The LENR-CANR ChatGPT is ON LINE!

I asked the public, general purpose ChatGPT the same questions about Miles that I asked the LENR-CANR.org version. The general purpose answers were better. This could be because that is Version 4 and the LENR-CANR one is Version 3. I was surprised to see the general purpose one has access to so much information about cold fusion. Here is the exchange:

ME: In his research at China Lake, Melvin Miles used a copper sheath in one of his calorimeters. Describe this.

ChatBot: Melvin Miles used a copper sheath in one of his calorimeters to help improve the accuracy of his experiments on cold fusion. A calorimeter is an instrument used to measure the amount of heat released or absorbed during a chemical reaction or physical process.

The copper sheath in Miles' calorimeter served as a container for the electrolytic cell where the cold fusion reaction was taking place. The sheath was made of high-purity copper to minimize any interference with the reaction, and it was designed to provide good thermal contact between the cell and the calorimeter.

The calorimeter itself was a double-walled vessel, with the outer wall filled with water and the inner wall containing the copper sheath and the electrolytic cell. This design allowed for accurate measurement of the heat produced by the reaction, since any heat generated by the cell would be transferred to the water in the outer wall of the calorimeter.

The copper sheath also helped to reduce any potential sources of error in Miles' experiments, such as heat loss due to conduction or radiation. Overall, the use of a copper sheath in Miles' calorimeter helped to improve the accuracy and reliability of his measurements of cold fusion reactions.

ME: Why did he use a sheath instead of temperature sensors in the electrolyte?

ChatBot: Melvin Miles used a copper sheath in his calorimeter instead of relying solely on temperature sensors in the electrolyte because temperature measurements alone are not sufficient to accurately measure the amount of heat produced by a chemical reaction or physical process.

[NOT QUITE RIGHT, I think. There were no sensors in the electrolyte.]

While temperature sensors can provide useful information about the temperature changes occurring in the electrolyte, they do not account for all the heat generated by the reaction. For example, heat can be lost due to conduction, convection, or radiation, and these losses can significantly affect the accuracy of temperature measurements.

By using a copper sheath, Miles was able to provide a more accurate measurement of the total amount of heat produced by the reaction, as the sheath acted as a heat sink, absorbing any heat generated by the reaction and preventing it from being lost to the surrounding environment. This allowed him to more accurately determine the energy output of the cold fusion reaction.

In addition, using a copper sheath also helped to reduce potential sources of error, such as thermal gradients in the electrolyte or changes in the thermal conductivity of the electrolyte over time. Overall, the use of a copper sheath in Miles' calorimeter allowed him to obtain more accurate and reliable measurements of cold fusion reactions than would have been possible with temperature sensors alone.

Here we see the LENR-CANR.org bot going off the deep end with hallucinations --

Hello! What can I do for you today?

In his research at China Lake, Melvin Miles used a copper jacket in one of his calorimeters. Describe this.

According to the data provided, Melvin Miles improved the China Lake calorimetry by using a copper inner jacket that acted as the integrator and replaced the H2O jacket. An insulating box over the cell top and bath greatly reduced the effect of changes in the room temperature. Furthermore, the lead wires coming out of the cell were thermally staked to the bath itself, thereby further reducing the effect of the room temperature. A copper outer jacket contacted the bath and minimized bath level. These improvements are illustrated in Figure 4 of the data provided.

What document is Figure 4 in?

Figure 4 is in the EPRI Proceedings: Fourth International Conference on Cold Fusion Volume 4: Theory and Special Topics Papers, TR-104188-V4, which was published in 1994 by the Electric Power Research Institute.

[There is no paper by Miles in Volume 4.]

Do you have a URL for EPRI Proceedings: Fourth International Conference on Cold Fusion Volume 4: Theory and Special Topics Papers, TR-104188-V4?

Yes, the URL for EPRI Proceedings: Fourth International Conference on Cold Fusion Volume 4: Theory and Special Topics Papers, TR-104188-V4 is http://my.epri.com/portal/server.pt?Abstract_id=TR-104188-V4. [This URL was provided by me. It is defunct.]

Do you have the title of the paper with Figure 4?

The title of the paper with Figure 4 is "Improved Calorimetry at China Lake" by Melvin H. Miles, which is included in the EPRI Proceedings: Fourth International Conference on Cold Fusion Volume 4: Theory and Special Topics Papers, TR-104188-V4.

[There is no such paper anywhere in the literature, as far as I know. The exact phrase "Improved Calorimetry at China Lake" does not appear anywhere in the literature, although variations of it do appear, in papers by Miles. For example, this reference p. 13 says:

https://lenr-canr.org/acrobat/MilesManomalousea.pdf

... An improved calorimeter recently developed at China Lake that can accomplish these tasks is described in the next section. In June of 1995, ...]

The vendor chatfast.io says they may be able to fix some of the problems I have encountered. Especially problems finding names and words in documents. Such as my futile search for giraffes. I thought these problems were caused by OpenAI.com but they might be a parameter setting or something. I don't know. I hope they can improve things.

It seems I have uploaded much more data into a bot than any other chatfast customer. The bot broke down several times when I was doing this. They had to reset it and change the parameters. I am stress-testing their software.

Here is a "gift copy" of an article from the New York Times describing AI hallucinations, and what causes them. "Gift copy" means it goes around the paywall.

When A.I. Chatbots Hallucinate

https://www.nytimes.com/2023/0…-wuyWH7Gl6&smid=url-share

I am replacing .pdf files in the database with .txt files. The vendor is improving the search function. This will slow down the Bot. I think accuracy is more important than speed. I have not yet uploaded the .txt files. I uploaded some in my Testbed Bot. Here are some responses from that Bot relating to https://lenr-canr.org/acrobat/BiberianJPjcondenseda.pdf

[After finding "a paper about an explosion" . . .]

. . . In the paper "Unexplained Explosion During an Electrolysis Experiment in an Open Cell Mass Flow Calorimeter" what kind of cathode was used?

In the paper "Unexplained Explosion During an Electrolysis Experiment in an Open Cell Mass Flow Calorimeter" by Jean-Paul Biberian, a hollow palladium cathode was used in the electrochemical cell designed to measure excess heat.

Describe the calorimeter in this paper.

The paper does not provide a specific description of the calorimeter used. Can you please provide the title and author of the paper you are referring to?

Describe the calorimeter in the paper "In the paper "Unexplained Explosion During an Electrolysis Experiment in an Open Cell Mass Flow Calorimeter."

The paper "Unexplained Explosion During an Electrolysis Experiment in an Open Cell Mass Flow Calorimeter" by Jean-Paul Biberian describes a constant-heat-loss-rate water calorimeter that was built for the specific purpose of the experiment. The electrochemical cell, located in the water of the calorimeter, was controlled at 30.0°C. The constant-temperature bath that surrounded the calorimeter was held at 21.0°C by flowing the water through a controlled-temperature circulator. A heat-loss-rate control gap regulated the rate of heat transfer between the calorimeter and the bath. The heat loss rate (Qc) was varied by placing air in the gap for levels of cell heat production up to 5 W and water in the gap for levels up to 30 W. The bath was thermally insulated from the surroundings by thick layers of styrofoam. The cover was insulated to prevent any significant direct heat loss to the ambient air through the top of the cell. The heat loss from the calorimeter was calibrated with a precision electrical heater that was placed into the electrolytic cell.

Here is another example for http://lenr-canr.org/acrobat/BiberianJPjcondensedzi.pdf

list all papers that discuss heat bursts.

The papers that discuss heat bursts in the data provided are:

- "Optical Observation of Spontaneous Heat Burst Phenomena during Hydrogen Desorption from Nano-Sized Metal Composite" by Takehiko Itoh et al.

- "Elemental Analysis for Elucidation of the Anomalous Heat Generation Phenomena" by Y. Iwamura, T. Itoh, J. Kasagi, M. Saito, S. Murakami, and Y. Shibasaki

- "Evidence for Surface Heat Release Reaction over Nano-sized Multilayer Metal Composite with Hydrogen Gas" by Y. Iwamura, T. Itoh, M. Saito, S. Murakami, and J. Kasagi

- The section "3.3. Heat Burst Phenomena" in an unspecified paper.

Okay! I have upgraded all issues of the CMNS. The ChatBot now finds data in them more readily. It finds Christy L. Frazier. She is no longer invisible to the robot.


ME: Do you have any information on Christy L. Frazier?

BOT: Yes, Christy L. Frazier was involved in the administration of support and technical materials for the attendees of a conference related to new energy. She also helped with the manning of the book desk and had a role in the major US media coverage of the conference.


Another response: Yes, according to the provided context, Christy Frazier of the New Energy Foundation helped with administration of some support and provided technical materials for attendees at a conference.

But the Bot is still confused . . .

ME: What was the title of the report Christy L. Frazier wrote?

BOT: There is no information provided about Christy L. Frazier writing a report.

ME: Do you have any information on the report "ICCF23 Held Virtually from China"?

BOT: Yes, according to the provided context, "ICCF23 Held Virtually from China" was a detailed report written by Christy L. Frazier and published in Issue 157 of Infinite Energy in August of 2021. The report covered the International Conference on Condensed Matter Nuclear Science (ICCF-23) which was held virtually due to the COVID-19 pandemic. The report provided information on the conference program, participants, and presentations.

The ChatBot may have confused Christy with Macavity, the mystery cat who was not there.

Macavity: The Mystery Cat by T. S. Eliot - Poems | poets.org

Here is another example of how to find things. I did not know about some of these events.

Have there been explosions during cold fusion experiments?

Yes, there have been reports of explosions during cold fusion experiments. For example, in 2015, Andrea Rossi's alleged cold fusion reactor, the E-Cat, ended explosively after the reactor heated to over 1000°C. There have also been unpublished reports of runaway events in LENR and related experiments by Francesco Piantelli, Stanislaw Szpak, Mark Snoswell, Vince Cockeram, and Brian Ahern. Additionally, there have been documented explosions during tests by S. B. Krivit and J.-P. Biberian.

In what papers did J.-P. Biberian describe explosions?

J.-P. Biberian described an unexplained explosion during an electrolysis experiment in an open cell mass flow calorimeter in a paper published in the Journal of Condensed Matter Nuclear Science in 2009. He also re-analyzed the explosion that occurred in his laboratory in 2004 via a review of the literature and dedicated experiments, which he described in a paper published in the same journal in 2016.

What are the titles of the two papers about explosions by J.-P Biberian?

The two papers about explosions by J.-P. Biberian are "Unexplained Explosion during an Electrolysis Experiment in an Open Cell Mass Flow Calorimeter" published in the Journal of Condensed Matter Nuclear Science in 2009 and "Re-analysis of the Explosion that Occurred in Our Laboratory in 2004 via a Review of the Literature and Dedicated Experiments" published in the same journal in 2016.

DId Fleischmann and Pons describe an explosion?

Yes, Fleischmann and Pons described an event in which their experiment was destroyed in a journal article. They reported that under the conditions of the last experiment, even using D2O alone, a substantial portion of the cathode fused, part of it vaporized, and the cell and contents and a part of the fume cupboard housing the experiment were destroyed. However, it is now clear that the event was a meltdown and not an explosion.

What was the title of the paper or papers in which Fleischmann and Pons described this explosion?

Fleischmann and Pons described the event in which their experiment was destroyed in a journal article titled "Electrochemically induced nuclear fusion of deuterium" published in the Journal of Electroanalytical Chemistry in 1989.

Do not believe everything the Bot tells you.

What is McKubre's equation describing the necessary parameters to produce excess heat?

McKubre's equation describes the variations in excess power production in electrochemical LENR experiments with three quantities: the electrochemical current density passing through the cylindrical cathode surface, the loading factor (specifically, the ratio of the amount of deuterium loaded into the cathode to the amount of palladium in the cathode), and an unknown proportionality constant. The equation is Pxs = K(I-Ic)αLβ, where Pxs is the excess power production, I is the electrochemical current density, Ic is a critical current density, L is the loading factor, and α and β are constants.

[WRONG EQUATION! I think it should be Pxs = M (x-x°)2(i-i°) |iD|]

What is the minimum loading needed to produce excess heat?

In the SRI experiments, it was found that cathodes that had not loaded to a peak D/Pd loading of about 0.95 were unlikely to produce an excess heat effect. However, in order to achieve appreciable (and measurable) power excess, loadings much higher than the threshold are needed (D/Pd ;;:: 0.9 and preferably ;;:: 0.95).

Various techniques are used to measure loading. List some of them.

Various techniques are used to measure loading, including resistivity measurement, dilation measurements, measurement of shift in rest potential, gravimetry, and in-situ measurements of lattice parameters by X-ray diffraction. However, most of these measurements tend to overestimate the measured D/Pd ratio when there is cracking or deformation, and any data presented must be used with caution.

[NOT SURE ABOUT SOME OF THESE METHODS.]

Describe the "orphaned oxygen" method of measuring loading.

The "orphaned oxygen" method measures the extra oxygen left behind after the deuterium released from D2O reacts with Pd. The volume of oxygen is measured by displacing oil onto a balance, from which the moles of deuterium reacted with Pd can be calculated. The temperature of the recombiner is used to determine how much of the decomposed H2O is present as unreacted H2 and O2, which when compared to the amount of H2O that is decomposed by applied current, results in a value for how much deuterium reacted with Pd.

The vendor changed the control parameters for the Chatbot (https://lenr-canr.org/wordpress/?page_id=2988). When you ask a question, it now takes longer to respond. I think it often takes about 30 seconds. It is worth the delay, because it is improved in various ways. It does a deeper search. I think the answers are more accurate. It remembers context better, from one query to another. You do not have to keep repeating the subject, or the name of the paper. It does a better job finding a URL (where that is available), or quoting a section of the paper verbatim. A verbatim quote is handy because you can use it in a Google search to locate the paper it comes from.

Here is an example, for the Index Bot (the one on top).


BOT: Hello! What can I do for you today?

ME: Has E. Storms written a paper about how to construct a calorimeter?

BOT: Yes, E. Storms has written a paper titled "How to Make A Cheap and Effective Seebeck Calorimeter" which describes how to construct a Seebeck calorimeter that is effective in measuring heat generation over a wide range of power with high sensitivity and stability. The paper also provides a successful example of constructing such a device, although it requires considerable investment of time. The paper was presented at the Tenth International Conference on Cold Fusion in 2003 and can be found on the LENR-CANR.org website.

ME: What is the URL of this paper?

[With the previous settings, the Bot would have asked, "what paper?" It lost track.]

BOT: The URL for the paper "How to Make A Cheap and Effective Seebeck Calorimeter" by E. Storms is http://lenr-canr.org/acrobat/StormsEhowtomakea.pdf.

ME: What does the abstract say? Quote it verbatim.

BOT: "The Seebeck calorimeter is very effective in measuring heat generation over a wide range of power and with high sensitivity and stability. Such a device can be constructed cheaply and easily, although with considerable investment of time. A successful example is described."


Note: If you don't say "quote it verbatim" a Chatbot is apt to summarize the abstract instead.

I have replaced some of the files with new versions in text format, derived from the Amazon AWS Textract OCR program. They work better than the previous version. I will replace all of the files in the coming days. If you want to test the new files and the new, slower but better performance, ask any question from any issue of the JCMNS. JCMNS issues are here:

https://lenr-canr.org/wordpress/?page_id=1495

For old documents, the AWS output is a bunch of undifferentiated lines of text, with no breaks between paragraphs. A Chatbot must have paragraphs, or it loses track. It cannot handle paragraphs longer than 512 tokens. I wrote a program to do various things, such as removing page headings and breaking the text into paragraphs. I used quick and dirty, hard coded methods, since the formats of the old papers are similar. The program is working pretty well, but this morning, I asked ChatGPT how to improve it. This is hardly what you would call a sophisticated program. It is more like a high-school assignment. But I did not feel like spending a few more days improving it. ChatGPT made some useful suggestions, and pointed out something I overlooked, which is that a paragraph ends with a punctuation mark. This is a simple job, but I am still amazed at how well ChatGPT understood it, and how it made useful suggestions.

Here is the . . . conversation? Exchange?

ME: Good morning! [TRIED TO CONFUSE IT WITH THIS.] I am using AWS Textract to OCR old scientific papers. Textract outputs all of the lines of the paper without breaks between paragraphs. Sample text is below. I need to add line breaks (\n\n) between paragraphs. I wrote a simple Pascal program to do this. It works pretty well, but I wonder if you can suggest some improvements. The program compares the length of lines of text. A line that is much shorter than the previous line is probably the end of a paragraph, so I insert a line break. Here is a snippet of the program:

{S is a line of text from the document. PaperRevised is the new version, with line breaks between paragraphs}

PaperRevised.Add(S);

ThisLen := Length(S);

LastLen := Length(Paper.Strings[RecPnt-1]); {Line before this}

NextLen := Length(Paper.Strings[RecPnt+1]); {Line after this}

if ThisLen<LastLen-12 then PaperRevised.Add(''); {New paragraph}

if NextLen>ThisLen+25 then PaperRevised.Add(''); {This line short and before a paragraph}

Here is some sample text. Please break this into paragraphs, and then suggest a better way to detect the end of a paragraph:

Possible tritium in the palladium was eliminated by pre-analyzing the metal.

This was done by sweeping any tritium out during anodic electrolysis (a process

known to remove tritium along with all other hydrogen isotopes[28]) and by analyz-

ing the palladium for Superscript(3)He, the tritium decay product. The absence of 3 He shows that

the anomalous tritium was not present in the palladium during prior years. The ab-

sence of tritium after anodic electrolysis shows that it was not dissolved in the palla-

dium immediately before the experiment. Only pre-analyzed, hence pre-purified,

palladium was used for the experiment.

If the tritium found in this cell, or that claimed in more than ten other well-

documented studies, can not be rationally explained as contamination, the existence

of a completely unexpected nuclear reaction must be acknowledged.

3.3 Helium Production

Of the nuclear products, only 4He has been found in sufficient quantity to

roughly account for the observed excess energy. Most helium is found in the gas

with lesser amounts being retained by the metal. [15; 30; 31; 32; 33; 34; 35] Charged

particles similar to energetic helium emission (a-particles) have also been detected

when suitable detectors were in place. [36; 37; 38; 39; 40; 41] Therefore, this product

appears to be a major part or perhaps all of the sought for nuclear ash required to

explain energy production.

ChatGPT

Sure, here's your text properly broken down into paragraphs:

Possible tritium in the palladium was eliminated by pre-analyzing the metal. This was done by sweeping any tritium out during anodic electrolysis (a process known to remove tritium along with all other hydrogen isotopes[28]) and by analyzing the palladium for Superscript(3)He, the tritium decay product. The absence of 3 He shows that the anomalous tritium was not present in the palladium during prior years. The absence of tritium after anodic electrolysis shows that it was not dissolved in the palladium immediately before the experiment. Only pre-analyzed, hence pre-purified, palladium was used for the experiment.

If the tritium found in this cell, or that claimed in more than ten other well-documented studies, cannot be rationally explained as contamination, the existence of a completely unexpected nuclear reaction must be acknowledged.

3.3 Helium Production

Of the nuclear products, only 4He has been found in sufficient quantity to roughly account for the observed excess energy. Most helium is found in the gas with lesser amounts being retained by the metal. [15; 30; 31; 32; 33; 34; 35] Charged particles similar to energetic helium emission (a-particles) have also been detected when suitable detectors were in place. [36; 37; 38; 39; 40; 41] Therefore, this product appears to be a major part or perhaps all of the sought for nuclear ash required to explain energy production.

As for a better way to detect the end of a paragraph, you could try using natural language processing (NLP) techniques to analyze the structure of the text rather than relying solely on line length. For example, you could look for sentence-ending punctuation marks such as periods, question marks, and exclamation points to indicate the end of a paragraph. Additionally, you could use a library or API that has built-in support for text segmentation and paragraph detection. Some popular options include NLTK (Natural Language Toolkit), spaCy, and CoreNLP.

ME: Can you suggest a way to detect a one-line section heading, such as "3.3 Helium Production" in this sample?

ChatGPT

Yes, here's a simple way to detect a one-line section heading like "3.3 Helium Production" in your sample text:

Check if a line has a period or colon at the end, which is often used in section headings.

Check if a line contains only digits and a period or colon, which is a common format for section numbers.

Check if the line is shorter than a certain length (e.g., 30 characters), which is often the case for section headings.

Based on these checks, you could add a condition to your existing program to detect section headings and insert an additional line break before them to separate them from the preceding paragraph.

Here's an example implementation in Pascal:

for i := 0 to Paper.Count-1 do

begin

S := Paper.Strings[i];

ThisLen := Length(S);

// Check if line is a section heading

if ((Pos('.', S) > 0) or (Pos(':', S) > 0)) and

(Length(Trim(S)) <= 30) and

(Length(Trim(AnsiReplaceStr(S, '.', ''))) = Length(Trim(AnsiReplaceStr(S, '.', '')))) then

begin

PaperRevised.Add('');

end;

// Check if line is end of paragraph

if (i > 0) and (i < Paper.Count-1) then

begin

LastLen := Length(Paper.Strings[i-1]);

NextLen := Length(Paper.Strings[i+1]);

if ThisLen < (LastLen-12) then

begin

PaperRevised.Add('');

end

else if NextLen > (ThisLen+25) then

begin

PaperRevised.Add('');

end;

end;

PaperRevised.Add(S);

end;

(This came out splendidly formatted on the screen, like so:)

ME: That is helpful! I derived the "LastLen-12" and "ThisLen+25" by measuring the average length of lines in these documents. Those are "hard coded" values applicable to these documents. Should I use some function such as standard deviation to derive more accurate values?

ChatGPT

Yes, using standard deviation would be a better way to determine the threshold values for detecting paragraph breaks.

Here's an example of how you could use standard deviation to calculate more accurate threshold values:

Calculate the mean line length and standard deviation of line lengths in your document. You can do this by looping through each line and calculating the length of each line, and then using the built-in Pascal functions to calculate the mean and standard deviation. For example:

var

LineLengths: array of Integer;

MeanLength, StdDev: Double;

i: Integer;

begin

// Loop through each line and calculate the length

SetLength(LineLengths, Paper.Count);

for i := 0 to Paper.Count - 1 do

begin

LineLengths[i] := Length(Paper.Strings[i]);

end;

// Calculate the mean and standard deviation of line lengths

MeanLength := Mean(LineLengths);

StdDev := StdDev(LineLengths, MeanLength);

end;

Set the threshold values for detecting paragraph breaks to a multiple of the standard deviation away from the mean line length. For example, you could use a threshold of 1.5 times the standard deviation away from the mean line length to indicate the end of a paragraph. This would look something like:

if (ThisLen < (MeanLength - 1.5 * StdDev)) then

begin

PaperRevised.Add('');

end

else if (NextLen > (MeanLength + 1.5 * StdDev)) then

begin

PaperRevised.Add('');

end;

By using standard deviation to calculate threshold values, you can make your program more adaptable to different types of documents with varying line lengths.

ME: So, I can test various threshold values starting at 1.5 to see which ones work best? Increasing or reducing from 1.5. Right?

ChatGPT

Yes, you can experiment with different threshold values to see which ones work best for your specific documents.

Starting with a threshold value of 1.5 times the standard deviation is a good starting point, but you may find that a different value works better for your documents. You can try increasing or decreasing the threshold value in increments of 0.5 or 1 until you find a value that works well for most of your documents.

It's worth noting that the optimal threshold value may vary depending on the specific characteristics of your documents, such as the font size, spacing, and layout. So it's important to test your program on a representative sample of documents to ensure that it's detecting paragraph breaks accurately.

ME: Got it. Bye!

ChatGPT

You're welcome! Don't hesitate to ask if you have any more questions in the future. Goodbye!

By the way, this program written by ChatGPT has errors:

// Calculate the mean and standard deviation of line lengths

. . .

StdDev := StdDev(LineLengths, MeanLength);

1. You should not use the reserved word "StdDev" as a variable.

2. The StdDev function takes only one parameter: ThisStdDev := StdDev(LineLengths);

People keep saying that ChatGPT does a good job writing program snippets. Just about every snippet I asked for had errors like this. I can spot them easily, and fix them. The compiler usually indicates the problem. But ChatGPT is not a foolproof way to generate code. The fact that it works at all is astounding.

All of the files for the ChatBot have been replaced with text files Performance should improve. It seems to find the URLs more readily.

It turned out several of the papers were not machine readable. The vendor told me that in some cases the internal files in the bot were filled with question marks and periods instead of text. That indicates the document format was set incorrectly, or it was way out of date. All of those files have been fixed.

I tested a number of files in English, Japanese and Chinese. It seems to find documents that it could not find before. In one case, I asked in English about a detail that is in only one paper by Zhang, in Chinese. It answered with the correct information translated into English. That is remarkable performance.