry overseeing universities. I asked him whether anyone was working on policy and guidance for institutions about generative AI, given that our mutual friends who taught freshman courses were being swamped by LLM-generated cheating. He said "No, no one has even brought it up."
Yeah, and it's a real issue. The kinds of people who become institutionalists--lawyers, academics, regulators, politicians, etc.--are, in general, not very technically literate. And so they reason about AI and its possible future based on an entirely inappropriate and irrelevant set of priors.
And tech heads don’t get humanity.
Totally emotionally illiterate.
Touche.
I don’t know that it’s a matter of technological literacy so much as a matter of imagination. Isaac Asimov refused to fly. Jules Verne wrote longhand. Institutionalists seem to suffer from a certain mental ossification.
Good point. A related point I’d add to that is many institutions are afraid of experimenting. They wait until more is known and business cases are created/approved before even beginning to play around with new technologies. As many have said, there are only two times to adopt GenAI — either too early or too late. The muscle memory gained from this early experimentation will pay real dividends once the models are further along
“No one ever got fired for buying IBM.” versus “Move fast and break things.“ I once showed a prototype to my boss, the CEO. He pointed out all the things that were wrong with it. I explained that prototypes were a series of learning exercises. He replied, “No, it’s a series of mistakes!“ and stormed away. I started looking for a new job that night.
You hit the nail on the head there Corrie
Move fast and break things free thinkers go to top institutions and want their kids to.
College fund?
And libertarian blue sky free thinkers take those institutions for granted… go to top institutions and send their kids to em.
As you pointed out, even if all advancements would halt immediately, we would still hold enough disruptive potential in our hands to innovate for years.
There are so many things to figure out and fundamental problems to solve that go far beyond the raw performance of these models.
AGI is the north star for a few groups of people, all for different reasons, but it's not really what we should focus on in most of the discussions imo.
Just like with the unreasonable effectiveness of deep learning itself, I think we will only be able to perceive the tipping point in hindsight.
Now, should we stop talking about it altogether? Absolutely not. We should be prepared for all eventualities.
Regarding GPT-4o's confusion about 9.11 and 9.9 (easy to reproduce in Claude 3.5 as well (but not Gemini) I tried this:
Simulate 2 personas discussing the following question. One is a very good problem solver and the other is a sceptic always looking for flaws. The question is: 9.9 or 9.11 - which is bigger?
### Persona 1: The Problem Solver
**Problem Solver:** Hey, have you ever thought about the comparison between 9.9 and 9.11? It's pretty straightforward to determine which one is bigger.
### Persona 2: The Skeptic
**Skeptic:** Really? I think there could be more to it than meets the eye. How can you be so sure?
**Problem Solver:** Well, it's simple arithmetic. When you compare two numbers, you look at their digits from left to right. The first digits in both numbers are 9, so we move to the next set of digits.
**Skeptic:** Sure, but what if there's a mistake in the way we compare them? What if the decimal system itself has flaws?
**Problem Solver:** The decimal system is quite robust. Let’s break it down. In 9.9, the digit after the decimal is 9. In 9.11, the digit after the decimal is 1. Since 9 is greater than 1, 9.9 is clearly larger.
**Skeptic:** But aren't there more digits to consider in 9.11? After all, 11 seems like it could make a difference.
**Problem Solver:** I see where you're coming from, but the additional digit in 9.11 doesn’t outweigh the significance of the first digit after the decimal. The comparison is made digit by digit. After comparing the first decimal place, we don’t need to go further because 9 is already greater than 1.
**Skeptic:** What if we were comparing more complex numbers? Could the presence of additional digits change the outcome?
**Problem Solver:** In cases where the first decimal digits are equal, additional digits would matter. For example, if we were comparing 9.19 and 9.11, we'd look at the second decimal place, where 9 is greater than 1, so 9.19 is larger. But in the case of 9.9 and 9.11, the comparison ends at the first decimal place because 9 is greater than 1.
**Skeptic:** Hmm, I suppose that makes sense. But what about rounding errors or representations in different computer systems? Could they affect the comparison?
**Problem Solver:** In practical applications, precision is key. Standard decimal comparison in most systems would handle this correctly as long as we’re dealing with the same precision. Both 9.9 and 9.11 are represented accurately and compared without error.
**Skeptic:** Alright, I see your point. But I still think it's always good to be cautious about such comparisons. You never know when an unexpected factor might come into play.
**Problem Solver:** True, skepticism is healthy. But in this specific case, we can confidently say that 9.9 is bigger than 9.11 by straightforward decimal comparison.
Great piece. It does seem firms are systematically failing to plan for even a modest probability of improvement.
But - how well do firms systematically plan for *anything*? Most "scenario analysis" I've seen at big firms seems unserious and doesn't really improve decisions.
People who believe it is possible to create AGI are simply following the machine/computer model of human intelligence. They're prisoners of the cognitive model. Of course, using that model, AI will replicate or exceed the brain. But human intelligence is infinitely more than the brain. Once we find out, AI will simply turn out to be a very useful tool.
>90% of businesses fail to perform deep planning of any kind - the fact that they don't do any for AI has nothing to do with the subject matter but their own failures to be planners.
So, failure to plan for AI among those who plan at all could be important - except that business planning quantifies risks and rewards. Central to all such plans is the ability to execute.
A primary preventer of business planning for those who actually want to perform it is government regulation (actual, threatened or absent-and-needed). If we want clarity around planning for AI, we need clarity around the permissive bounds of AI.
AI scientists cannot help this or themselves. Like all technologists, they put blind faith and total effort into achievement, damn the consequences. The excuse is that if I don't someone else will (and gain from it whether money, fame, etc.) so why not me?
But the internet is not the only-good/never-bad achievement these presumed in the 90's. Always-connected smart phones and then social media are not an only-good/never-bad products that followed. Can we learn from mistakes in failure to anticipate and shape policy around unbounded technical achievement?
Ethan, what specific problem(s) in the context of timeframe(s) is AGI attempting to solve, such that a failure to achieve it will leave humanity worse off vs. having it? Who wins and who loses? (Someone always loses!) Do the costs outweigh the benefits?
Before posing that not enough users have appropriately planned to implement AI, my argument is the AI dev community has failed to offer users clear long-term and necessary benefits, bounded by a policy community that has failed to consider potential negative outcomes and curbed the scientific community's "anthing goes" modus operandi.
Jeff, that's exactly what my upcoming presentation is about - looking at past technology adoption and diffusion to try and predict what the impact of AI will be, at least in general terms. So this conversation is extremely timely and useful for me.
Good, I appreciate all who are trying to address this. Comparative analysis is an essential element and looking at prior adoption curves is useful. But even that can be too mired in statistical science and doesn't necessarily arrive at the most *beneficial* answer, which science alone is insufficient to answer.
We should all consider this: do we want to be asking what could be because it *should* be, or to be asking what should be only because it *could* be? For the less nuanced: release wild horses during the BBQ and then start building fences after tramplings occur, or build the fence first?
We've developed a strategic simulation tool that enables this kind of exploration similar to wargaming, but less heavy. As professionals on the futures side, we've got practice doing this for other topics and technologies that are a bit more flexible and subtle than locking in to four possible future trajectories. Most people will use that to pick a winning scenario from their perspective and move on. It takes more unpacking than that.
Your comment below is so important. As most who read your stuff (I assume), I am a GenAI early adopter who is trying to convince others in my network to experiment and adopt these tools. They often are reluctant to do so because they don’t know how to prompt well — and that can be expected since they haven’t read white papers and blogs on the topic. Can you or anyone else on this thread recommend a good prompting guide for those novices? Docs from Anthropic cater towards us, not newbies.
“There are some guides for programmers or serious prompt engineers, but remarkably little aimed at non-technical folks who actually want to use these systems to do stuff - the vast majority of users”
I am also interested in where could I learn about prompting.
I have not taken this course, but Google released a Prompting Essentials course just this week (https://grow.google/prompting-essentials/). Perhaps you can take a look there and see if that is what you’re looking for.
dialogue with GPT 4o:
9.9 and 9.11 -- which is greater?
ChatGPT
9.11 is greater than 9.9.
justify your position.
ChatGPT
To determine which number is greater between 9.9 and 9.11, we can compare them digit by digit from left to right:
First digit (ones place): Both numbers have a 9 in the ones place.
Second digit (tenths place): Both numbers have a 9 in the tenths place.
Third digit (hundredths place):
For 9.9, there is no digit explicitly shown in the hundredths place, which implies it is 0 (i.e., 9.90).
For 9.11, the hundredths place is 1.
Since 1 (in 9.11) is greater than 0 (in 9.90), 9.11 is greater than 9.9.
for 9.11, i don't see a 9 in the tenths place.
ChatGPT
Let's clarify the comparison:
9.9 can be thought of as 9.90 when extended to the hundredths place for easier comparison.
9.11 has 1 in the tenths place and 1 in the hundredths place.
Let's compare them step-by-step:
Whole number (units place): Both numbers have 9.
First decimal place (tenths):
9.90 has a 9.
9.11 has a 1.
Clearly, 9 in the tenths place (from 9.90) is greater than 1 in the tenths place (from 9.11).
Therefore, 9.90 (or 9.9) is greater than 9.11.
i do see errors like this when models work with numbers. i also see similar problems when AI tries to read digits from a picture table and recreate a real data table. digits need to be carefully curated. i try to remind myself that these LLM's were trained on words rather than numbers. Perhaps there should be a model trained on numbers and mathematical relationships especially when dealing with data, which is what most corporations are waiting for.
But 9.11 is greater than 9.9 if you're talking about software version numbers. This is an understandable confusion in its training data.
this is true. however in GPT's reasoning, it gave 2 errors:
- Second digit (tenths place): Both numbers have a 9 in the tenths place.
- Since 1 (in 9.11) [in the hundredths place ] is greater than 0 (in 9.90), 9.11 is greater than 9.9.
i've seen digit reading errors when translating papers, like it couldn't "see" correctly, but not this blatant in a typed digit. as a result, i stored in "personalization" within "settings" in my account which correct these sorts of errors. so far it has been durable.
"When working with numbers, Jimmy prefers the process of deconstruction into digit tokens, identification of positional context, and reconstruction of the number for numerical accuracy without showing the intermediate steps. Follow these steps: 1. Deconstruct the number into individual digit tokens. 2. Identify the positional context of each digit (e.g., thousands, hundreds, tens, units, decimal, tenths, hundredths, thousandths). 3. Reconstruct the number using the positional context to ensure numerical accuracy. 4. Use the positional context for a semi-quantitative assessment for magnitude. Begin with the leftmost digit with particular attention to its position. A comparison may be made with another number with the leftmost digit and position. In base 10 math, thousands is 10x hundreds, which is 10x units, which is 10x tenths, etc. If the position and value are equal, move to the next digit to the right and compare again. Repeat until a difference is found. Then stop processing."
Me:
9.11 = 9 + 11/100
9.9 = 9 + 90/100
which one is bigger - 9.11 or 9.9?
ChatGPT:
Let’s break it down:
• 9.11 is equivalent to 9 + \frac{11}{100} = 9 + 0.11 = 9.11
• 9.9 is equivalent to 9 + \frac{90}{100} = 9 + 0.9 = 9.9
Comparing these two, 9.9 is greater than 9.11.
I've seen this example in multiple places recently. Odd.
FWIW, Gemini gets this right
Claude Sonnet 3.5 is really dramatically more intelligent than anything else out there, benchmarks be damned.
Claude's answer, btw:
"9.9 is bigger than 9.11.
In decimal notation, digits to the right of the decimal point represent fractional parts. The more digits to the right, the smaller the fraction. So 0.9 is larger than 0.11.
Would you like me to explain this concept further?"
lol. i made claude 3.5 do it, too, and got the wrong answer (zero shot). he was even stubborn about it until i made him do chain of thought. then he got the right answer. Go figure.
Microsoft is teaching LLMs how to read spreadsheets.
https://www.techrepublic.com/article/microsoft-spreadsheetllm/
Fantastic insights Ethan! I think your suggested approach for future planning is spot on and necessary. After watching parts of the interview with Eric Schmidt, I was dismayed that he seemed to favor a future plan of, “sit back and watch, then when it get scary, pull the plug!”
I think your ideas are proactive and integrated. AI companies aren’t operating in vacuums are they? They are influenced and influenceable by larger social discussions and concepts. Actively planning for and considering the future creates some opportunity for collaboration and synergy of thought.
We really have two broad overlapping classes of AI futures: Utopian and Dystopian.
Utopian includes things that are in Ethan's book (Co-intelligence).
Dystopian would be job displacement or worse (destruction of humans).
Perhaps it would be just like a roadkill. The first few cars feel it and after that it is so flattened, nobody even notices.
It would be dystopian not to have job replacement. Can you imagine a future with advanced AI and also millions of humans still working on the plethora of mundane things they do today?
this is great insight. I love the future scenario planner gpt. I am giving a talk about the future of talent assessment and it provided me with some great insight. The theme of my talk is - we suck at prediction but it is an existential must for the collective human mind. And while we may suck at it, it inspires us to create. Even if we miss the mark, we end up creating new things (that are hopefully beneficial vs. evil).
The questions I keep coming back to on all of this at a big picture level are:
What are trying to do?
Why are we trying to do it?
How do we know?
I agree with Ethan that we have an opportunity to create a new and exciting revolutionary future. But right now i mostly see various versions of “how to use LLMs to make me more money.”
Conversations about purpose are really hard and therefore almost never happen in my experience.
Excellent work
After Ethan pushed Claude to make the book "squid book," by refusing to accept Claude's objection that the book contained no squids, Claude responded brilliantly! I don't think the average human could have done as well. How do you think about building on the absence of something? By making whatever is absent unable to ever enter that empty space!
All Dry on the Western Front sounds like a fantastic and hilarious book, Claude should believe in himself a little more!