ChatGPT, Startups, and Pivots

I write monthly updates about our startup journey to collect my own thoughts, hopefully to inspire you to chase your dream too, and of course in the spirit of #buildinginpublic; follow my newsletter and get notified next month!

In this update you will read:

• ChatGPT and how we pivoted our product
• Our drastic UX redesign
• Rebuilding our backend for performance and scale
• LLMs for clinical knowledge and other "journal club" topics we covered
• Data structure for knowledge sharing, and other "deep random talk" episodes
• Ultradian cycles and other "podcast episodes" I enjoyed

Asks

1. If you are starting a new ML project and want to get a headstart in designing your pipeline, the private version of our product can help. Reach out for details.

2. If you know ML / DS managers, Data product people, or ML / DS Consultants, I would love an intro to get their feedback on our product

Previously at Aggregate Intellect...

In?December '22?...

• We wrapped up our community round
• We did our second CDL session
• We, once again, discovered how important stories are for startups
• We narrowed down further on our commercial use case
• We started to think about what ChatGPT meant for our product

See also:??September '22 | October '22

Here is what happened in January 2023:

So, ChatGPT reached 100M monthly active users in Jan - Reuters, Feb. 1 2023

Say what!?

Like anything else that goes explosively viral, this has created significant conversation around AI systems that approach human level language capabilities. Note that I'm not calling it cognition or intelligence because I do believe to qualify as those our smart systems need to have a much better level of reasoning that they do have now. See my rant here.

Philosophical discussion aside, this kind of virality poses a significant stress on existing systems. I'm not even talking about how established businesses as well as startups will be impacted by products like this, as people have been speculating; I think the much shorter term consequence of this will be a radical shift in the way users expect to interact with products, especially knowledge-based ones.

We have to wait and see what the new "normal" will be like in terms of user experience going forward (and we are recording at least one episode of DRT talking about that). We, however, instead of waiting to see what will happen, decided to define the future that we are stepping in, at least as far as knowledge-intensive products like ours are involved.

We have to scrap whatever we've done on the UX and go back to the drawing board and come up with the simplest, and most intuitive user experience we can - me, talking to my product team hollering them in out of their new year break

The team moved really fast and reimagined the UX, everything that needed to change on the backend to enable this, and we started running experiments with a completely re-imagined system. [scroll up to the "asks" if you want to have a sneak preview]

I am very excited about the pivots we made this month and some of the creative ideas we are using to quickly validate details of the new direction.

Unsung Hero

In this section I say thanks to someone who is not directly involved in every day operations of our company but has significantly helped us:

For this month it has to be Suhas Pai ! He always patiently helps me think through a lot of nuanced NLP questions. Thank you, Suhas! you rock!

What were the highlights of the month for the team?

Ammar: I am proud of my work on the prototype re-design, in a way that reduced our time to value by 10x

Sara: My proudest contribution this month was re-designing the library multiple times and iterating on it and prototyping it really quick to get users feedback

Percy: My proudest contribution this month is designing the architecture for our new system involving chatbots, LLMs, traditional information retrieval systems, and knowledge graphs.

Samuel: My proudest contribution this month is our quick collaboration and iteration on the new backend to improve our modularization, scalability and performance.

Bereket: My proudest contribution this month is modeling and implementing our various databases and APIs.

Eyob: My proudest contribution this month has to be working on the new backend system with new technology that I haven’t worked on before. This will make our system cleaner and ready for future use cases.

Maryam: I'm really proud of how we came up with a new way to validate market traction by creating a scaled down MVP. We can now provide a use-case specific technical insights report, backed by expert validated resources, to help ML developers build their pipelines faster! The response has been very interesting as we roll out this limited service for feedback. (CTA: Is your ML team starting a new project soon? reach out to ?? Amir Feizpour , Ammar Khan or Maryam Farooq )

Journal Club

We run two series of weekly sessions to educate ourselves about topics that are important to our success.

Public ML Weekly Series

You can join this series every Friday at noon ET [add to your calendar]

We recently discussed the following papers:

• Solution Patterns for Machine Learning
• DEMIX Layers: Disentangling Domains for Modular Language Modeling
• Prompting Is Programming: A Query Language For Large Language Models?
• Large Language Models Encode Clinical Knowledge
• A Review on Language Models as Knowledge Bases
• Demonstrate-Search-Predict: Composing retrieval and language models for knowledge-intensive NLP
• [next week] ReAct: Synergizing Reasoning and Acting in Language Models

Internal Product Dev Weekly Series

(this is not a public series but if you are really interested in these topics message me and we will work something out)

We recently discussed the following topics:

• The biggest bottleneck for large language model startups is UX
• Product and Business Aspects of Collective Intelligence
• Sharing Knowledge And Expertise
• Effect of DevOps and ModelOps on Product Development

Deep Random Talks

DRT is our YouTube show where we have deep but friendly conversations on various topics. The episodes have timestamps and summarized notes in their descriptions.

• [Startup Fundraising]?Unfolding the Maze of Funding Deep Tech; Metafold - DRT S2E14
• [Knowledge Management]?Data Structure for Knowledge = Language Models + Structured Data - DRT S2E13
• [NLP]?EVE - Explainable Vector Embeddings - DRT S2E12
• [DeSci]?LabDAO - Decentralized Marketplace for Research in Life Sciences - DRT S2E11
• [Product Dev]?Data-Driven Behavior Change and Personalization - DRT S2E10

Podcast Episodes I really Liked

• Leveraging Ultradian Cycles, How to Protect Your Brain
• Dr. Sam Harris: Using Meditation to Focus, View Consciousness & Expand Your Mind
• Dr. Eddie Chang: The Science of Learning & Speaking Languages
• Machine Learning for Thinking
• Exercise, Heat, Cold & Other Stressors for Longevity

Parting Thoughts...

This past month was a remarkable coincidence of disassembling our approach to our solution, and reconstructing it in a way that is way better than before. Not only has this drastic change has positioned us to have a much better UX, but it also has opened up our hands to do concrete experiments with our target customers and get some real world battle feedback. This has in turn changed our approach and momentum around sales and now we have a much healthier pipeline and process. All this is not accidental though; we have been working hard to establish systems around rapid experimentation and rigorous and candid decision making processes. We have also created intentional spaces where we teach each other all sorts of topics relevant to our work which creates a fertile ground to find solutions to our complex product and business decisions.

Other Resources

Get Started with AI.SCIENCE

Follow?this RECIPE?to know more about what you can do on our platform

Want to talk about any of the above? startups? building products? or about your career in ML?

I do (free) 20 minute calls with people to talk about everything?ML?or?startup; if you know anyone who is trying to get into ML / DS, or figuring out how to build an AI product, or thinking about getting into startup life, AND wants to chat about it with me, please pass this link to them:?https://calendly.com/amirfzpr/20min-ama