Vector search is an opportunity to cut out
__BELOW_IS_AI_GENERATED__
ベクトル検索は切り出しの機会になる 2023-10-06 14:15 omni.icon
note
Vector searches are an opportunity for new discoveries.
digest
Vector searches provide new perspectives.
fragments
pVectorSearch
UI for humans is not interesting, question answering has evolved toward "don't answer what you don't know" and lost its fun factor, create an API and return relevant pages and images by vector search as a tool for future self to use. Returning images is particularly interesting. For example, requesting "false dichotomy" returns the URL of a related image. Also, instead of randomly, a vector search can be performed on the current sentence being written, and a blind spot card can be drawn. This is possible because Scrapbox stores pairs of "images" and "sentences" in the "page" format. pVectorSearch2023-04-29~05-31
The human-oriented UI is not interesting, and question answering has evolved toward "don't answer what you don't know"; create an API and return relevant pages and images via vector search as a tool for future self use. Returning images is particularly interesting. For example, requesting "false dichotomy" returns the URL of a related image. Also, instead of randomly, a vector search can be performed on the current sentence being written, and a blind spot card can be drawn. This is possible because Scrapbox stores pairs of "images" and "sentences" in the "page" format. Transverse Vector Search Experimental Memo 2023-09-20.
Experiments were conducted to increase the amount of data to be searched for red link extensions by AI. The search targets include other people's Scrapbox data, books, and papers. Since there is a possibility that some problems may arise in releasing the generated results as they are without review, the output destination was a private project that is closed to the public.
Vector search in Nishio.
As an example of a cross search, a search for "computer assisted deliberation and consensus building" yields specific results. The "share" link in the search results will open a page of such results.
next study group.
Vector search is good for ambiguous searches in natural language conversations, but it has a high cognitive load for users because it also hits "different things that look very similar" in "the name of the company I visited for sales" or "the product model number".
Examples where vector search was useful.
When the topic of "social security costs" came up, I recalled that I had talked about the high cost of social security in a past lecture, but could not find anything by searching for "social security costs." However, a vector search for "social security payments scientific research" yielded a hit. I also searched for "hole in the wall" to talk about a hole in the wall, but could not find it, so I did a vector search for "hole in the wall, you can't see it unless you get close" and found it.
digest
From the fragments of "pVectorSearch," "pVectorSearch 2023-04-29~05-31," "Cross-Vector Search Experimental Memo 2023-09-20," "Nishio's Vector Search," "next study group," and "Examples where vector search was useful," vector search has lost its human-oriented UI lost its fun, and question answering evolved in the direction of "don't answer what you don't know". However, it is possible to create an API and return the corresponding page or image by vector search as a tool for future self use. Returning images is particularly interesting. It is also possible to perform a vector search on the current sentence being written and pull a blind spot card, rather than randomly. This is possible because of Scrapbox's "page" format, which stores pairs of "images" and "sentences. However, although vector search is good for ambiguous searches of natural language conversations, it has a high cognitive load for users because it also hits "different things that look very similar" in "the name of the company you visited for sales" or "the product model number. thought
Vector search provides a new perspective and is a useful tool for the future self to use. However, the user's cognitive load must be considered.
title
Vector Search: Providing New Perspectives and Considering Cognitive Load."
extra info
titles: ["pVectorSearch", "pVectorSearch2023-04-29~05-31", "Transversal Vector Search Experimental Memo 2023-09-20", "Vector Search in Nishio", "next study group", "case studies where vector search was useful"]
code:fragments
### pVectorSearch
pVectorSearch
2023-04-29
Consider this Scrapbox [vector search
Using a UI for humans to do it is not very interesting.
Loading prompts and answering questions has become less and less interesting as people have been trained to "don't answer what you don't know" more and more as a halcyonation measure.
First, create an API
As a tool for my future self to use.
Vector search and return the corresponding page
Cross-search Scrapboxes, blogs, etc. other than your own
Vector search and return a picture
This looks interesting.
For example, loading a request with "false dichotomy" will return https://gyazo.com/80d45ec33ca8cbb138108d71ad7eec02 in the response.
誤った二分法.icon
blind spot card can be drawn in a vector search in the sentence I'm writing, rather than randomly. This is possible because of the accumulation of "image" and "text" pairs in the form of "pages" in Scrapbox
2023-05-04
Examples of vector searching and looking at images
2023-05-11
2023/5/16
A system that recognizes speech when you are speaking and automatically queries it to search this Scrapbox and display matches.
### pVectorSearch2023-04-29~05-31
pVectorSearch2023-04-29~05-31
pVectorSearch2023-04-29~05-31
2023-04-29
Consider this Scrapbox [vector search
Using a UI for humans to do it is not very interesting.
Loading prompts and answering questions has become less and less interesting as people have been trained to "don't answer what you don't know" more and more as a halcyonation measure.
First, create an API
As a tool for my future self to use.
Vector search and return the corresponding page
Cross-search Scrapboxes, blogs, etc. other than your own
Vector search and return a picture
This looks interesting.
For example, loading a request with "false dichotomy" will return https://gyazo.com/80d45ec33ca8cbb138108d71ad7eec02 in the response.
誤った二分法.icon
blind spot card can be drawn in a vector search in the sentence I'm writing, rather than randomly. This is possible because of the accumulation of "image" and "text" pairs in the form of "pages" in Scrapbox
2023-05-04
Examples of vector searching and looking at images
2023-05-11
### Cross-sectional vector search experiment memo 2023-09-20
Transverse Vector Search Experiment Memo 2023-09-20
We conducted an experiment to increase the number of data to be searched for in Extending the Red Link with AI, which we have been running publicly on this Scrapbox. Searches include other people's Scrapbox data, books and papers
relevance
Since there might be some problems with publishing the generated results as they are without review, the output destination is a private project that is not open to the public.
Subsequent updates changed the search hits to output directly to Scrapbox, making it clearly unpublishable.
https://gyazo.com/bb1a0bd6cb151fa1b6ffa2d7c498744c
The page is automatically generated by AI at the destination of the red link created like this
mounting
Load from local pickle
It takes about 5 seconds to read 100,000 records
About 7 seconds to perform a local vector search
It's slow when you think of it as a web app response, but with the recent "throw a keyword or page that comes to mind, work on something else, and look at it after a while" style, it's not a problem.
There's about 10-25 minutes between when you throw the query and when I come back to check the results (I don't measure it).
### Vector search for Nishio
Vector Search in Nishio
Example of Cross Search
For example, a search for "computer assisted deliberation and consensus building" result. You can open a page of search results like this one with the "share" link in the search results
Term of Service
The query string POSTed to this service can be viewed by administrators. We may also create a public view of search results in the future.
Because we prioritize new knowledge over compatibility in our development, we cannot guarantee that a feature that was available at one point in time will be available at another point in time.
History
Only those who have been given the password to gain privileges will be searched for more.
2023-06-07 Added /tkgshn /halsk /yuiseki to search
Now you can only do a cross search of the whole, the checkboxes for the search target are not operable.
### next study group
Vector searches are good for vague searches in natural language conversations because they produce a flurry of hits, but they also produce hits for "the name of the company you visited on a sales trip" or "the product model number" because they produce hits for "different products that look very similar," which places a high cognitive burden on the user.
The following is a summary of the information enteredgpt.icon
7/29: AI writes to Scrapbox, introducing the concept of Scrapbox Agents.
Early to mid-August: Omoikane study group, discussion on working with AI, writing research notes with AI, evolution of vector search, and topics on handling AI-generated pages.
Mid to end of August: topics related to updating and search, taking into account multi-head, page memory, and user cognitive load.
Beginning of September: Explore AI-user interaction, note management, and relevance between different content.
Mid-September: various topics related to multi-head thinking, the intellectual production techniques of engineers, and working with AI.
Late September: LLM vs. other models, discussion on human concepts, optimal use of Scrapbox, and feedback on non-public tools.
Overall, this period seems to focus on Scrapbox and AI integration, particularly the concepts of vector search and multi-heading, and AI thinking and interaction with the user.
### Examples of where vector searches have been useful
Cases where vector searches were useful
Case 1
It's all about "social security spending."
I recall a past lecture in which he said, "I've talked about the high cost of Social Security."
The focus was on the ratio of investment to science and technology.
I'd try to find it, but I can't find it by searching for "social security payments".
https://gyazo.com/34b6beae7ed4d7d5d95fd0d4d08aaff6
The expression was "38 trillion for medical care, 24 trillion for welfare and others, and 120 trillion in total."
There's also "Science and Technology" a little further down the road.
I made an excerpt page with "Social Security Expenses" in the title.
Case 2
I'm trying to come up with a story about a hole in the wall, and I'm searching for wall or hole, but I can't find it.
Vector search for "hole in the wall, you have to get close to see it."
Second candidate found.
https://gyazo.com/9c55e099867137986e3be8ead02a9fdb
generated: 2023-10-06 14:15
---
This page is auto-translated from /nishio/ベクトル検索は切り出しの機会になる 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.