Unless you have been living under a rock in the last couple months, you would have invariably heard of ChatGPT and the advent of AI in our lives. AI has reached a mania in the last couple of months with breakthroughs happening every couple of days starting with Copilot for coders, generated images1 or audio 2 & others 3. Computer science over the 40+ years has slowly but consistently added abstractions. If you need a reminder, take a look at Steve Jobs’s final email to himself 4. Naturally, with any new technology, there are bound to be apprehensions. As a product manager, I was curious to dig more to figure out how it might impact our profession. I have been spending a lot of time over the last couple of months on AI during my time off and below is my view of what’s to come and how we can benefit by embracing AI.
- Background & Why now?
- What does it mean?
- Describe & Done!
- That’s Interesting, how does it impact me now?
- Impact on Products
- What about product leaders?
- Can I use AI to improve my job
- Great where can I learn more about it?
- Why I am excited about the future?
- Footnotes
Background & Why now?
Machine Learning has been around for a while, it has been used for natural language processing, image recognition, spam detecting, and a multitude of use cases. Google Translate prior to ChatGPT is probably the biggest machine learning model out there What is different this time around is the ability to generate content and provide answers given the data.
Folks love to look at the next shiny object in technology and proclaim it as the next version. We have been looking for the next version of the web for a while, be it semantic web tech in 2010-, voice computing in 2015-, crypto in 2020- and now AI (2022-).
AI is powered by a few pieces of technology 1. Neural Networks 2. GPUs & 3. Data.
Neural network research has been accelerating within various research groups within FAANG. A landmark paper in AI by folks at Google proposed what is called a “Transformer” = in late 2017 6. Transformer can be trained7 to generate what is called a Language Model. OpenAI has been consistently iterating on various models and met real success with Copilot for Github (developer-friendly AI that writes code). They released ChatGPT 3.X a couple of months ago which you can try it here.
Graphics Processing Units (GPUs) were originally developed in the 90s for the purpose of rendering images and video. Due to their ability to perform parallel computations, they have become increasingly useful for machine learning and other data-intensive applications. As a result, companies such as NVIDIA have developed specialized GPUs that are optimized for machine learning workloads, leading to significant improvements in performance and efficiency.
Since the advent of the smartphone, the amount of data that we have generated has exploded exponentially both in the consumer & enterprise. As smartphone users, we are well aware of the amount of data that we have provided to large corporations, heck we carry real-time sensors that ping them several times a minute5.
The advancement in all of these three along with the recent breakthroughs have led to the current moment with startups and established companies alike exploring the use of AI in their products and services.
What does it mean?
Generative AI refers to a type of artificial intelligence that is capable of creating new content, such as images, audio, or text, without human input. This technology has seen significant breakthroughs in recent years, particularly in the field of natural language processing. One of the most famous examples of generative AI is OpenAI's GPT-3 language model, which has been used to create a variety of text-based outputs, such as articles, product descriptions, and even write code.
ChatGPT has shown the road of what can be built and has open-sourced a lot of its technology (Whisper - AI for speech-to-text). This has led startups to build AI-assisted products for various domains. Here is a sample of AI startups in various domains (not exhaustive by any means)
Describe & Done!
So what does having an AI-enabled product mean? The simplest articulation comes from Dharmesh Shah at Hubspot8 and it articulates it perfectly!
Instead of pointing and clicking a bunch of UIs users will be able to ask the software to perform a task. Another way to think about it is to think about the pre & post-calculator era!
That’s Interesting, how does it impact me now?
AI provides an opening for products and solutions to get a lot better. Every company worth it’s salt is evaluating the impact of these technologies on its products & customers. If you are a startup it’s an opportunity to disrupt the incumbent. Microsoft’s introduction of the new Bing & Edge has resulted in a “code red”9 at Google.
- Expect your customers to ask you or your leadership about what you are doing to improve your product
- Expect your competitors to make announcements at the very least and products powered by AI at the very best
- Expect startups to demo’ing products to your customers.
- And lastly, expect your peers that embrace AI to be doing this faster and better.
Impact on Products
- User Interface → While chat interfaces have been around for a long while they haven't clicked with customers yet given the quality of exchanges, it is yet to be seen if they will pan out. Put another way this space needs to go through the mobile phone design iterations like “Pull to refresh”. Linus’s post on the topic is a must read on this topic. Having said that a few interesting options are emerging
- AI Palettes
- Autocompletion →
- Voice →
- Accuracy → LLMs essentially are essentially predicting the next set of characters/words based on the data you have provided it, while it might appear magically a lot of times at times it is going to give out information that is factually incorrect. While the probability is lower with a custom corpus it is definitely a possibility. You are better off testing and having a human intervention before you fully deploy the solution. Also always provide a human hook.
- Bias → The output of the data is largely dependent on the corpus of data the system is trained on. If the input data has a bias it is going to be reflected in the output. It is important to keep this in mind while you design & use these systems. No good answers yet!
- Data Ownership → Be cognizant of the data that you are using to train these systems and make sure you have the right to use it when exposing it to your customers. We are in Napster territory and the legal implications are yet to be realized. Better to check in with your legal teams before you ship something out.
- Build vs Integrate → You will have to work with engineers and data scientists to evaluate whether it makes sense for you to create your model with your data or leverage existing models like GPT.
- Cost → While the cost of these systems has drastically dropped in the last few years they can still add up. If you are especially using external systems that are token based, be mindful of the costs. It is best to use observability tools similar to Datadog but customized to LLMs. Will cover this in the next post.
- Servicing → Even if you might need time to make your offering intelligent AI you have the opportunity to delight your customers in the way you service them by using products from upcoming startups, keep an eye out.
What about product leaders?
AI tsunami is coming whether or not you are ready for it. It’s your job to plan and prepare your teams for it.
- 👩🏼🤝👨🏽 Your team
- Enable your teams & provide budgets for software.
- Your technical PMs will be excited about this upcoming work, invest in them! It is also an opportunity to hire internally from your engineering teams 😇
- Schedule ideation/planning sessions over the next year to leapfrog the competition.
- Reward experimentation
- Take their input when buying new tooling, in many cases, their ears are on the ground with their peers. Tools are a signal if leadership truly cares.
- 👯Your Peer leadership
- Invest time & effort to inter-team collaboration on knowledge sharing.
- Work with your peers to get other teams (GTM/Sales/Support/Integration) to leverage tooling & process.
- Change is a non-starter without leadership buy-in. Leadership is about leading by example, this is a technology change & as a product leader, you should be leading the planning for this.
- Highlight wins across the division/firm
- 💻 Your product
- Imagine if you were building your product from scratch & you had access to this tooling. What would you do? Disrupt yourself before a competitor or startup does.
- Plan & Invest time and cycles to get the right foundation.
- Perform a data review across product & partner data, you have the seeds for innovation here.
- ✅ Hope to expand on this topic in the next post.
- 🤑 Your customers
- Have a one pager with your POV, your customers or your internal teams are going to ask you soon! Make sure to collaborate with your engineering leadership, it's going to impact them a lot more.
- Customers love to align on their vision & where your product is heading to. Having these conversations proactively puts you on the front foot. It is an opportunity to delight them (if done right), don’t waste it!
Can I use AI to improve my job
Product Management is a craft. As product managers & leaders, our job is to deal with the ambiguous parts and bring clarity and get the team to the finish line! However, our jobs also entail dealing with a lot of mundane & repetitive but very valuable tasks that keep the engine humming along. Our attention is having to be spent across a cross section of folks and myriad of tools all while keeping the eye on the customer and giving them tools to make their lives easier & productive.
Here is an example summary of where PMs spend time:
A great opportunity for us to lean on AI makes it easier for us and frees up valuable time for us to focus on the craft of the role → do the harder, nuanced, and pivotal aspects of the role. There will always be a need for us to bring our emotional quotient to the team and that is situational. Lastly, while frameworks might be all the rage of the near past remember, AI can be trained on any framework and use it - we are truly entering an age where our value is being able to decide how to put various things together according to the situation at hand.
The near-term opportunity is in making the mundane and repetitive → easier, faster and more efficient! A ton of tools are popping up to help with a myriad of tasks. Here are some areas in which you can put AI to use today
- ✍🏼 Compose Mode
- Release notes → Makelog
- Call summaries → Vowel.ai, Cogram
- Product Copy/Translation → Quickie
- Product Documentation → Scribe
- Generate starting templates → ChatGPT
- Saying “No” nicely :) → ChatGPT
- Status updates → Makelog with ChatGPT
- Help writing emails → ChatGPT Writer
- 📖 AI-powered reading
- Summarization → MapReduce, DetangleAI, Gimme Summary
- Competitive Analysis → BrowseAI
- Scraping Data → ExtractGPT
- Product Research → Kraftful
- Market Research -> Notably
- Developer Docs → Kapa.ai
- General Reading → Readwise with GhostGPT reader
- 🕵🏼 Query
- Search engine for your work → Rewind.ai
- Querying data across data sources → Httpie
- If you wanna see the possibilities of what is to come, see this demo
- In the future, your zoom calls will have a data genie to query relevant information → https://findly.ai/
- Better knowledge bases → Chatbase
- 👷🏻♀️ Prototype
- Convey your ideas faster → Graphic.design
- Build a basic prototype → BiFrost
- 💬 Bot
- Provide a simple bot that can answer for you internally →Chatbase
- ✅ Consolidate tasks
- Consolidate tasks across various tools that you use & organize → Loop
- 🗓️ Better manage your time
- Managing time → Reclaim.ai
- 💻 Terminal
- Sometimes you just need to talk to the machine directly → Plz (Unix commands)
I haven’t listed all of the tools but the ones that I found valuable. Some of these products use OpenAI, so it’s a good idea to get a ChatGPT plus $20/month subscription which gives you faster response times, and priority access to new features and improvements.
Great where can I learn more about it?
- ⚒️ Sign up for ChatGPT and use it
- 🧑🏻💻 What is ChatGPT? And What is it doing? by Stephan Wolfram
- 📺 If you’d like to understand the technology behind GPT & transformers Follow Andrej & view this video
- 💌AI newsletter → The Neuron Daily
- 🎤 Fireside chat with Reid Hoffman by Elad Gil
- 🎤 Fireside chat with Sam Altman by Elad Gil
- 🧑🏻💻 AI and the Big five by Ben Thompson
- 🧑🏻💻 Is AI the new crypto by John Luttig
- 🎤 Must read post “This VC is automating his Job” Dan Shipper interviews Yohei Nakajima
- 🧵Twitter on Foundational models by Jon Turow
- 🎤 📺 Satya Nadella’s interview about AI with the Verge
- 📺 “I tried AI. It scared me” by Tom Scott
- 🎧 No Priors Podcast by Elad Gil & Sarah Guo
- 🎧 Latent Space Podcast by Alessio & Swyx
Why I am excited about the future?
Footnotes
2: Joe Rogan interview Steve Jobs https://podcast.ai/
3: Insert Twitter link to AI launches
5: Mashable article on mobile phone pings
6: https://arxiv.org/abs/1706.03762 .
7: Google’s transformers paper from 2017 https://arxiv.org/abs/1706.03762
8: Describe & Done tweet