DAYMN — 19 Sep 2021
Data Articles You Might Need. Subscribe & follow here!
After being away for a week, I was surprised to see how many people reached out asking about DAYMN. I had initially estimated that number to be close to ZERO, but it was anything but. So, thanks to all of you — who take the time to read DAYMN every week and also encourage me to do this more often.
Now, here are the top 5 articles from this past week — please do share your feedback & thoughts!
1. How OpenAI Sold its Soul for $1 Billion
OpenAI has been the pioneer and inspiration for a lot of people and organisations looking to do “the right thing” with AI technology and also focus on making it an equitable area of research. However, a recent massive investment and a shift in their profit-focus, are now a bit worrying. It brings up the question, “can there ever be a movement or organisation for AI and its applications in society who cannot be swayed by money or power?”
This eye-opening article from Alberto Romero shines a light on this concern.
https://onezero.medium.com/openai-sold-its-soul-for-1-billion-cf35ff9e8cd4
Call out? In 2019, OpenAI became a for-profit company called OpenAI LP, controlled by a parent company called OpenAI Inc. The result was a “capped-profit” structure that would limit the return of investment at 100-fold the original sum. If you invested $10 million, at most you’d get $1 billion. Not exactly what I’d call capped.
2. New approaches to delivering packages accurately
As a big fan of using intelligent ways to enhance supply chain and operations research problems, this article recently on amazon.science was very interesting for attempting to use a completely different tech to solve a totally different problem. The researcher uses a tangential approach of Learning To Rank (L2R), which is typically employed in search and information retrieval algorithms — to solve for the problem of identifying the “most accurate delivery drop GPS location” for a customer.
https://www.amazon.science/blog/using-learning-to-rank-to-precisely-locate-where-to-deliver-packages
Call out? There’s a difference between the popular learning-to-rank methods in information retrieval and my adaptation of them, however. In the search engine scenario, the algorithm may be sorting through tens of thousands of documents or products to produce a ranking. In the geospatial case, however, we typically compute offline, and we usually have fewer than 100 sufficiently distinct candidate locations to consider for each delivery address. This makes it practical to compare each candidate location against all the others at inference time: we select the one that wins the most pairwise comparisons
3. Bayesian Media Mix Modelling for Marketing
I have built and delivered dozens of Media Mix Models, and Multi Touch Attribution models for digital marketing teams of various sizes and shapes. Despite using Bayesian models to build some of these media mix models, one thing that was always missing - was the lack of good, clear and articulate material that could describe the why, what and how of using a complex and hard-to-understand technique like Bayesian modelling for marketing teams.
This article from PyMC Developers is exactly that. I highly reading it to understand how modern approaches can make marketing optimisation, so much better and efficient.
https://www.pymc-labs.io/blog-posts/bayesian-media-mix-modeling-for-marketing-optimization/
Call out? Another significant benefit of the Bayesian approach is that we know how confident we are. Yes, we want to make data-driven decisions. Still, we only want to do so when the data provides us with a sufficient level of confidence in our understanding. Parameters pertaining to Marketing channels that are estimated with high level of uncertainty can motivate targeted marketing efforts (i.e. incrementality tests) to resolve this uncertainty.
4. State of AI in Creative Automation
Apart from sensational headlines of SkyNet-esque takeover of AI, the applications of real-world AI tools are either fairly sparse or not as significant yet. Most of the applications that gain attention are customer-choice focussed, operational-efficiency aligned or revenue-maximising driven. One area of intelligent application that has been the sole bastion of human intelligence “superiority” has been the creative arts. Of course, AI research has been making good in-roads here too, but mostly what we know are superficial.
This wonderful digest from Eze Vidra on VC Cafe is the answer to the question — “Tell me everything about what’s happening with the use of AI in images, videos, art, gaming and more?”
https://www.vccafe.com/2021/09/08/the-state-of-creative-automation/?amp=
5. What is Productivity? And how does one define it?
Thanks to this wonderful share from Karthik Subramanian, it got me thinking about what “productivity” really meant, especially in the context of large teams with a lot of autonomy and unrealistic workloads. What the folks at GitHub have done is to draw out a scientific and “wholesome” way to frame and measure productivity. In their own words,
The authors assert that by measuring developer productivity using a multidimensional framework, SPACE, we can more accurately measure productivity and make better decisions, and capture insight into the many layers of organizational, team and developer productivity. The SPACE framework presents five categories important to consider when measuring productivity.
S — Satisfaction & Well Being
P — Performance
A — Activity
C — Collaboration & Communication
E — Efficiency & Flow
https://github.blog/2021-03-10-measuring-enterprise-developer-productivity/
Would you measure your own work or your teams’ work using this framework? If not this, how would you adapt or change this?
Call out? No single measure can be taken alone to draw conclusions about productivity. Teams and organizations should leverage at least three dimensions to gain deeper insight into how teams are working.
Have a wonderful week ahead everyone, hope at least one of these articles is exciting reading material for you, and made you think for a moment
