Fun and polite exchange. Unfortunately for mainstream astronomy, the JWST is lending much more support to the EU than anyone would have expected, and the "cosmology crysis" manifests as a growing silence and surprise from the experts.

IMO, this was an incredibly informative exchange between @Diadon and AI. Hope everyone reads it. I learned of this ChatGPT only a few weeks ago, and my mind is simply blown by it's potential. Not only to LENR, but all sciences.

On the other hand, if "she" gets involved in social issues, and politics, it may very well be the end of us.

Quote from Shane D.

On the other hand, if "she" gets involved in social issues, and politics, it may very well be the end of us.

We need to start thinking how to ingrain Asimov's three laws in every AI right now.

Quote from Curbina

We need to start thinking how to ingrain Asimov's three laws in every AI right now.

And they are?

Quote from Shane D.

And they are?

They are for robots but apply to AIs as well:

First Law
A robot may not injure a human being or, through inaction, allow a human being to come to harm.

Second Law
A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.

Third Law
A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.

The robots of Asimov's books are electromechanically embodied autonomous AIs.

(I now realize how a huge Nerd I am, I was reading Asimov whenI was 10 years old, and I assumed his three laws of robotics are part of the pop culture by now, LOL).

This exchange however, reveals that ChatGPT is not yet an AI. It is indeed remarkable but it still can't have original ideas stemming from current knowledge. It can compare point's of view, but it can't develop it's own. It can't create new knowledge.

Quote from Curbina

This exchange however, reveals that ChatGPT is not yet an AI. It is indeed remarkable but it still can't have original ideas stemming from current knowledge. It can compare point's of view, but it can't develop it's own. It can't create new knowledge.

There seems to be some debate about whether AI can be creative. There are some examples that it can. While the experts decide that issue, I see room for the community to exploit it as it is now.

For one, it is a great research tool. Unlike other branches of science, LENR history is largely unorganized, and difficult to navigate. The few good books about it were written years ago. That, IMO, leads to unnecessary duplication of effort. This CharGPT AI will go a long way to alleviating that problem. A new researcher, or one of the ARPA-E funded teams, can be spun up on their area of interest in literally minutes, avoiding "reinventing the wheel", and wasting their time as so many others before have.

And since AI has such command of the volumes of literature scattered all over the internet in an instant, it is able to tie things together we can't. Just this one chat with Diadon taught me more about the EU, what it was, and how it might incorporate LENR, and other phenomenon, than my years of being tuned in to the subject.

And it is only going to get better at what it does.

What I see interesting in AI is how it questions what is human intelligence. For me the errors of AI are so funny because humans make so similar errors, have similar hallucinations, illusions...

LLM intelligence is far from humans, like is the intelligence of a supercalculator solving differential equations, but since language is invented by neural networks with blood, not surprising neural networks understand each other so well through languages.

Some also remind us that human creativity is mostly remix, and AI do that well.

the way I start to use ChatGPT is to talk with a document. Without a document, talking wit ChatGPT is like talking with a librarian... On LENR my experience is that it is full of consensual sentences. LLM won't discover cold fusion theory, but it may find commonalities in LENR experiments...
Seriously, a well funded team should feed a LLM with LENR papers and ask for regularities, ideas, missing parts... the hard job that nearly nobody did, except maybe few librarian or reviewer...

Quote from AlainCo

What I see interesting in AI is how it questions what is human intelligence. For me the errors of AI are so funny because humans make so similar errors, have similar hallucinations, illusions...

LLM intelligence is far from humans, like is the intelligence of a supercalculator solving differential equations, but since language is invented by neural networks with blood, not surprising neural networks understand each other so well through languages.

Some also remind us that human creativity is mostly remix, and AI do that well.

the way I start to use ChatGPT is to talk with a document. Without a document, talking wit ChatGPT is like talking with a librarian... On LENR my experience is that it is full of consensual sentences. LLM won't discover cold fusion theory, but it may find commonalities in LENR experiments...
Seriously, a well funded team should feed a LLM with LENR papers and ask for regularities, ideas, missing parts... the hard job that nearly nobody did, except maybe few librarian or reviewer...

Display More

Finding common threads and parallels in already existing LENR research is what Bob Greenyer often talked that AI would really be useful for, but he includes all fields of science in the search as he thinks LENR has been hiding in plain sight. I agree this is probably true, but my direct experience is that ChatGPT is 95% of the time unavailable, so it's already a "scarce resource".

Quote from Shane D.

There seems to be some debate about whether AI can be creative. There are some examples that it can.

ChatGPT is a Large Language Model (LLM) AI. I think the consensus of opinion is that LLM cannot be creative. Based on what I know of it, I do not see how it could be. However, there are other AI models, and some of them are definitely creative. Decades ago, one of them was programmed to design electric and electronic circuits. It was given a goal, and it reinvented some patents filed by AT&T around 1900. The people at AT&T in 1900 were among the most creative in the world.

I expect that sooner or later the creative models will be incorporated in Chatbots.

ChatGPT lacks logic. I expect that will be fixed with things like the Wolfram plugin.

ChatGPT Gets Its “Wolfram Superpowers”!

Wolfram plugin gives computationally accurate answers to ChatGPT queries using Wolfram's data knowledgebase and language. Custom visualizations are given as…

writings.stephenwolfram.com

Quote from Curbina

I agree this is probably true, but my direct experience is that ChatGPT is 95% of the time unavailable, so it's already a "scarce resource".

You can upgrade to ChatGPT Plus for $20 a month.

Quote from Diadon Acs

It is not an AGI by any means, but still quite an incredible tool like Google. I wanted to actually feed it more data, as it had very little information on Taakaki Matsumoto and others.

I gave it the URL of a paper by me, and asked it to summarize the paper. It did a poor job.

I am still hoping I can arrange to have a local copy that looks at the papers in LENR-CANR.org only. I do not know if I can arrange that.

How To Build Your Own Custom ChatGPT With Custom Knowledge Base

Step-by-step guide on how to feed your ChatGPT bot with custom data sources

How To Build Your Own Custom ChatGPT With Custom Knowledge Base

Step-by-step guide on how to feed your ChatGPT bot with custom data sources

betterprogramming.pub

JedRothwell The above might be a useful tool for linking LENR-CANR to ChatBPT? I cannot comment on the time this would take, but it looks like a long job.

Quote from Alan Smith

The above might be a useful tool for linking LENR-CANR to ChatBPT? I cannot comment on the time this would take, but it looks like a long job.

Okay, going through the comments at that article, I have found this service:

ChatFast - Create GPT chatbot from your own data

Looking at their pricing plans, I do not think they can handle the LENR-CANR.org database of 1,166 documents, 2.4 GB total, at LENR-CANR.org. I sent them a message asking if they can deal with this.

I am a very bad programmer. I learned a long time ago to program in L.S.E. (Symbolic Teaching Language) This is a kind of primitive BASIC. I was using a Bull Mithra computer. It was the firing computer of the French main battle tank.

Programs were entered with punched cards. I kept the plastic boxes that were used to transport the programs in the form of punched cards. I use them to store my underwear and my socks. (picture)

All this to say that my computer skills are a bit dated. Jed will correct me if I say too much nonsense in this area:

But I remember that the innovation that is the basis of all AI is the work of a French researcher (I forgot the name of this woman)

Pattern or sentence recognition programs gave bad results because the system found non-optimal solutions, and then got stuck there.

You know that I like to reason by analogies. It's like a gold digger with his sieve. If you put graded sand in the sieve, and if the grains are smaller than the size of the holes, all the sand should pass through the sieve. This is not what happens: if you shovel the sand into the sieve, you get a nice pile of sand on the sieve and nothing underneath.

This metastable equilibrium must be destroyed by shaking the sieve. And there, the sand passes.

To get out of the partial solutions of AIs, you have to “shake” them, i.e. artificially introduce errors again and again, until the optimal solution is found, by a kind of computer Darwinism.

It works, and GPT4, Dall-e and Midjourney are here to prove it. But the problem is that they are black boxes, we don't know what's inside.

This is why I think that AIs will quickly show creativity, because creativity is the result of a cascade of unconscious errors that take us off the path well traced by our predecessors.

And we probably benefit from outside help, but that's another story, and if we are helped, why shouldn't the AIs also benefit from this providential help, too?