Thursday, February 2, 2023


TL;DR -- As promised, we are getting technical. And, given our long time and wide expanse related to life on these shores (and beyond), that covers a whole lot of territory. We have started to look at ChatGPT as a representative of the most modern of approaches (take that with a grain of salt). Actually, we see lots of weaknesses which have been brought up before. Was this a waste? Not really, as we need failures to learn. This is not a failure so much as a work in progress that has further to go than they thought would be the case. Except, their mission statement now mentions the uncertainty. ... See Remarks on 02/11/2023, for an opinion on the matter. 


We are focusing on technology as the major theme for research. Roots Tech is an example of an application of technology. We haven't attended, as our interest are with the deeper issues. Say, with DNA, the whole of the technological basis plus the computational lens required to support all of the various ways of making interpretations. Computing in general? The whole of the framework that provided support for development as we have seen that ubiquity is more than mere presence. No science is done without some type of artificial routine kicking to action. 

Our daily lives? About the same thing. For good or bad, that's our world. 

I have been listening to discussions about OpenAI. Other than noting that Musk provided some of the funding and that Microsoft just put their dollars on the line, I have merely touched upon the foundational aspects. NNs go back a long way. The mathematics involved in these new approaches goes back even further. What's new? Hardware is one thing and is far beyond what was thought possible. Software and those in-between modes has a very interesting history, too. 

ChatGPT is from OpenAI and will tell you that when it uses 'I' to express itself and try to explain its ways. Don't expect depth, yet. As an article in The Atlantic noted, it's a toy now. But, some are trying to use it as a tool (several APIs are provided). That raises hackles for good reasons. For a few years now, even the brainy have weighed in with their dread of these approaches. On the other hand, the marketing view has grown to be worse than unsustainable, as hype eventually eats up all of the oxygen in a room. But, how long that will play out is, per usual, unknown. But, it gives us two sides to watch as things evolve. 


Except, technology needs the middle way which we are up to. Lots to explain there, but systems development is a good example. From the top, we have various demands that get studied, and decisions can be made as to coming up with a product (applies across the board). This is a top-down position that is always the one of power, not necessarily of truth. Make it happen, we see the Star Trek captain saying; or, he uses something of that ilk. 

BTW, expectations come this way and are very hard to control, let alone try to manage rationally. It's apparent always, especially given the news ability to harp. On the other hand, it is from this side of things that we get the systems management and all of those creative things dealing with function and form. We will go into examples of this as we go along. 

But, at some point, things have to happen. Arm-chair quarterbacking will not get a game won. Thank God for the industry of engineering which gets thing done. But, before that, we need the basic knowledge that comes into play (science is both top-down and bottom-up). At the same time, we need the ones with the abilities to actually do something with their hands, with or without tools. 

Getting a system done requires lots of abilities. And, a system will get (usually or hopefully) to the state where it does what and how as is realizable. Does that state necessarily agree with the earlier thoughts coming from the top-down crowd? If you think yes, well, that is not the usual case. The middle is where adjustments get made for the bottom-up to be driven by management while those above figure how to get their views allowed (aligned) with actual potentials and not marketing dreams. 

One might say that this is old-time management science. Well, yes, in a sense. The new perturbator? The computer brings whole new classes of issues that need attention. As it becomes more capable, so too does the computer's problems rise. That's our current situation where we see, on the horizon, looming changes which may or may not have desirable results. 


ChatGPT was released for public use and comment back in November of 2022. We watched, mainly, but knew that we had to get involved so as to make our own assessments. Yesterday was the first exposure which started analysis with respect to the work involved, past, current and future. We found it shallow, to the extreme. That comes from the approach which we can discuss. The interface is polite in its verbiage and strives for a 'neutral' position in its responses. 

A huge reminder is there: this is an AI system created by OpenAI and has known limits. We will get into that, but they say that all of the training data comes from before 2021. Okay. There is more that they ought to acknowledge; the input data is a hodgepodge. How will such formulate into a coherent view seemingly willy-nilly? Oh, patterns. We can immediately show that mathematics is not just about patterns; nor, is science of that ilk when done at its core. 

Don't expect creative output, of a valuable nature. Though, I might add, ChatGPT can feed into the graphic generator which comes up with pretty pictures. In case this is not known, the training results in a composited model(s) from which one can obtain what is a photo. Say, of people, and it gives one a 'fake' person. What is the accomplishment? There is no reality behind that. 

Wait, there could be which we will explain, at some point. Actually, what we are suggesting would be a type of analytic stance of benefit. One thing that alarms people is that we can get 'fake' video which is a series of photos. As noted by the discerning eye, the morphs needed are not very sophisticated. That is, one can see the effects of mods that generate behavior through incremental changes to the underlying structure, say a face getting its lips moved to look as if speaking. 


ChatGPT, 2 Feb 2023
Our first attempt ended in some errors, as did today's. Yesterday, we were testing its knowledge of itself which it handled very well, in most cases: I am ChatGPT created by OpenAI and only know what the trained data allows me to know. Nothing more; I do not reason (yet). 

I pulled some of the log from queries related to Thomas Gardner. Before this, there had been queries related to similar themes. 'Query' is my question; the response follows. Notice that it was in a 'terse' mode which is nice. Then, I asked for a 'verbose' answer to the prior question. 

Query: Who is Thomas Gardner of Salem MA?

The specific Thomas Gardner of Salem, Massachusetts you are referring to is unclear. "Thomas Gardner" is a common name and there may have been multiple individuals with that name living in Salem, Massachusetts. Could you please provide additional context or information to help identify the Thomas Gardner you are inquiring about?

Query: Thomas Gardner married to Margaret Fryer. He died in 1674. 

I apologize, but with only this limited information, it is difficult to determine the specific Thomas Gardner you are referring to. The name "Thomas Gardner" and "Margaret Fryer" were both common during that time period and there may have been multiple individuals with those names in Salem, Massachusetts. If you have access to any additional details, such as birth dates or occupation, it would be helpful in identifying the Thomas Gardner in question.

Query: Did you read about Roger Conant?

Yes, I have knowledge of Roger Conant. Roger Conant was a colonial leader and one of the founding fathers of the Massachusetts Bay Colony. Born in England in 1592, he immigrated to the Massachusetts Bay Colony in 1623 and was an early leader in the colony, serving as governor of the settlement at Salem. Conant was a key figure in the establishment and growth of the Massachusetts Bay Colony and played an important role in the early history of the colony and the development of New England. He died in 1679 at the age of 87.

Query: Thomas Gardner was the lead before Roger Conant of the expedition that settled Cape Ann in Massachusetts in the 1623/24 timeframe? 

Yes, that is correct. Thomas Gardner was the leader of the expedition that established a settlement on Cape Ann in Massachusetts in the 1620s, prior to the arrival of Roger Conant. Gardner led a group of settlers to Cape Ann in the early years of the Massachusetts Bay Colony, and the settlement he established there was one of the first permanent English settlements in the area. Roger Conant later became governor of the settlement at Salem, which was located near Cape Ann, and played a prominent role in the early history of the Massachusetts Bay Colony.

Query: Verbosely, please.

Thomas Gardner was a colonial leader and one of the early settlers of Massachusetts Bay Colony. In the 1620s, he led a group of settlers to establish a settlement on Cape Ann, which is located in present-day Massachusetts. This settlement was one of the first permanent English settlements in the area and was established prior to the arrival of Roger Conant, who later became a governor of the settlement at Salem and played an important role in the early history of the Massachusetts Bay Colony.

Gardner was a visionary and a pioneer, who saw the potential of the New World and was willing to take risks to make it a reality. He played a key role in the establishment and growth of the Massachusetts Bay Colony and helped to lay the foundation for the development of the region. Despite the challenges he faced, Gardner persevered and helped to build a thriving community on Cape Ann.

Today, Thomas Gardner is remembered as one of the early settlers of the Massachusetts Bay Colony and as an important figure in the early history of New England. His legacy serves as a testament to the bravery and determination of the early settlers who helped to shape the American landscape and lay the foundation for the growth and development of the nation.

On the last? It's not quite right, but then it is a composite from several sources. This type of approach might appear to be of interest but is not. Debatable, of course. 

Now, we will get to the specifics, but the training data set was large and came from the internet: pages, sites, books, and such. What a heap! Right? That would confuse anyone. 

There is so much to discuss. We got it to say that the training set was a subset and not just the whole mess. After all, the internet is a mess, for several reasons and of various types. One thing that sticks out is that the 'domain' aspect of life has been ignored. That is, architecture with respect to this type of technology is very weakly evident, perhaps nonexistent. 

One failure, last night? I suggested to the thing that we and it make errors. Right. But, to know how to assess its errors, we need details related to 'provenance' in the proper sense. And, ChatGPT admited that it cannot specifically respond to anything that would be like a 'footnote' request. Alas. ... On trying to respond to that, the system crashed (red text, emphasized - several times). On getting things going again, I was told that I had exceeded the number of queries per hour. Oh well, it and I were in sync ;>). 

So, I was thrown out and the party ended. 

Remarks: Modified: 02/11/2023

02/05/2023 -- There is a lot to this. We will focus on GPT3 which is that which ChatGPT invokes for its content. Too, Google has a player, too: LaMDA (remember fall 2022 and claims of sentience?). 

02/11/2023 -- This is not an official TGS opinion, it is mine. And, I (JMS, starter, with Ann, of the TGS, Inc.) need to give some overview. Why? I have been decades doing advanced work with computers where the focus was involved in 'real' issues related to application of these things. It has been a long road that these types of programs (meaning systems-level focus) and projects have hoed. Some results have been remarkable, say MRIs and other imaging, or CAT/CAE/CAM (advance mathematics applied to design and engineering), communications (look at the 'cloud' for instance - but with rose-colored glasses as that place is a mess), scientific experiments (but the issue of non-repeatability is there; and, other issues; be aware that the framework's strength will keep problems from being obvious), medicine (too many to note, however I will say that I was in top-notch surgical environments back in the '60s in the U.S. Army and saw the state of the art then; to me, robotic surgery is far from something that we ought to trust), and much more. Essentially, computing is everywhere. That is why immature and irresponsible acts are and ought to be concerning to anyone thinking of the future and their progeny of that time. Pollution is one analog such as we got the EPA for. There are other descriptions that I will do later. ... Truth engineering is what the TGS is trying to raise to focus. ... Disclosure: my focus is not Comp Sci. No, it comes from the underlying mathematical frameworks that are being used. Necessary care does not seem evident since a product ought not be released until fully analyzed with respect to several factors, one of which is safety. The main issue? Allowing API access which is hidden from view for something that is not really a product, nor has it any ability as expected by most. Cute, but poisonous to our mental health, might be one depiction of the state of the thing. 

No comments:

Post a Comment