The Future Stack AI Newsletter #6
Welcome to the sixth edition of the newsletter. This newsletter is about talking about, predicting, and creating the future of AI together. Because the best way to predict the future is to create it!
⚡New this week
For practitioners - Embodied AI, Fine-Grained Visual Categorization
For business leaders - Data & AI Strategic mistakes
For investors - Look for data-native companies
For the Australian Industry - we need more courses and trainers that teach professional skills
💗Mental Health Check-in
In this world of FOMO and information overwhelm, find a problem you are happy working on for the next 10 years and go all in! You will thank me later 😉
👩💻AI for practitioners
🦾🦾🦾Embodied AI and why you should know about it?
Embodied AI describes the shift from “using data, images, and text to train algorithms” to “training agents in physical systems to learn through interactions with their environment from an egocentric perception similar to humans”. Such systems are mostly part of the field of robotics and differ from reinforcement learning systems in the fidelity of the simulators they use. Click below to know more.
🔗A Survey of Embodied AI: From Simulators to Research Tasks
Organizations working on Embodied AI: NVIDIA, Logical Robotics, Boston Dynamic, Meta AI, CMU, Apple, USC, GaTech, SFU, U Padova, AI2, Intel, UW, Google, UCSD, CSIRO, Amazon, UIUC, Stanford, Neuromation.
This field is a culmination of years of research in computer vision, reinforcement learning, robotics, and other fields. Right now humans have to get on their phones or laptops to interact with AI systems, but reading about the work in this field made me realize how much closer we are to interacting every day with physical AI systems like robots.
🌿🌿🌿Fine-Grained Visual Categorization
According to the computer vision group at Cornell, “Fine-grained categorization, as a sub-field of object recognition, aims to distinguish subordinate categories within entry-level categories. Examples include recognizing species of birds such as “northern cardinal” or “indigo bunting”; flowers such as “tulip” or “cherry blossom”. Fine-grained categorization often requires efforts from different aspects compared with generic object recognition.”
This task requires a high level of domain knowledge and can be difficult even for humans. The tasks of FGVC involve differentiating different species of animals, plants, art, etc. The main challenges in such tasks lie in domain generalization, exploring new data horizons, as well as learning complementary models that can boost performance.
For those interested in exploring more, CVPR 2023 hosted a workshop on this task.
🔗Link to the CVPR 2023 workshop
🔥AI for business leaders
This time I have a shameless plug in place of some advice. I am currently working with three different startups - one in the field of healthcare and two in the field of education advising on their data strategy.
So if you are looking for any strategic advice on how to start building a machine learning capability in your organization, reach out to me through my LinkedIn.
🕴️AI for investors
🧐🧐🧐How to identify if a startup has a moat?
Patents or proprietary technology: This includes unique algorithms that offer a significant advantage over what is available that is significant from a business requirements perspective.
Data & Data Network Effects: Exclusive data belonging to a niche that is not easily available as well as data network effects make it difficult for competitors’ products to be at par and create a big enough barrier. Always look for unique data advantages or potential for such an advantage.
Strategic alliances: A company that has alliances to get the data and expertise they will need (maybe with universities) has a unique advantage. Also, look for a startup offering products/APIs to its competitors making it one of a kind in the ecosystem.
Automation: A startup that has automated its processes and succeeded in improving the productivity of its talent gets extra brownie points.
Agility: Agility matters a lot where innovation happens at a rapid pace. Look for a strong team that pushes out updates faster than the market is evolving.
🦘AI for the Australian industry
🎰🎰🎰 How to train the future workforce of AI?
If we talk about the Australian workforce, the skills penetration is much less than India, China, and the US. The reasons mainly lie in the government’s unwillingness to invest in such an important technology for the country and the general risk-averse ecosystem we live in.
One way is to incentivize private equity to flow in Australia. Australian firms need to expand their markets and think globally. Once money starts flowing in, the demand for talent is going to increase and that will incentivize people enough to start focusing on emerging technologies. Another way is for leaders in the government to seriously consider the future of our country if we continue to be laggards in risk-taking and how that will impact the quality of life of Australians.
The learning process needs to change too! Not just in Australia, but everywhere!
I think most of the learning we do - either in online courses or even universities is not applied enough. When it comes to bringing ideas to life and building a proper product or service from technology, there are many skills missing in an academic context that need to be either self-taught or learned on the job.
Also, skills quantification is something that academic institutions don’t do a good job of. A skill is something that allows you to do something. It is very hard to measure that unless we understand that, define it, and put it to a practical test.
For example, what is the basic skill someone who claims to know prompt engineering should have?
*“Use LLMs and prompts to create end-to-end backend NLP Applications”
*This is a definition derived from many LinkedIn job posts that were looking for a prompt engineer.
It is defined not in terms of a tool or a language, it is a capability!
The best way to test a capability is a practical setting that utilizes this capability.
A whole new model of education has started to evolve in the dark streets of the startup world that will change the way we learn. Personalized learning will become more commonplace and learning will lead more and more toward building capabilities that will be quantified much better. I see more instructor-led, experience-based courses becoming commonplace and more and more companies open to work-integrated learning opportunities for students straight out of high school.
🔧AI for Talent
If you are someone trying to figure out where to apply your skills, go to this website 80000hours.org. They talk about how you can work on an issue that not only motivates you but also has a huge impact on the current problems that the world is facing.
I have been reading a lot about them and I have a matrix that I use to rank the issues based on my skills, affinities, and growth prospects. If you are interested, you can email me at tanya@viterbi.ai to get the matrix (a bit personal to share it here :) ).
👩💻 About me
I am an ML engineer in Canberra and my most intense obsession is to help companies and individuals become competitive by investing in data and AI. Subscribe to my newsletter if you have similar ideas on making a dent in the universe. All the content that I create is free and I intend to keep it so!
Follow me on LinkedIn here!