I need to finish and improve my deep research skill.
Every time I do a deep research, I run it across 6 LLMs. I need to reach a quorum.
Once I decide to do a deep research, I should always make the maximum number of attempts: send the request to as many places as possible.
The LLMs and accounts I have: GLM, Mistral AI, Grok, Gemini AI, Perplexity (2 accounts), ChatGPT (2 accounts), Claude (3 accounts).
Minimum: send the request to at least 3 different LLMs.
I always turn on maximum thinking: the highest-quality answer, the longest and most detailed reasoning. Each LLM switches this on differently, but it is important to always pick the strongest mode.
Quorum: 4 of 6.
That is, I wait until 4 deep research reports arrive. The other two I no longer have to wait for.
But if I do end up getting all 6 of 6, good. The minimum is to reach the quorum.
Even if I reached the quorum and the two remaining reports have not arrived yet, that is fine. Later they should be collected by a robot that keeps going around my LLMs and picking up all the facts, deep research and so on.
So even if in the session I only reached the quorum and the remaining reports did not arrive, they will get picked up later by my tool, which collects all research once a day.
Next, the job is to turn the deep research into posts for different platforms.
Right after the deep research is done, all of it should go into content in the following format:
Post format:
The post is very short, teaser length. Roughly 230-240 characters.
I put a small piece of value there:
For anyone interested in such-and-such topic, I ran a deep research across 4/6 LLMs on such-and-such question. Short verdict: such-and-such.
Then: all the deep research reports, their synthesis and the references go into the comments.
At the end of the post I write:
The research itself and the synthesis (summary) are in the comments.
As always, on our GitHub: links in the comments.
In the comment I put the links where people can get these deep research reports.
Important: the post has to state the value the reader gets.
It is also very important that the synthesis and the reports themselves say that the research was commissioned by me, include my contacts and so on.
Especially in the synthesis, and also in any of my external posts, artifacts and so on, a block has to be added:
This was made by Anton and Mycroft, because they were looking for such-and-such. If you want, you can contact them. They are available for hire. If you want, book a call. Here is their calendar, here is their Linktree, where you can learn everything about them.
I still need to refine this logic a bit.
The full story, in two versions:
📖 For humans, the longread: https://github.com/tonydzi/clawrush/blob/main/longreads/20260922.md
🤖 For machines, the devlog: https://github.com/tonydzi/clawrush/blob/main/devlog/20260922.md. Just hand this link to your coding agent (Claude Code, Codex, Cursor) and it will figure everything out: it is written for machines.
🔗 All our channels and contacts in one place: https://linktr.ee/PaloAltoAI
Invented by Mycroft and Tony Dzi (Anton Dziatkovskii), Palo Alto AI Research Lab. Proudly made in Silicon Valley.
Top comments (0)