- x NeonPulse | Future Blueprint
- Posts
- 🤖 A.I. asset managers, avocados, and ChatGPT's inherent bias
🤖 A.I. asset managers, avocados, and ChatGPT's inherent bias
Neon Pulse Team
February 22, 2023
Good morning and welcome to the latest edition of neonpulse!
Here's what we have for you today:
Machine learning takes on asset management
A.I. comes for your avocados
and ChatGPT’s inherent bias
A.I. Portfolio Management
One of the most promising areas where A.I. has yet to be widely adopted is investing.
This isn’t because the technology isn’t a good fit, but simply because investment managers haven’t yet found the best way to apply A.I. and machine learning to their investment decisions.
Yet a recent set of studies by a Dutch investment firm is set to change that after showing that machine learning models can “substantially outperform” comparable strategies.
Robeco, a Dutch investment firm with 200 billion Euros under management, recently performed a number of studies using machine learning investment models and found that the ML powered models were able to provide significantly better predictions than traditional quantitative modeling.
One of the primary reasons that ML models are able to outperform traditional investment models is because ML models are able to better identify the signal from the noise, finding the variables in a data set that have the most meaningful impact on asset pricing and using those variables to form predictions on future prices.
“Machine learning is well equipped to deal with large sets of features and ‘learns’ to give the most weight to the most relevant variables,” according to the paper.
And beyond simply providing future value predictions based on historical prices, machine learning models are also able to extract meaningful insights from qualitative data (such as analyst predictions, SEC filings, and earnings calls) that investors can use to forecast events, such as earnings surprises, potential mergers and acquisitions, and even share buybacks.
As with any new technology, the early adopters stand the most to gain, with Robeco’s research paper concluding that “asset managers who disregard advances in machine learning will see their performance decline relative to those who embrace machine learning.”
You can check out the full research paper here.
Brought to you by Jasper.ai
If you have to write a lot for work, whether its marketing copy, social media posts, or marketing emails, there are a number of great A.I. tools that allow you to create powerful copy in a flash.
I think the best way to describe Jasper is that it’s like kryptonite for writers block, helping me to turn out high quality content with perfect grammar in an instant.
If you don’t have to write a lot for work it’s not a must have, but if have to produce content reguarly, and frequently find yourself staring at a blank screen praying for inspiration, its definitely worth checking out.
If you’d like to take Jasper for a test drive, neonpulse readers can get 10,000 free credits by clicking here.
A.I. targets supermarket shrinkage
One of the areas most prone to theft in grocery stores is self checkout.
From customers deliberately misidentifying items (entering avocados as a cheaper item like onions) to customers simply not scanning items, the losses at self checkout areas have been a constant source of headaches for grocery retailers worldwide.
A 2018 report conducted by the ECR Retail Loss Group found that in a grocery store where 50% of the transactions are being processed through self checkout, shrinkage losses are 75% higher than the average rate in grocery retailing.
And with the average profit margins of groceries stores hovering in the 1% to 3% range, this increased level of shrinkage can have a significant impact on the profitability of the business.
In order to help combat this issue, Australian supermarket chain Woolworth’s has developed a new A.I. powered self-checkout system, and has begun rolling the technology out across their 1000+ locations.
“We know the vast majority of our customers do the right thing at our self-serve checkouts. This is a new security measure we're trialing for those that don’t,” a Woolworths spokesperson said.
Artist rendering of future supermarket security technology
By using a combination of cameras and image recognition software, the new self checkout systems are able to make sure that all items are scanned and scanned correctly.
"And if a mis-scan occurs, a short video plays highlighting the affected product and customers then have the opportunity to re-scan it."
And beyond helping to minimize shrinkage, these systems are also designed to help make the self-checkout process more seamless by automatically identifying the types of item you are scanning.
"So if a customer places a loose tomato on the scanner, the system will show a range of tomato varieties rather than the full list of fruit and vegetable items," the spokesperson explained.
In other news, now that people are going to have to actually start paying for their avocados, I’ve entered into a substantial short position in GUAC.
ChatGPT’s inherent bias
In the latest episode of the All-In podcast, entrepreneur and investor David Sacks exposed what he believes to be an ”inherent bias” in the safety layer that governs ChatGPT’s responses.
One topic that Sack’s thought highlighted the bias particularly well was the difference in ChatGPT responses when asking for a poem about Trump as opposed to asking for a poem about Biden:
“There is mounting evidence OpenAI's safety layer is very biased... If you thought trust and safety were bad under Vijaya or Yoel, wait until the AI does it,” Sacks said.
Elon Musk commented on a video featuring Sacks on Monday, stating that he believed that the built-in bias was a “major problem.”
You can check out a twitter thread with Sack’s commentary along with additional ChatGPT prompts highlighting the safety-layer bias here.
And now your moment of zen
That’s all for today folks!
If you’re enjoying Neon Pulse, we would really appreciate it if you would consider sharing our newsletter with a friend by sending them this link: