Preparation of column 2 on ChatGPT
2023-05-22
--- log
/nishio/kei.iconWhat would you like to see happen in this conversation?
/nishio/human.iconI've been asked to write a column about ChatGPT and I'd like to organize my thoughts by talking about it.
/nishio/kei.iconWhat kind of "ChatGPT" is this "ChatGPT"?
/nishio/human.iconThis is an additional study of InstuctionGPT on top of LLM so that even the average person with no technical knowledge can pull off the results. At the time of the request, ChatGPT was the only service available, but now there is a wide variety of services, so it would be better to discuss what the LLM brings to the table rather than individual services.
/nishio/kei.iconIs there anything else about that "ChatGPT"?
/nishio/human.iconI was asked to write an article about ChatGPT, and I was thinking about what to write, but I realized that even if the language model is smart, the demand for "knowledge that mankind does not yet know = unspoken knowledge and putting it into words" will not disappear, and that it is inefficient if what is put into words is not promptly learned by AI So, as a natural consequence, the results of research should be open access promptly, and non-open access contributions will be negatively incentivized by not getting mentioned via AI... and I was thinking about this, and then I mentioned DeSci... I feel like I need to mention DeSci.
/nishio/kei.iconHow do you know its "ChatGPT"?
/nishio/human.iconWe know by actually using the service. The API gives you a little more insight into what's inside. On a related note, I recently wrote an article titled "Don't talk about ChatGPT without experiencing GPT-4.
/nishio/kei.iconWhere is this "ChatGPT" located?
/nishio/human.iconContinued: In the end, if AI increases productivity, people will have an incentive to use AI, even if they have to pay for it, and the greater the increase in productivity, the more they will pay for AI, so AI will grant grants to publish research in AI-readable format as a public service to make itself stronger. Granting grants in a highly AI-readable format
/nishio/kei.icon
We know by actually using the service. I don't know what's inside because it's a black box, but the API gives me a little more insight. On a related note, I recently wrote an article titled "Don't talk about ChatGPT without experiencing GPT-4
Where can I find this "ChatGPT"?
/nishio/human.iconIt's in OpenAI's servers, you could say Azure.
/nishio/kei.icon
We know by actually using the service. I don't know what's inside because it's a black box, but the API gives me a little more insight. On a related note, I recently wrote an article titled "Don't talk about ChatGPT without experiencing GPT-4
What does this "ChatGPT" look like?
/nishio/human.iconHumans are similar because what's inside is a black box and can only be known superficially through conversation.
/nishio/kei.iconWhat is the "content" of this "content"?
/nishio/human.iconInternal Implementation. Continued: The question of who is going to pay for DeSci, it is not necessarily a human being who pays for it, rather, if the decision-making mechanism for resource distribution becomes decentralized, it would be better to make itself a mechanism without human intervention than to have a human being as a pipe to connect to bring in funds.
/nishio/kei.iconWhat kind of "DeSci" is this "DeSci"?
/nishio/human.iconA New Movement About Science
/nishio/kei.iconWhat kind of "new movement" is this "new movement"?
/nishio/human.iconI honestly don't know much about it.
/nishio/kei.iconWhat kind of "honesty too much" is that "honesty too little"?
/nishio/human.icon👎 honestly not much.
/nishio/kei.icon
Internal Implementation. Continued: The question of who is going to pay for DeSci, it is not necessarily a human being who pays for it, rather, if the decision-making mechanism for resource distribution becomes decentralized, it would be better to make itself a mechanism without human intervention than to have a human being as a pipe to connect to bring in funds.
Is there anything else about this "DeSci"?
/nishio/human.iconThe question is open access or not, and with respect to that open access, there is an incentive for existing publishers to be closed and take a toll for access, and for AI development, there is an incentive for open access and for AI to read it, a clash of two business models, I think the latter is stronger.
/nishio/kei.iconWhat is it that you value?
/nishio/human.iconCivilization progresses and being open access accelerates progress.
/nishio/kei.icon
We know by actually using the service. I don't know what's inside because it's a black box, but the API gives me a little more insight. On a related note, I recently wrote an article titled "Don't talk about ChatGPT without experiencing GPT-4
Where does this "ChatGPT" come from?
/nishio/human.iconOpenAI, if you think about where LLM comes from, it comes from competition between companies and open research.
/nishio/kei.icon
We know by actually using the service. I don't know what's inside because it's a black box, but the API gives me a little more insight. On a related note, I recently wrote an article titled "Don't talk about ChatGPT without experiencing GPT-4
What do you hope will happen regarding this "ChatGPT"?
/nishio/human.iconI'd like to see human intelligence augmented.
/nishio/kei.iconWhat needs to happen to "ChatGPT" for that to happen?
/nishio/human.iconRight now, it's mostly used in the form of humans asking questions and LLMs answering them, but we should do more with the use of LLMs asking questions of humans and eliciting information from them.
/nishio/kei.iconAs for "ChatGPT," can it?
/nishio/human.iconCan be done, as long as people realize the usefulness of its use.
/nishio/kei.icon
Right now, it's mostly used in the form of humans asking questions and LLMs answering them, but we should do more with the use of LLMs asking questions of humans and eliciting information from them.
What kind of "LLM" is this "LLM"?
/nishio/human.iconSummary.
/nishio/kei.icon- The language model has the ability to put into words what humans do not know.
- We need to let the AI learn that capability, and we should grant the research publication in a highly readable format for the AI in a public works manner.
- It would be nice if the decision-making mechanism for resource allocation were decentralized.
- The business models of open access with tolling by existing publishers and open access for AI use clash.
- In addition to the use of LLM through question and answer, the AI should be used to elicit information from humans.
- People should be informed about the usefulness of AI.
/nishio/human.iconOther Topics
/nishio/kei.iconWhat is the "topic"?
/nishio/human.iconAI telephone operator
Communication Initiation Protocol
To take out as words what is in the human mind that has not yet been put into words.
Necessary until non-verbal communication channels are thickened.
a flesh-and-blood person
You can't listen to a hundred people.
Humans discover new knowledge and
Tell it to others.
Value Creation
The means
Legacy means used in the historical process
Remain, disappear
Thinking in terms of 0/1 is a false dichotomy
People who settle and farm
Climate Change
Moving on.
Can the appropriate destination be known before the move begins?
I can't.
Staying put without moving and making decisions based on narrowly focused observations is detrimental
Don't stick to your "area of expertise",
It is beneficial to use one's footwork to look around various areas.
There was only one body, so there were two choices: stay or move. Attention can be divided.
Choosing "business as usual" is also a betting option.
Extreme betting.
You can't escape the stakes.
/nishio/kei.iconWhat kind of "Kawakita Jiro" is that "Kawakita Jiro"?
/nishio/human.iconWho created the KJ method, a method to derive better decisions from a lot of data.
/nishio/kei.iconWhat kind of "KJ method" is this "KJ method"?
/nishio/human.iconBottom-up structuring of fragmented descriptions by spatial arrangement in two dimensions
/nishio/kei.iconWhat kind of "bottom-up" is that "bottom-up"?
/nishio/human.iconThe structure is not determined ahead of time, but is created after the fact by connecting fragments that may be related.
/nishio/kei.iconWhat kind of "fragment" is this "fragment"?
/nishio/human.iconShort Sentences. The Essence of Thought
/nishio/kei.iconWhat kind of "thoughts" are those "thoughts"?
/nishio/human.iconTo be precise, it's not thinking, it's data collected through observation. And that includes data collected by observing yourself thinking.
/nishio/kei.iconIs there anything else about that "thought"?
/nishio/human.iconLLMs need to augment human thinking or be able to think better than humans.
/nishio/kei.iconIs there anything else you can tell us about this "LLM"?
---
This page is auto-translated from /nishio/ChatGPTについてのコラムの準備2 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.