OpenAI Deep Research
https://gyazo.com/82357e75c92856d99e94097400c4f619
ukkaripon I just played around with Deep Research and o1 pro. nishio a meeting to try out Deep Research. If you use it well, you can dig deep and make discoveries. If used poorly, it is simply a flat collection. If you pick up commodities in your initial search, you get into commodities. Either make them think deeply first, or collect examples with questions that don't get into commodities, and then stand back and think deeply. relevance
You should be aware of whether you want to collect them flat, fill in the gaps, or pile them up.
nishio I saw a lifehack article that said, "Don't give detailed instructions to o3," but that "giving detailed instructions that are of lower quality than what they think will reduce overall productivity. But it's just "Don't give detailed instructions at all, just throw them out! I wonder why it is so extreme...(or does the writer of the article not have enough understanding?) What you get from throwing a whole lot is a commodity because the people who throw a whole lot get the same as well.
Someone said it's good to have them report in a dialectics way. That's one piling up way to do it. A case study was also shared related to "picking up commodities in your first search will get you hooked on commodities" and "reading an AI-generated article that is halucinating and taking it to heart."
I guess I should clarify whether I'm asking you to brainstorm various articles and report miscellaneous information, or whether I'm asking you to land on a reliable source.
I had ChatGPT's Deep Research put together a "Tips for Getting Adopted" and they had a pretty good story to share!
nishio No really, I think it's amazing that you can come up with this level of report just by throwing the right prompts and waiting 10 minutes. I think there are more people who could spend a whole day and still not be able to write this report. It's well organized from a mentor's point of view. nishio And an AI that reads these Tips will write a proposal at this level, and an AI that reads that proposal will give an existing case study at this level. ImAI_Eruel By the way, I have a comment about DeepSearch from OpenAI, which is really quite amazing. When I asked him to research a field of study that he knew almost everything he could gather on the internet to be completely correct, he came up with information that I once had quite a bit of trouble finding, and then he found a story that I didn't know either.
knshtyk Ah Deep Research, this is a higher level version of search. It performs the steps of searching, gathering information, and building knowledge systematically at hand or in your mind, instead of humans, and generates a summary from the information. The information search capability is higher than web search, so you can skip the time-consuming search and selection process... kajikent I've been using OpenAI's "Deep Research" and I've found that while Google's "Deep Research" function just goes through a large amount of resources and summarizes them in a way that covers the entirety of the information, OpenAI's function goes a step further and aggregates and reports information with its own opinions and hypotheses. OpenAI's feature goes one step further and aggregates information with its own opinions and hypotheses and reports them to us. Conclusion, super useful for primary reporting for all operations.
2/4
Agentic RAGs like Deep Research are extremely useful in cases where there is already information in the public domain that can be found by search engines, and not being able to find it is the bottleneck.
Next comes the phase of using this for non-public internal information, your own library, etc.
nishio When Deep Research can search non-open access papers and even PowerPoint documents in internal groupware ...the dawn of Singularity is finally dawning! https://gyazo.com/e1ba46951ecd4babdb0797f67bf5d99e
Deep Research can go from noticing a gap in the context of the KJ method, which is a type A to type B conversion, to finding something to fill the gap.
---
This page is auto-translated from /nishio/OpenAI Deep Research 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.