We've all been inundated recently with the sensational news surrounding the new generative software tool, Chat Generative Pre-trained Transformer, or ChatGPT, a chatbot developed by privately held OpenAI. On the positive side, such a tool has the potential of a query aid for students in a range of technical topics (I give a few examples based on my own interaction with the tool below). On the negative side, ChatGPT has produced B-grade essays and marketing material that serve as plagiaristic content for the user to pass on, or worse, "plausible sounding but incorrect or nonsensical content" as Wiki aptly states. Similar generative tools, such as DALL-E (so-named after the famed surreal artist Salvador Dali), also from OpenAI, generate digital images based on databases of images and artistic styles, again subject to plagiaristic outcomes and copyright violations.   

My own take on such tools is that they hardly represent the AI that might be productively useful, such as expert AI training systems that interactively teach and train humans on specialized topics, such as how to weld a joint and check it structurally, or how to give CPR and other EMT tests in an emergency, or even advanced techniques, such as how to prepare a biological assay or DNA sequencing test, conduct routine surgical procedures, or write specialized computer code and algorithms (and no, not produce the algo, but to teach a human how to produce it, with levels of understanding and proficiency checked). What humans lack is basic knowledge and proficiency, and AI training systems can fill in the deficiencies with productivity and patience. (I wrote about the need for this concept in the midst of the Covid pandemic in April 2020, "It's Time to Train.") Imagine an interactive AI that is focused on teaching mathematics or physics to young students, with the patience and breadth of techniques that serve even the least motivated or most deficient students, or even disabled students. That sounds like a productive progress for AI, even though it may be years away, given the progress to date. It requires an AI with understanding of subject matter and the ability to interact with and understand humans, far beyond anything we've seen.

One may argue that ChatGPT is not AI, but rather a collection of machine learning algos and a large language database, producing human-like responses to queries. It lacks understanding and context. This is abundantly clear with the latest salacious responses it has offered users bent on forcing nonsensical content out of the tool. That ChatGPT has been integrated into a major search engine with the plan for wide distribution is at the least disappointing, and at the most terrifying. What is the solution? Don’t offer such ----. Focus on developing AI that is specialized, expert, productive and can actually interact with and understand humans, ostensibly for teaching and training, a high need across many cultures and classes.

One solution for generative tools like ChatGPT or DALL-E is to embed digital signatures or watermarks in the text or images, such as cryptographically. This idea isn't original, apparently OpenAI may have already toyed with the idea (see HERE), but has not yet implemented it. (If it had, its recently released text checking app to educators would probably not be so poor at detecting ChatGPT-produced content!) DALL-E images are apparently digitally watermarked, but this can be edited out with an image editor. Embedding a pseudorandom cryptographic signature into generated text is possible, may be even difficult to defeat, and probably worth the effort if OpenAI wants to sell an app that checks for the signatures. However, I’ve seen enough content generated by ChatGPT that makes it somewhat easy to detect based on typical patterning of the responses, and if the human passing such content off as original doesn't have the smarts to match the level of generated material, well, that's an easy flunk.

Concluding, this is yet another grand challenge to the AI community, one I'll repeat from my last essay on the subject: "It's time to train, and in developing valuable human capital to be productive and innovative, it might just mean recruiting AI systems to do it, and do it BIG. Our future depends on it." No more unproductive BS please.

A few examples in my sessions with ChatGPT:

Assessment: ChatGPT is really a Q&A query tool, producing answers that closely mimic Wikipedia but in a slightly friendlier (but wordy) writing style. You may note, like I did, that the answers have a consistently typical verbose writing style that is easy to detect beyond Wikipedia, which is written by many different human contributors, with no consistent style. I prefer Wikipedia. I also prefer Google Search for basic questions with the range of material provided that I can choose or reject based on my own human intellect. ChatGPT can answer directed/tailored questions on topics, but it fails at open-ended questions that require sheer speculation (see my example below, where it renders an immediate fatal error, without recovery). When I asked Google Search the same question, I got back pages and pages of links, including a Wiki on "Initial singularity." ChatGPT is not really a training aid, since there is little human interaction and it cannot assess or understand the user. In fact, when asked follow-on questions to answers it provided for more technical clarity, it also returned a fatal error. Not ready for prime time, even as an effective Q&A query tool.

First Chat Session:

Q: Explain complex analysis of the Riemann zeta function

A: The Riemann zeta function, denoted by zeta(s), is a complex function of a complex variable s, defined as: zeta(s) = sum_{n=1}^{infty} frac{1}{n^s}, where s can be any complex number with a real part greater than 1. The Riemann zeta function is one of the most important functions in mathematics and has deep connections to number theory, complex analysis, and physics.

The complex analysis of the Riemann zeta function involves studying its properties and behavior as a function of a complex variable. In particular, the study of the Riemann zeta function is closely related to the distribution of prime numbers.

One of the most important results in the complex analysis of the Riemann zeta function is the Riemann hypothesis, which states that all non-trivial zeros of the zeta function lie on the critical line Re(s) = 1/2. This hypothesis has been a central problem in mathematics for over a century and remains one of the most important unsolved problems in the field.

In addition to the Riemann hypothesis, there are many other results in the complex analysis of the zeta function. For example, the zeta function has a meromorphic continuation to the whole complex plane, except for a simple pole at s=1. This means that the function can be extended to the entire complex plane in such a way that it remains holomorphic everywhere except for a single point.

Another important result is the functional equation for the zeta function, which relates its values at s and (1-s). This functional equation plays an important role in the study of the zeta function and its connections to other areas of mathematics and physics.

The complex analysis of the Riemann zeta function is a rich and fascinating topic that continues to attract the attention of mathematicians and physicists alike. While many important results have been obtained, there is still much to be discovered about the behavior of this function and its connections to other areas of mathematics and physics.

Q: How secure is a one-time pad cipher?

A: The one-time pad (OTP) cipher is theoretically unbreakable and provides perfect secrecy when used correctly. This means that, in principle, an OTP-encrypted message cannot be decrypted by anyone who does not have access to the key used to encrypt the message. This level of security is due to the properties of the OTP, which is a stream cipher that uses a truly random and secret key that is at least as long as the message being encrypted.

The security of the OTP cipher relies on two key assumptions: 1. The key is truly random and secret. 2. The key is used only once and then discarded. If these two assumptions hold, then the OTP cipher is unbreakable, since there are an infinite number of possible keys, and each key is used to encrypt only a single message. This means that any attempt to decrypt the ciphertext without the key is effectively impossible.

However, in practice, it is difficult to meet these two assumptions perfectly. For example, generating truly random keys is not always possible, and securely distributing and storing such long keys can be a challenge. Furthermore, it can be challenging to ensure that the key is used only once and then discarded.

Despite these challenges, the OTP cipher is still considered one of the most secure encryption schemes available. However, in practice, other ciphers, such as the Advanced Encryption Standard (AES), are often used instead of the OTP cipher, since they are easier to use and have other security properties, such as resistance to side-channel attacks.

Q: How do I prepare a biological assay?

A: Preparing a biological assay involves several steps to ensure accurate and reliable results. The specific steps will depend on the type of assay you are performing and the biological system you are working with, but here are some general guidelines:

1. Plan the experiment: Define the research question, select the biological system (cells, tissues, or organisms), and choose the assay type. Consider the number of samples you need, how to allocate them, and the control samples to use.

2. Prepare the samples: Collect and process the biological material, such as cells or tissue, according to established protocols. Be careful to minimize changes that could affect the results, such as degradation, contamination, or exposure to environmental stressors.

3. Choose the reagents: Select the appropriate reagents, such as assay kits, antibodies, or probes, that are specific to the analyte of interest. Test the reagents for cross-reactivity, sensitivity, and specificity.

4. Optimize the conditions: Optimize the experimental conditions, such as the concentration, timing, and temperature of the reagents, to ensure that the assay is sensitive and specific.

5. Run the assay: Follow the protocol and perform the assay according to the instructions. Record the results for each sample, including the controls.

6. Analyze the data: Calculate the results and analyze the data using appropriate statistical methods. Compare the results to the control samples and evaluate the significance of the differences.

7. Interpret the results: Interpret the results in the context of the research question and the biological system. Consider the limitations of the assay, the potential sources of error, and the reproducibility of the results.

8. Draw conclusions: Draw conclusions based on the results and provide recommendations for further experiments, if needed.

Overall, preparing a biological assay requires careful planning, attention to detail, and expertise in the biological system and the assay type. It is essential to follow established protocols and quality control measures to ensure that the results are accurate and reliable.

Q: What happened before the Big Bang?

A: ! An error occurred. If this issue persists please contact us through our help center at help.openai.com.

At this point I had to clear the conversation and request a new chat a few times, plus reload the page before the tool cleared the error. Perhaps it really does expect a directed question with no speculative nature!

New Chat:

Q: How do I spot weld a joint?

A: Spot welding is a process used to join two metal pieces together by applying heat and pressure to a specific area of the pieces. The process involves the use of an electrical current, and it's commonly used in the automotive, construction, and manufacturing industries. Here are the basic steps for spot welding a joint:

1. Prepare the metal pieces: The first step is to prepare the metal pieces that you want to weld. You should make sure that the surfaces are clean, smooth, and free from any debris or contaminants. Also, ensure that the two pieces are in contact with each other and properly aligned.

2. Position the electrodes: Position the electrodes of the spot welding machine on either side of the joint. The electrodes are the metal rods that will deliver the electrical current to the metal pieces. You should make sure that the electrodes are aligned with the joint and that they are in contact with the metal.

3. Apply pressure: Press the electrodes together with a significant amount of force. The amount of force required will depend on the thickness of the metal and the type of joint you are making.

4. Apply electrical current: Turn on the spot welding machine and apply an electrical current to the electrodes. The current will flow through the metal pieces, generating heat at the point where the electrodes are in contact with the metal.

5. Hold the current: Hold the current for a few seconds until the metal pieces reach the desired temperature. The duration of the current will depend on the thickness of the metal and the type of metal being used.

6. Release the pressure: After holding the current for the desired time, release the pressure on the electrodes and remove them from the metal pieces.

7. Inspect the joint: Inspect the joint to make sure that it has welded properly. If the joint has not been welded correctly, you may need to repeat the process.

It's essential to follow the safety precautions when performing spot welding, such as wearing appropriate personal protective equipment and ensuring that the area is well-ventilated.

Q: How do I inspect the spot weld or test the integrity of the weld?

A: ! An error occurred. If this issue persists please contact us through our help center at help.openai.com.

Oops! This was a valid question given the prior answer. When I asked Google Search the same exact question, "How do I inspect the spot weld or test the integrity of the weld?", I got back many suitable links and even a direct answer: "In the automotive industry, the most common method for inspection of spot welds is ultrasonic testing, but experience shows that the reliability of the test is often depending on the time, place, testing situation, and skill of the operator." No kidding!

P.S. I don't work for Google, OpenAI, or any other competitor in the space. This essay was written solely based on my personal interest in the subject matter, and the concern for the direction and progress of AI.