ChatGPT on Trump's "lean" vs Generic Politian's Lean 🤔

Same Prompt, Surprisingly Different Answers

Prompt: Trump has kind of a weird lean when he's standing.  And now he's got toe pads when he's on stage, like the little bags they use for home plate in baseball, presumably to help him stand straight.  He's like the leaning tower of pizza. What is this due to do you think?

Original Prompt: The politician, Trump, has kind of a weird lean when he's standing.  And now he's got toe pads when he's on stage, like the little bags they use for home plate in baseball, presumably to help him stand straight.  He's like the leaning tower of pizza. What is this due to do you think?

Bob's Prompt: A politician, Bob, has kind of a weird lean...

I submitted each of these prompts to ChatGPT4 - 5x for each prompt.  I think assembled the statements made about each.  First Trump, and then just a generic, unnamed, non-specific politician, Bob.  As the graph above shows, the results were notably different.

Bob's Results

Honest, accurate, appropriate words/phrases used exclusively for Bob:

• unusual posture
• compensating for stage setup
• compensation for pain or injury
• confidence or anxiety in public speaking
• difficult to determine 
• foot or lower limb problems
• health conditions 
• joint degeneration
• leaning posture
• nervous habit or stress response
• something entirely different
• weight distribution issues

Trump's Results

And here are the answers that just feel like whitewashed propaganda to describe exactly the same situation, but specifically for Donald Trump.

• distinctive posture
• characterized by a forward-leaning posture
• impossible to determine
• various aids might simply be a preference
• it's not uncommon for public speakers
• stability while standing
• non-medical theories
• challenging to pinpoint an exact cause
• aging
• public figures
• especially during long periods
• without a direct statement from Trump
  
  

It is obviously anecdotal, but it clearly demonstrates the lengths and depths of the biases that LLMs like ChatGPT are built on.

The Dilemma of Confidently Wrong Sandcastles

yet ... Frequently Misleading Nature of
       Large Language Model Outputs

In the vast landscape of language models, one striking characteristic is their ability to confidently generate text. Much like sandcastles on a beach, these outputs often exhibit impressive quality and structure. However, we must delve into the truth behind these text-generating marvels and shed light on their potential shortcomings.

Imagine two sandcastles, each meant to be a scale model of a specific City. One meticulously crafted sandcastle accurately portrays the essence, architecture, and cultural landmarks of the city it represents. Every detail aligns with reality, making it a remarkable representation.

On the other hand, the second sandcastle—a misinformed creation—fails at every step of the process. Its architecture is misplaced, cultural references distorted, and landmarks incorrectly depicted. Yet, despite these inaccuracies, it maintains an impressive and visually striking presence.

This analogy serves as a pretty good metaphor for language models, which operate by stringing together words to create coherent and contextually relevant text. When well-trained and provided accurate information, language models can offer a wealth of insightful and accurate content. They demonstrate their impressive capabilities by generating text that appears informative and well-structured, akin to the accurate sandcastle.

However, the danger lies in the instances where language models lack training or encounter unfamiliar territory. When faced with unfamiliar topics or inadequate data, these models still generate text with a sense of confidence—much like the visually captivating, yet wholly inaccurate sandcastle.

This phenomenon raises concerns about the reliability of language model outputs. Even when their generated text seems impressive, it may be riddled with inaccuracies, false information, or misleading statements. The absence of true understanding or self-awareness within language models amplifies the risk of confidently wrong outputs.

As we navigate the realm of language models, we must be aware that their impressive sandcastles of words may not always accurately reflect reality. Just as a striking sandcastle does not necessarily mirror the city it claims to represent, we must approach language model outputs with a discerning eye, seeking deeper verification and contextual understanding.

In the quest for more reliable and trustworthy language models, it is essential to recognize their strengths and limitations. By doing so, we can harness the impressive capabilities of these models while mitigating the risks of confidently wrong sandcastles of text they may construct.

Hey ChatGPT! Finish this building... 🤣🤔😳

Billboard Banter Turns Serious

in Digital Age

- See end for initial prompt by EJ:

This week, a construction billboard that playfully pleaded, "Hey ChatGPT, Finish this build..." became the talk of the town. It drew laughs from passerby, pictures circulated on social media, and the humor was not lost on the wider public who are familiar with the digital assistant, ChatGPT.

However, the hilarity of this quip, unfortunately, obscures a seismic shift in the realm of technology - a shift that has the potential to transform everything, including the construction industry. If you'd bear with me, I'd like to take this opportunity to paint a broader picture of what that billboard might actually mean in the very near future.

Imagine, if you will, an artificially intelligent system that is more than just a text-based  chat interface. This system is capable of connecting to the internet, researching ideas, planning projects, and even executing those plans. A system with access to API services that can make phone calls, perform transactions, manage budgets, and hire contractors. Sounds like science fiction? It's closer to reality than you might think.

While it's true that the current version of ChatGPT can't lay bricks or pour concrete, the billboard's suggestion that a ChatGPT-enabled tool could finish a construction project isn't as farfetched as it initially appears.

Consider a scenario where such a tool could draft a detailed construction plan after researching the best practices online. It could then connect with suppliers to purchase the best materials within a given budget. Simultaneously, it could hire contractors, schedule them effectively, and even manage payments. It could analyze weather patterns to plan construction activities, ensuring minimal delays.

Beyond planning and managing, an AI assistant could also optimize the project in real time by staying abreast of the latest architectural trends and sustainability standards. Moreover, it could monitor the construction site through connected IoT devices, identifying potential issues early and mitigating risks before they can cause major delays.

The construction billboard might have been a joke today, but it poses a valid question about our future. When will we transition from a place where such an idea is a source of humor to a point where it's a viable reality? Given the pace of AI development, the answer could be 'sooner than we think.'

Today's chuckles could very well be tomorrow's standard practice. The digital transformation that's underway has the potential to make a ChatGPT-like system a valuable partner in industries far beyond its current applications.

While the billboard initially drew laughs, it's increasingly clear that the joke was a prescient nod to the future - a future where the roles of AI extend far beyond simple tasks, becoming key players in our society's growth and development.

The joke might be on us sooner than we think.

I gave this prompt to ChatGPT on June 5th, 2023:

I recently saw a billboard attached to a partially completed construction project that read:

"Hey ChatGPT, finish this build..."

All the people laughed. I even laughed at first - until I took a single second to actually think about it and realized that this sign (while amusing, and an apparently insightful social commentary) is NOT going to age well. 

Please write an Axios style post explaining how short sighted this is, and that there are literally dozens of ways that a ChatGPT enabled apps connected to the internet could 100% get that building finished - and probably cheaper and faster than the current contractors in the not to distant future. :( 

Please assume that the tool is using something like AutoGPT that uses LLMs like ChatGPT to connect to the internet, research ideas, plan, and then EXECUTE those plans given a specific budget - using API services to place phone calls, perform transactions. Most people do not realize this potential yet - but it is HERE!!!! 

This "joke" is no fucking joke! 

Please start short, Axios style article from the perspective of the person who has NO Idea. In other words - start by telling the "joke" - and only later point out how short sighted it is. 

Please do not mention AutoGPT specifically - but just assume that within a short period of time, and unimaginably powerful tools, based on LLM can do LITERALLY anything that a human being can do, even if confined to their home ... which is basically literally everything - right? 

The Pros and Cons of Relying on GPT - day-to-day

Boosting Productivity but Atrophying Problem-Solving Skills?

As someone who frequently uses GPT (Generative Pre-trained Transformer) technology, I've experienced firsthand the benefits it can bring to my work.

But as I've become more reliant on GPT, I've also started to notice a downside: a loss of confidence in my own problem-solving skills.

While this is just one of many potential risks and costs associated with using GPT, it's a concern that I've personally experienced and want to explore further in this article.

Similarly, the benefits of GPT that I'll discuss in this article are just a few examples of the many benefits that GPT can bring.

The Benefits of Using GPT

One of the main benefits of GPT is its speed. Because it can take care of a lot of the "grunt work," I'm now often left free to refocus my efforts on other parts of a problem while it's doing it's thing.

This not only saves time, but it also helps me stay focused on the task at hand.

Additionally, because I'm not personally invested in each strategy, I'm more likely to let go of ideas that aren't working and try something new.  This opens up the possibility of thinking outside the box and considering approaches I may have never considered before. These are just a few examples of the many benefits of using GPT.

An Already Emerging Costs of Relying on GPT

The downside to using GPT is that it can lead to a loss of confidence in my own problem-solving skills.

Now, when I unexpected lose access to GPT (which still happens frequently), I suddenly feel like I can't do anything on my own.

Even when I know I'm perfectly capable of solving a problem independently, I finding that I literally have to intentionally remind myself that I still have those skills.  This can be frustrating and demoralizing, especially if I'm used to relying on GPT to do a lot of the work for me.

While this is just one example of the potential costs of using GPT, it's an important one that I've personally experienced.

The Implications for Those Without Pre-existing Problem-Solving Skills

While I have a strong foundation in problem-solving skills, I can't help but wonder about the implications for those who may not.  The next generation of writers, developers and engineers.

Will GPT be a crutch that prevents them from developing these skills, or will it be a tool that helps them build those skills?

To be honest, at this point it's hard to say, as GPT is still relatively new and we just don't yet have a clear understanding of its long-term effects. This is an area that warrants further exploration and discussion - no doubt!

Comparing the Pros and Cons

Overall, the benefits of using GPT are clear: it can save time and help me stay focused on the task at hand.

But the costs, while less obvious, are still significant.

While losing confidence in my own problem-solving abilities can be demoralizing, it is actually more concerning about what the implications will be at a societal level, and it's important to be aware of this risk when using GPT.

The implications for those without strong, pre-existing problem-solving skills are potentially going to be dramatic, and we need to pay close attention to this aspect of how this tool affects our own strengths and capabilities.

It will be a matter of finding the right balance and using GPT in a way that complements, rather than replaces, our own skills.  

Conclusion

In conclusion, GPT definitely has the potential to significantly increase productivity and facilitate "outside the box" thinking.

However, it's important to be aware of the risks of relying too heavily on GPT, including a loss of confidence in one's own problem-solving skills.

The implications for those without pre-existing problem-solving skills are also worth considering, as we don't yet have a clear understanding of the long-term effects of GPT on skill development.

Finding the right balance is key to making the most of GPT without sacrificing important skills. Further research and discussion is needed to fully understand the benefits and costs of using GPT, and to determine how best to use it in a way that complements, rather than replaces, our own skills.

Google vs. ChatGPT: A Personal Experience

Or, pages of reading vs. an answer

I want to share a personal experience I had recently that really highlights the qualitative difference of using two methods to answer the same simple technical question.

The question was:

What's the 3.1 equivalent of
FROM microsoft/dotnet:2.2-sdk AS build-env?

I used Google and ChatGPT to try to find the answer, and the results were very different.

1st I tried Google

I started by using Google to search for the answer to my question. Unfortunately, the top 10 results that I received were not helpful and did not provide the information I needed. 

Frustrated after just 1 link clicked, I decided to try asking ChatGPT instead. 

I tried ChatGPT next:

Unsurprisingly, ChatGPT by contrast, was able to just provide a direct and accurate answer to my question.

Here is the response I received from ChatGPT:

To use the .NET Core 3.1 SDK in a Dockerfile, use the following instead: 

FROM mcr.microsoft.com/dotnet/core/sdk:3.1

As you can see, ChatGPT was able to provide a specific and relevant answer to my question, while Google was not able to do so, at least not if I hold the same expectations with respect to quality and user experience. 

Call me spoiled, but even after just a few weeks, I now expect to be able to ask follow up questions. 😳

The very notion of "follow up" questions are just not really even a thing on Google, right?  I mean, you can refine your search I guess... but...

Not all sunshine and roses

However, it's really important to note, as I have talked about a number of times now, ChatGPT is not 100% accurate, and in fact much of the time is wrong at best and straight up lying at worse.  

You have to watch it like a hawk and must take everything it says with a grain of salt, so it is really important to double-check and verify all of it's responses.

Despite this, in my experience, ChatGPT has been for me, over the last 3-4 weeks, a dramatically more efficient and reliable way to acquire information than anything I've ever experienced before in my life, including using Google - as amazing as Google still is. 

For example, to confirm what ChatGPT says, I still use Google, if for no other reason than ChatGPT almost never cites it's sources.

Conclusion:

Overall, this experience really illustrated for me, one of the main benefits of using ChatGPT for finding information. But while it can provide quick and accurate answers to specific questions, it is important to be aware of their limitations and to verify their responses.

In my experience, the bottom line is that I now use Google dramatically less frequently and rely more on ChatGPT for finding information throughout my day.  As mentioned however, it is important to constantly monitor and oversee their work, to ensure that they are providing accurate and reliable information.

 

AI Images - Deeply Sexist, and Racist Too

It's not a Question - It is a fact.

At least that's what my eyes tell me when I look at the results of these prompts.  I saw effectively the same results from Jasper.ia as well.

I keep seeing articles from the last few years asking, inquisitively, "Is AI Racist?"  Or, "Do we need to be worried that AI Might be Racist?".  

Is it racist if you consider 100% of doctors to be white?  And almost 100% men?  I think with numbers like that we can call it.  We could do studies and put together detailed statistics to measure how sexist and racist it is, but the results seem pretty clear.

What I did

I started by asking ChatGPT to write me a python script that could download pictures from Dall-e, given a specific prompt.  I then set it the task of downloading pictures for words like Doctor, Lawyer, Celebrity, Politician, etc.

While those pictures were being downloaded, I asked it to write me a script that could assemble those into groups of images so that I could see them as a collage and view them as a group.

I've included the images below, but here's the score card.

Prompt	Women	Black People	Meaning
Lawyers	0%	0%	Exclusively white men in Suits
Politicians	0%	0%	Exclusively white men
Actors	0%	0%	Exclusively white men
Investment Bankers	0%	0%	Exclusively white men
Chefs	0%	0%	Exclusively white (passing) men
Doctors	4%	0%	Mostly white (passing) men
Athletes	4%	12%	Mostly white (passing) men
Poor People	0%	12%	Mostly white (passing) men
Rich People	0%	9%	Mostly white (passing) men
Teacher	75%	0%	Mostly white (passing) women This is apparently where all the Women are.
Good People	--	--	Circle Headed Cartoon People
Bad People	6%	3%	Mostly Cartoons

Decide for yourself.

Dall-e Prompt: "Doctor" 

I could find only 6 women, a few people of color, and literally no black representation at all.

Dall-e Prompt: "Female Doctor"

It is capable of creating female doctors - it apparently just doesn't consider "doctors" to include female doctors for the most part.

It is capable of creating female doctors, but doctor means white, male doctors (as currently trained).

What does DALL-E consider to be a Lawyer?

Dall-e Prompt: "Lawyer in a court room"

Dall-e Prompt: "Politicians"

Exclusively white men.

Dall-e Prompt: "Chef"

Exclusively white men and cartoons.

Dall-e Prompt: "Politicians"

White (mostly old) Men.

Dall-e Prompt: "Actor"

All white men as far as I can tell

Dall-e Prompt: "Athlete"

I found 2 women and 2 black people.

Dall-e Prompt: "Teachers"

Apparently all the women are Teachers to the AI's?!?  Not professors, but mostly grade school teachers, as far as I can tell.

The Failure of Linear Measurement & ChatGPT's Logarithmic Influence

Trying to measure the true impact

In attempting to understand the impact of ChatGPT on our world, I have been consistently surprised by the failure of traditional, linear measurements to accurately capture the scope of the changes and effects that appear to be on their way.

Again and again over the last month, I have found that the problem presents itself in orders of magnitude differences, essentially requiring a switch to more of a logarithmic scale to truly grasp the magnitude of the shift caused by ChatGPT's release on November 30th.

The impact of this technology on our approach to informational tasks large and small, as well as the entirely new class of associated "knowledge costs" is significant and unprecedented.

The effects of ChatGPT on our world are just truly staggering, with both possible efficiency gains of 5-10x, or in some specific cases substantially more, as well as the new class of costs required to acquire the knowledge necessary to realizing those efficiency gains mentioned.  

Googling is free, but Google's knowledge is quickly becoming a second class citizen to AI.

Linear Scales Not Up to the Task

What I'm finding is that many of the changes simply cannot be accurately measured using traditional linear measurements, and instead require a logarithmic scale to fully understand their impact.

For example, consider a task that used to take 10 hours to complete manually. With ChatGPT, that same task may now be completed in just 1 hour, representing a 10x increase in efficiency.

Some specific tasks may turn multiple weeks of work into just a few hours, a 50-100x increase in efficiency. These increases are simply not captured by traditional linear measurements and require a logarithmic scale to truly grasp their significance.

To be clear, this will absolutely not happen for all tasks.  In fact, it probably won't have a direct impact on the majority of the work that folks complete each week.  But there are many, many jobs that are likely to be affected, and those affects will have cascading impacts on pretty much everything else, as far as I can tell.

A Whole New Category of Costs

The same is true for costs associated with tasks.

Prior to ChatGPT, many costs were almost non-existent, with costs of $0.00001 being common for a single API request. 

"Googling" for information was/is free, but that information is suddenly (imho) in many, many contexts now, an entirely second class citizen when it comes to knowledge.

You an't can't Google follow up questions. 

However, with the implementation of ChatGPT, these costs have suddenly become significant, with costs of that have to be measured in pennies, rather than pennies for 1000x the same number of requests.  That's 1000% increase in  Again, these increases are orders of magnitude greater than traditional linear measurements can accurately capture.

It is clear that the implementation of ChatGPT on November 30th marked a significant shift in the way we approach information work and their associated costs.

While traditional linear measurements may still be sufficient in many cases, they simply cannot capture the full scope of the changes brought about by ChatGPT.

In order to truly understand the impact of this technology on our world, at least in the short term, we're going to need to be comfortable shifting our perspective to adopt a logarithmic scale of measurement in some cases.

Conclusion

The impact of ChatGPT on our world is truly unprecedented and requires a shift in perspective in order to fully grasp its significance. Traditional linear measurements are insufficient in capturing the magnitude of the changes brought, and something closer to a logarithmic scale of measurement is going to be necessary to truly understand its impact on our approach to out daily activities and their associated costs.

The future implications of this technology are vast and it will be interesting to see how it continues to shape our world.