Q: Hi Jon, you are definately onto the footsteps of Socrates here :) Really good concept đź‘Ť
Two questions please: 1) Sorry for sounding dumb but even after reading the Q&As here and reading about Deep Dive on your website, I still can't comprehend fully what it actually does. I do get it that it reads documents step by step so no section is missed. For my benefit and any others who didn't quite grasp it, please can you elaborate this feature, especially in terms of quality of output and use cases? Is it good for summarising or simply asking questions? Would DeepDive generate/repurpose better content from the context compared to simple document chat? 2) Once a Deep Dive is done, it stays in Socrates' memory forever? i.e. it has "learned" this knowledge, so we can always come back to it and ask questions? Thanks Jay
A: Thanks for the question. 1) When you have a question about a complex document, there are usually multiple categories of questions. 2 main categories are: 1) specific questions, and 2) broad questions. Normal chats excel at specific questions, like, 'What recipes include bananas?'. Deep Dive chats excel at broad questions, like, 'Summarize this' or 'Look through the entire document to find...
Aha! thanks Jon, finally understood it! Deep Dive is siloed to the chat but can we keep adding more documents to the same chat? So let's say I had a research paper that I did a deep dive in a chat but then found another relevant paper. Can I add to the same chat for deep dive and then ask questions to BOTH papers so that Socrates picks the best answer from both papers?
So am I right to say that unless I use Deep Dive enabled, the results would be really bad? Because that's what it seems from the Hamlet. The summary was so bad - much worse than many other LLMs I've tried - that it's not worth even using it (without Deep Dive).
If your queries are more specific like "Who is Hamlet?", the results actually are better with normal chat than Deep Dive. That's why Socrates chooses the 'memory' that will perform better if Deep Dive is enabled.
Q: How do you do this?
"This tool provides excellent output. The way it summarizes an entire document thoroughly from beginning to end, without losing steam towards the end of even a 300 page document, is impressive. Very few tools can do this, and the one or two I tried did not do nearly as good a job (they would still skip large sections). Based on the results and the founder, I think this tool has a bright future, and is well worth it."
Someone wrote this review above. Could you tell me how this is done and why your method works while others don't? Thanks.
A: See my response to JayTheRainmaker, but to quickly summarize, we repurpose the content to a proprietary format (and don't use the same techniques everyone else uses) to create a second 'memory' from which Socrates can access. Socrates will choose the best 'memory' to use depending on the question asked.
A: Hello, it just means "Rows" in the table. In Tier 4, each table (1 per chat) can have 100 rows (i.e., documents) to run prompts on. In other words, you can easily compare up to 100 docs at one time.
Q: How many collums for table AI are supported?
Jon_Socrates
Oct 2, 2024A: Unlimited
Q: Hi Jon, you are definately onto the footsteps of Socrates here :) Really good concept đź‘Ť
Two questions please:
1) Sorry for sounding dumb but even after reading the Q&As here and reading about Deep Dive on your website, I still can't comprehend fully what it actually does. I do get it that it reads documents step by step so no section is missed. For my benefit and any others who didn't quite grasp it, please can you elaborate this feature, especially in terms of quality of output and use cases? Is it good for summarising or simply asking questions? Would DeepDive generate/repurpose better content from the context compared to simple document chat?
2) Once a Deep Dive is done, it stays in Socrates' memory forever? i.e. it has "learned" this knowledge, so we can always come back to it and ask questions?
Thanks
Jay
Jon_Socrates
Oct 2, 2024A: Thanks for the question.
1)
When you have a question about a complex document, there are usually multiple categories of questions. 2 main categories are: 1) specific questions, and 2) broad questions. Normal chats excel at specific questions, like, 'What recipes include bananas?'. Deep Dive chats excel at broad questions, like, 'Summarize this' or 'Look through the entire document to find...
Verified purchaser
Aha! thanks Jon, finally understood it! Deep Dive is siloed to the chat but can we keep adding more documents to the same chat? So let's say I had a research paper that I did a deep dive in a chat but then found another relevant paper. Can I add to the same chat for deep dive and then ask questions to BOTH papers so that Socrates picks the best answer from both papers?
Verified purchaser
We don't currently support this, but this workflow sounds rather useful, so we are prioritizing this feature on our roadmap!
Q: Why this results?
I went to the demo and viewed Hamlet.pdf
I asked it to give a detailed summary of every act and scene.
It skipped a lot of scenes - most of them actually.
Why are the results like that?
Google AI Studio could do a better job.
Jon_Socrates
Oct 2, 2024A: Hello,
I believe the Hamlet demo doesn’t have deep dive enabled. You’ll get a much deeper summary with Deep Dive
So am I right to say that unless I use Deep Dive enabled, the results would be really bad? Because that's what it seems from the Hamlet. The summary was so bad - much worse than many other LLMs I've tried - that it's not worth even using it (without Deep Dive).
Verified purchaser
If your queries are more specific like "Who is Hamlet?", the results actually are better with normal chat than Deep Dive. That's why Socrates chooses the 'memory' that will perform better if Deep Dive is enabled.
Q: How do you do this?
"This tool provides excellent output. The way it summarizes an entire document thoroughly from beginning to end, without losing steam towards the end of even a 300 page document, is impressive. Very few tools can do this, and the one or two I tried did not do nearly as good a job (they would still skip large sections).
Based on the results and the founder, I think this tool has a bright future, and is well worth it."
Someone wrote this review above. Could you tell me how this is done and why your method works while others don't? Thanks.
Jon_Socrates
Oct 2, 2024A: See my response to JayTheRainmaker, but to quickly summarize, we repurpose the content to a proprietary format (and don't use the same techniques everyone else uses) to create a second 'memory' from which Socrates can access. Socrates will choose the best 'memory' to use depending on the question asked.
Q: What does "documents for table AI" mean?
Jon_Socrates
Oct 2, 2024A: Hello, it just means "Rows" in the table. In Tier 4, each table (1 per chat) can have 100 rows (i.e., documents) to run prompts on. In other words, you can easily compare up to 100 docs at one time.