Yes, a brilliant article Kevin and some pithy questions to get better quality from AI. It would be great to develop a library of questions…and I am more than happy to contribute.
One suggestion: Analysts tend to specialize, so in the priming phase I adopt the following:
You are an experienced analyst specializing in the <insert industry>.
And there are questions where the requester might state ‘and I am an experienced investor requiring a high degree of confirmation as to fact based data’
Two comments of mine (I use free perplexity, but also notebooklm and Claude, all free atm)
1) the spaces feature is available also on free
2) I like the idea of the hedge fund short prompt. I wonder if an AI analyst battle would be fun: 2 rooms, both get the files you uploaded and get the appropriate printing. Then, e.g. create an investment thesis for a company that is then provided to the other to put holes into... The reason for suggesting two rooms is that I wonder if the creation of e.g. a long-sided thesis will act as priming for the short-side prompt that follows... Idk
Hm maybe I explained it not well enough: a room to set the basic priming, then create a first thread in that room creating a thorough long oriented research report. Then, and that's what I meant with contest, feeding the long report into a second thread in the same room, i.e. it has the same priming, to criticize and act as a short.
Alternatively, one could think of three threads in the same room: one long one short, and a third acting as judge/referee/portfolio manager making a final investment decision...
Your approach to using Perplexity for stock research is fascinating, especially the emphasis on proper priming and context setting. I've been experimenting with AI tools for investment research, and your structured methodology makes a lot of sense.
I'm curious about how you handle the potential hallucinations in AI-generated reports. Do you have any specific verification processes you use to cross-check the information?
The hallucinations usually appear later on, when you try to go deep and more detailed. You can use another tool to ask for factchecking. But even that is not 100%.
hi, Perplexity and other deep research tools can now be mounted in any home made applications through the remarkable Python package GPT-researcher. Local and constrained research (list of url to use as sources) modes are also available. This is what I include on my own personal research platform Chat with Fundamentals. To mitigate hallucinations risk, I prefer to maintain separated a pure data interpretation pipeline powered by reliable financial data provider. Future of analysis in my view lies on the knowledge of what is in the blackbox. Best
I am reaching out to people I think will be interested in knowing not just the latest but future trajectory of Nvidia and the semiconductor industry. If that's you SUBSCRIBE and you get a weekly bulletin and monthly State of Nvidia and Industry Magazine. That's all.
Really helpful article, thank you! Out of interest have you tried the prompts on other platforms? I enjoy using Grok for customer sentiment etc from X and I seem to be getting longer form answers from Grok and Chat GPT so interested in your thoughts
I have used Grok from time to time, and it seems to offer great results. Grok has the advantage of being able to use the posts on X, but I haven't made a 1-1 comparison using the same prompts. I'll see if I can whip something up in the future.
Perplexity is one of the first tabs opened on my browser. It's my little force multiplier and it has a knack for helping me frame problem sets, E.g. Uncovering rocks i didnt even know were there. I hope Apple does not buy it outright. They'll ruin it and steal my soul...😮💨
Your approach to competitive analysis using Perplexity is quite thorough. I've been using similar tools for market research, and your method of explicitly asking for opposing views is particularly valuable.
I'm interested in how you balance the AI-generated insights with traditional research methods. Do you find that using Perplexity has significantly reduced the time you spend on initial research?
Yes, of.course, after the initial phase, you still need to do the heavy lifting. Perplexity cannot help build conviction. But it helps going faster through the funnel....
The example of analyzing Cipher Pharmaceuticals is very insightful. I particularly appreciate how you break down the research process into clear steps, starting with understanding the basic business model.
I wonder about the effectiveness of the "explain like I'm 8" approach for complex business models. Have you found this method works well for all types of companies, or are there certain industries where it's less effective?
Great article! On the prompt, if you’re aiming for research summaries that approach human-level quality, I highly recommend trying prompt chaining. Also, using structured output methods can help ensure consistency and clarity in the results.
Yes, a brilliant article Kevin and some pithy questions to get better quality from AI. It would be great to develop a library of questions…and I am more than happy to contribute.
One suggestion: Analysts tend to specialize, so in the priming phase I adopt the following:
You are an experienced analyst specializing in the <insert industry>.
And there are questions where the requester might state ‘and I am an experienced investor requiring a high degree of confirmation as to fact based data’
That's a great idea! Let me figure out a way to collaborate on this, maybe with a voting mechanism to test the quality of the prompts.
All I want is to share Nvidia love with the world and maybe get a sub or two. Wanna be #1? Subscribe or say 'YES' to join the ride.
Hi, great tool, good article 👍😃
Two comments of mine (I use free perplexity, but also notebooklm and Claude, all free atm)
1) the spaces feature is available also on free
2) I like the idea of the hedge fund short prompt. I wonder if an AI analyst battle would be fun: 2 rooms, both get the files you uploaded and get the appropriate printing. Then, e.g. create an investment thesis for a company that is then provided to the other to put holes into... The reason for suggesting two rooms is that I wonder if the creation of e.g. a long-sided thesis will act as priming for the short-side prompt that follows... Idk
Usually threads and spaces cannot interact I believe. But somehow, we should be able tontest this right?
Hm maybe I explained it not well enough: a room to set the basic priming, then create a first thread in that room creating a thorough long oriented research report. Then, and that's what I meant with contest, feeding the long report into a second thread in the same room, i.e. it has the same priming, to criticize and act as a short.
Alternatively, one could think of three threads in the same room: one long one short, and a third acting as judge/referee/portfolio manager making a final investment decision...
Idk just a fun idea on a friday
No. It was me. I didn't get it 😀. We should test it!
Great article. If anybody wants to learn more about how LLMs work, Andrej Karpathy does a very good job of explaining it.
https://www.youtube.com/watch?v=7xTGNNLPyMI
If there was a GPU for heartbreak, I’d need one. 30 days, no fans. Help fix this subscribe or say ‘YES’ and I’ll bring you the goods.
Your approach to using Perplexity for stock research is fascinating, especially the emphasis on proper priming and context setting. I've been experimenting with AI tools for investment research, and your structured methodology makes a lot of sense.
I'm curious about how you handle the potential hallucinations in AI-generated reports. Do you have any specific verification processes you use to cross-check the information?
The hallucinations usually appear later on, when you try to go deep and more detailed. You can use another tool to ask for factchecking. But even that is not 100%.
You scroll, I post. Let's turn this moment into something beautiful. Subscribe or message 'YES' and become part of the Nvidia crew!
hi, Perplexity and other deep research tools can now be mounted in any home made applications through the remarkable Python package GPT-researcher. Local and constrained research (list of url to use as sources) modes are also available. This is what I include on my own personal research platform Chat with Fundamentals. To mitigate hallucinations risk, I prefer to maintain separated a pure data interpretation pipeline powered by reliable financial data provider. Future of analysis in my view lies on the knowledge of what is in the blackbox. Best
That's the most important: know the limitations of each tool and the data you use.
This isn’t just a call to action it’s a tech love story waiting to happen. Subscribe or reply ‘YES’ and let’s script greatness.
A comparison side by side with another tool, e.g., Gemini, ChatGPT might be more persuasive.
I’m on it
I’m on day 30 of talking to ghosts about Nvidia. Wanna haunt this feed with me? Subscribe or say 'YES' for ghost-free updates!
Great tool bro...The Dr. House reference hooked me in... huge fan of the show!
Damn, now I feel I have to rewatch the show, for the third time😀
lol
It’s not begging. It’s passionate GPU evangelism. Subscribe or type 'YES' for daily chips, dips, and data.
Do you use Google maps? What
Of course. You mean like for a retailer, to analyze their stores?
No, I meant personally, when you look for a museum or how to get to somewhere.
Tired of boring feeds? I serve hot Nvidia insights daily. Subscribe now or just shoot a 'YES' for a no-strings-attached test run.
Is this a scam?
Y would you think that?
As I don't know how is this relevant here - not to be mean, I'm just careful
I am reaching out to people I think will be interested in knowing not just the latest but future trajectory of Nvidia and the semiconductor industry. If that's you SUBSCRIBE and you get a weekly bulletin and monthly State of Nvidia and Industry Magazine. That's all.
Really helpful article, thank you! Out of interest have you tried the prompts on other platforms? I enjoy using Grok for customer sentiment etc from X and I seem to be getting longer form answers from Grok and Chat GPT so interested in your thoughts
I have used Grok from time to time, and it seems to offer great results. Grok has the advantage of being able to use the posts on X, but I haven't made a 1-1 comparison using the same prompts. I'll see if I can whip something up in the future.
They appear to be working adequately on CoPilot.
Great article! Helps a lot!
Thank you!
Posting Nvidia magic with zero followers feels like playing co-op alone. Be player 2 subscribe or hit me with a ‘YES’ and let’s win this.
I use Perplexity, however, my priming and questions are not as detailed as these so i will see the difference in being more prescriptive.
Perplexity is one of the first tabs opened on my browser. It's my little force multiplier and it has a knack for helping me frame problem sets, E.g. Uncovering rocks i didnt even know were there. I hope Apple does not buy it outright. They'll ruin it and steal my soul...😮💨
Your approach to competitive analysis using Perplexity is quite thorough. I've been using similar tools for market research, and your method of explicitly asking for opposing views is particularly valuable.
I'm interested in how you balance the AI-generated insights with traditional research methods. Do you find that using Perplexity has significantly reduced the time you spend on initial research?
Yes, of.course, after the initial phase, you still need to do the heavy lifting. Perplexity cannot help build conviction. But it helps going faster through the funnel....
The example of analyzing Cipher Pharmaceuticals is very insightful. I particularly appreciate how you break down the research process into clear steps, starting with understanding the basic business model.
I wonder about the effectiveness of the "explain like I'm 8" approach for complex business models. Have you found this method works well for all types of companies, or are there certain industries where it's less effective?
It works for most. But sometimes it oversimplifies things
Great article! On the prompt, if you’re aiming for research summaries that approach human-level quality, I highly recommend trying prompt chaining. Also, using structured output methods can help ensure consistency and clarity in the results.
Thanks. I will try that!
You should also try "Deep Research" in Gemini. It's amazing - and free of charge.
Thanks, will compare the results with Perplexity