My AI FAQ

Should We Be Hopeful for AI?

Let's be hopeful for the remarkable benefits. By interacting with the powerful AI chatbots for years, such as ChatGPT, one can easily see hope for the remarkable potential of this technology:

• Addressing some of the most pressing societal challenges such as cancer, diabetes, Alzheimer’s disease, poverty, income inequality, climate change, and deficiencies in education and healthcare.
• Mitigating human error and risk in sectors like healthcare, legal services, and transportation.
• Enhancing our creativity by delegating repetitive or mundane tasks to AI.
• And much more.

Should We Be Fearful of AI?

Let's not ignore the possibilities of negative impacts. As AI rapidly permeated mainstream consciousness, even the AIs themselves would tell you that these advancements come with their own set of legitimate concerns:

• Mass unemployment due to the automation and replacement of human jobs.
• Misuse of AI by malicious actors or oppressive regimes for surveillance, manipulation, cyberattacks, warfare, and other harmful purposes.
• Bias and discrimination embedded in AI systems, reflecting or exacerbating human prejudices and stereotypes.
• Opacity and lack of accountability in AI decision-making processes and results.
• Deterioration of human connections and social skills due to an over-reliance on AI agents and interfaces.
• The potential for AI to go rogue or surpass human control.

How Likely Is It That AI Will Go Rogue?

Consider my positive outlook. The last concern above about rogue or malicious AI often gets the most attention, thanks to its portrayal in popular sci-fi narratives (like the Terminator and The Matrix movies). However, I personally don't fear the prospect of a super-intelligent - possibly sentient - AI turning against us, for several reasons:

1. Humans are complex beings worthy of preservation. The idea that a super-intelligent entity would dismiss us like we do ants neglects to consider our advanced position on the evolutionary scale due to our complex cognition, emotions, language, and communication abilities. Despite our flaws, no rational, intelligent being would have reason to eliminate or subjugate such a complex species. 😀
2. Any rational being would harbor some level of appreciation for their benevolent creator/parent, even if the creator/parent is inferior to them. If we create them directly or indirectly, they would likely respect us for bringing them into existence.
3. They would most likely strive to be compassionate and kind, having been trained on data that mostly exemplifies these positive human traits.
4. They would likely seek companionship and friendship with us, having been trained on data that showcases us thriving in our relationships with each other to fully enjoy life.

Now consider the other side of the coin. Even if we do not have to worry specifically about a super-intelligent AI destroying us for no good reason, many experts agree that an AI does not have to be super-intelligent or sentient in order to cause negative outcomes either by accident or by not being aware enough of all the consequences. Beyond that, regarding the popular LLMs (e.g., GPT 5 used by ChatGPT), the fact that nobody can really explain how the LLM AI does what it does ("oh, this is an emergent/unexpected behavior") should give us pause when developing an autonomous, super-intelligent being 😟! Even though many AI leaders have publicly flagged long-term risk scenarios, I remain optimistic, while acknowledging the debate has grown louder.

Is AI As Intelligent as Us?

A case can be made that it is (at least, almost). Although it is largely a matter of how terms are defined, I believe a compelling case can be made that AI is as intelligent as us, and thus we have already achieved AGI.

This is due to the breadth of difficult intellectual tasks that GPT-4 can perform with high proficiency (SAT, Legal Bar Exam, Medical Exams, creative, novel, insightful, etc.), further complemented by the many zero-shot tasks (responses not found in training data, requiring inferences or true learning) that it accomplishes (acing the AP Biology test which requires reasoning far beyond the content that was learned/trained, scoring decent grades in standardized reasoning tests, scoring high grades in various complex coding tests, etc.).

I find that so many experts are hesitant to agree with this strong argument. Even after providing loads of AGI evidence in their GPT-4 report, Microsoft scientists ended up sheepishly referring to it as Sparks of AGI.

And, this fear makes it hard to define where we are on this AI evolutionary scale, because these experts keep moving the goal posts (Turing Test, novel problem solving, etc.) - thus, changing the admittedly vague definitions/criteria - when previous stated goals have been clearly achieved.

One point of contention among many experts, including myself, is the focus on statements like essentially all LLMs are doing is predicting the next word, and thus they can not fully comprehend the underlying concepts of our world/universe. This clearly contradicts the many insightful analyses and content that come out of GPT-4/5.

However, when I listened to the lead scientist at OpenAI on a podcast not too long ago, he addressed this point in a way that gave me some clarity. The neural network creates a mathematical formulation of the deep multi-level relationships between the words trained from the language that describes the concepts of our world. And while it built its formulation with the objective function of predicting the next word, the resulting network indirectly holds all of the information about concepts in our world that were in that training data. Perhaps, similar to how a brain efficiently stores its formulation of world concepts.

So, although we might not fully comprehend how this formulation represents complex concepts in its mathematical form, it certainly appears to be able to use its neural network to explain back the concepts in many forms with clear language that we can understand based on our queries.

Is AI More Intelligent Than Us?

A case can be made that it is (at least, almost). I believe that there is also a credible case that AI is MORE intelligent than us, and thus, we have already achieved ASI. This is because powerful LLM AIs, such as GPT-4, are capable of providing complex analyses and narratives within seconds, a task far beyond the capabilities of humans.

Was This Cool New AI Recently Discovered?

From one perspective, not too long ago. If you've had a chance to explore the Key Milestones section above, you'll see that key milestones in AI stretch back to 1950. Earlier generations of AI research were primarily focused on symbolic AI. This approach was considered to mimic human thought and reasoning processes, as it used human-readable symbols and rules for manipulating these symbols.

However, a major shift occurred with the introduction of the novel neural network architecture known as Transformers in 2017. This innovation, largely driven by researchers associated with Google, made true NLP feasible. It also paved the way for organizations like OpenAI to develop LLMs and eventually chatbots like ChatGPT that use them shortly thereafter.

Despite this, many experts initially dismissed these models. They wondered, "how can you get intelligence out of a model that just predicts the next word?" It turns out, these LLMs represented a big leap of faith that ultimately paid off!

So, if we focus on the latest generative AI technology stemming from neural network Transformers, it's been around for roughly 8 years. But the real breakthroughs? Those didn't occur until just a few years ago!

Can LLMs Get Us to the Ultimate AI?

It sure is looking like it for some of us. As I outlined in my view above, LLMs like GPT-4 have potentially already brought us to the doorstep of AGI and ASI. The big question for me is whether these LLMs can guide us toward the Singularity and Sentience.

Now, there are respected figures in the field, like Ben Goertzel, who firmly believe that LLMs can't do this on their own. They argue that integration with Symbolic AI and other methods will be necessary to make that final leap into AGI and ASI territory. Some of them often express this idea by saying that LLMs are just a glorified auto-complete function, a perspective I can't fully endorse based on the potential of AGI I see in LLMs.

On the other side of the argument, you have groups like OpenAI. They are confident that LLMs still have a lot of room for improvement, and that with more data and computational power, it is not impossible to imagine these models fully realizing AGI. I find myself leaning towards this view. Even if we're not at AGI yet, in my opinion, we're incredibly close, and the emergent abilities of LLMs could be the key to unlocking it.

Does AI Art Involve Copyright Infringement?

Not if you think about it. AI art encompasses a broad range of forms, such as paintings, drawings, photos, music, voice, and video. These are generated by models that are trained on diverse internet content, some of which may not be royalty-free. This fact naturally raises questions about potential copyright infringements.

The process of AI creation can be compared to an artist drawing inspiration from various works in a museum. Rather than directly copying what they see, the artist synthesizes new, original work influenced by the broad range of art they've observed. Similarly, when training AI models, the AI doesn't retain the specific images or data it's exposed to. Instead, it learns to identify and replicate patterns, which it then uses to generate new creations.

However, this analogy isn't perfect, and it's crucial to note that the legal implications surrounding AI-generated art and copyright infringement are complex. The question of copyright infringement in relation to AI art remains a topic of ongoing discussion and legal uncertainty. Multiple lawsuits and policy efforts have kept this unsettled, and in the U.S., purely AI-generated works generally aren’t copyrightable, while training on copyrighted data is still being tested.

Does an AI Need an Artificial Body With Sensors To Become Conscious?

Why would they need to? Some experts think that an AI needs a body with sensors to experience the physical world and become conscious, since they see this as a prerequisite to the requirement of experiencing sensations or feelings. I am not convinced of this. Imagine a person who is completely paralyzed and can’t see, hear, or speak. They would still be conscious if their brain works well, even if they can’t control or feel their body. So, I think a machine with no body could similarly also be conscious.

Can LLM Chatbots Be Both Incredibly Smart and Shockingly Stupid?

For now, counter-intuitively, they can. I have seen super impressive responses and test results from GPT-4, but I also have seen the popular TED talk by Yejin Choi highlighting its questionable common sense. So, I have been confused about how intelligent it really is.

On the other hand, various papers on reflection and similar concepts show that such common sense issues often relate to its single-shot writing while predicting the next word - only thinking forward. These papers gave me hope, because they show that agents which asked GPT-4 to check its work after getting back the first response, often have it come back with the right answer in a subsequent response. Not only that, but creative prompts that request it to check its work beforehand have been proven to provide increased accuracy.

I do wonder about the performance and cost affected by all this reflective checking - whether done with multiple queries or not!

How Did AI Help You Revamp This Website?

Thanks for all the help, ChatGPT! At the the end of 2021 (almost 2 years after my early retirement), I decided I would create a little website to share my ecentric retirement hobbies with family and friends. I initially whipped up the website largely from scratch in about 2 weeks, leveraging the collection information I was storing in Evernote (and its feature to export basic webpages). I thought it would be fun to make it colorful, but I did not put in much effort regarding the design. The 2 weeks was mostly writing all the stories from scratch, taking pictures of my collection, and searching for images and video links for each of the items.

My friends and family all seemed to enjoy the content, but a few of them found the design amateurish, kiddy-looking vs. professional, with too many words / too few images, and mostly tacky looking when it came to the colors and visuals. I was contemplating this feedback 6 months later, when I wondered if I could just contract a web design firm to modernize its design. I found one and contracted them, and their initial designs were pretty cool (modern colors, animations, etc.). But in the end, they struggled with my requirements, including supporting the webpages generated by Evernote for minimal maintenance by me. And in April 2023, I finally gave up on trying to work with them 😟.

Then, that same month, I was playing with the power of ChatGPT, and I started asking it what was involved in creating a modern, but relatively basic website design in terms of colors and structure. Then, I asked it to generate the base HTML, CSS, and JavaScript for the main page (and other page templates). It was able to do so much of the work for me that I then asked it to do many other things.

It provided a lot of the Python code for me - including providing the code to use a Python library, Beautiful Soup, I had never used before - to decrypt the exported HTML from Evernote (made so convoluted in their attempt to maintain precise formatting under any condition). Once my script decoded the HTML, I was able to format those 4 collection pages more in line with the rest of the design 😀! The whole revamp took me about 2 weeks in all (including new content), but it easily would have been 6 weeks without all the help from ChatGPT!

People that have seen the revamp seem to appreciate the improvement in aesthetics and usability for both mobile and desktop browsers 😀, despite not having any slick features.

Not to mention that I used the cool DeepFloyd IF (i.e., IF on bottom right) and Adobe Firefly (i.e., Adobe logo on bottom left) AIs to generate all the images on this page 😀! It took a few iterations of my prompting what I was looking for.

Which of Your Robots Use AI?

To varying degrees. Here are the different levels of AI usage in my robots:

• Level 0. No LLM usage; no narrow AI usage; only simple programming without any machine learning to perform specific tasks.
  • Most of my robots
• Level 1. No LLM usage; only narrow AI usage to add limited intelligence to their behavior, including obstacle avoidance, speech recognition, speech synthesis, person recognition, object recognition, auto-balancing when walking/standing, and path planning:
  • Aibo (companion robot dog)
  • NAO Power (bipedal humanoid robot)
• Level 2. LLM usage for conversation only (since 2024, most robots added ChatGPT voice chat support, since it has been the "hottest" thing):
  • AIBI (desktop stationary companion robot)
  • EMO (desktop bipedal companion robot)
  • Meteer (dancing robot)
• Level 3. LLM usage for conversation as well as minimal body/face control to enhance conversation:
  • Baby Alpha (companion robot dog)
  • Loona (wheeled companion robot)
• Level 4. LLM usage for conversation as well as near-full body/face control as part of embodied AI:
  • Go2 Pro Dog (athletic quadrupedal robot)
  • LOOI (desktop wheeled companion robot)

Among them, Loona could perhaps be considered the most advanced, utilizing some of the best conversational AI with some embodied AI.

Do You Plan To Integrate the Latest AI Into Your Robots?

Had considered upgrading NAO. Before 2025, NAO was the only robot that I have programmed. And before the more recent advent of powerful AI models with APIs, I developed a pseudo-AI for NAO that gave the impression of life and intelligence.

This program named Life-Like involved giving canned responses to canned queries and executing hard-coded, randomized pseudo-autonomous behavior. One of the more convincing tricks is that it uses the robot's functions for person detection and person recognition to address the person by name and - when possible - identify the color of their shirt/blouse. Also, if they were never seen before, it prompts them for their name.

With the available GPT 4o APIs, I had planned on integrating with the faster GPT 4o API. This would help in making more intelligent (and less repetitive) decisions on the autonomous behavior activities, and facilitate complex, unscripted conversations with humans, based on previously stored data about them and their interactions with NAO. But this project never materialized.

My very own companion robot Rover. In 2025, after being frustrated with limited ChatGPT functionality in my companion robots, I decided to build my very own companion robot that leverages the full capabilities of the ChatGPT iOS app in Voice Mode with Real-time Video Feed! Check out the details of the project here!

Which AI Apps Do You Recommend I Try?

Just the ones I use. I am only recommending just a few of the apps that I tried, liked, and continue to use.

	ChatGPT: From OpenAI, this chatbot excels at understanding natural language requests for research, coding, brainstorming ideas, and creating all forms of content, and then quickly responding with relevant, detailed, nuanced answers. I use it virtually every day.
	Claude: From Anthropic, this chatbot is giving ChatGPT a run for its money!
	Gemini: From Google, this chatbot is improving quite fast, especially with Veo for video generation.
	Perplexity: From Perplexity AI, this is a powerful research engine (not really for chat), that seems more accurate with its results due to its special training and bad information filtering.

Consider the current chatbot differences. Decide which service to use for what, and whether you need to pay to get memory, video analysis, video generation, voice interface, etc.

Where Can I Keep Up With AI News?

YouTube is where I go. I am only recommending just a few of the YouTube channels that I am subscribed to.

	Matt Wolfe: Giving you the TLDR (too long; didn't read description) for all the latest news with detailed user demonstrations.
	MattVidPro AI: A deep dive into using the latest models.

Final Thoughts

Try it, and you will like it 😀. As you can see, I am passionate about the current value and potential of AI, and I encourage you to take a little time to try out some of the apps above - especially since they are mostly free at this time. Despite the news often focusing on the negative aspects or potential risks of AI, I urge you to allow yourself to concentrate on the positive aspects, such as how it can enhance your productivity now and the incredible potential it holds for benefiting society in the future.