CookBookGPT🍳[ HTC #62]
Welcome to Hold the Code #62!
HTC would like to welcome two new writers: Mark and Kimberly! Featured in today’s article, we discuss a new application from Google that implements computer vision in combination with Native languages. Could social media be in control of finances? Take a look at our next article on social media and it’s influence on bank runs! Lastly, our writer Mark walks us through his use of ChatGPT to prepare a new recipe!
Practice Native Languages with a Snap!
Image Source: Google’s Woolaroo
Language is an important element in describing our surroundings, our relationships with people and places, for many communities, in the practice of storytelling too.
In 2019, the United Nations declared it as an International Year of Indigenous Languages. Most recently, the United Nations Educational, Scientific and Cultural Organization (UNESCO) declared 2022 the beginning of the International Decade of Indigenous Languages.
Google’s Woolaroo pilot app experiment, which uses Vision AI, uplifts the act of preserving Native languages with machine learning. In 2022, they made seven more Indigenous languages available on the Woolaroo app. What does the development of Woolaroo’s platform mean in navigating ethical use of AI and uplifting cultural practices including the use of language?
What is Woolaroo?
Woolaroo, is Google’s open-source, software with source code that is publicly accessible, photo-translation platform powered by machine learning and image recognition.
The app uses Google Cloud Vision API, a tool that has been used across 1612 and counting of Google’s services.
Google Cloud also allows for customers like The New York Times to access pre-trained models including Vision API, a key feature of Vision AI.
The pre-trained machine learning models under Vision API make it possible for quick image recognition and the classification of everyday objects, product logos and more to take place.
Woolaroo’s current platform allows you to translate from 5 languages to 17 languages:
Translate from: English, French, Spanish, Arabic, or Italian
Translate to: Yugambeh, Maya, Yiddish, Tamazight, Rapa Nui, Nawat, Māori, Calabrian Greek, Sicilian, Louisiana Creole, Yang Zhuang, Tepehua, Vurës, Potawatomi, Serravallese, Sanskrit, or Kumeyaay/Diegueño
Inspiration behind Woolaroo
More than 7,000 languages are spoken worldwide and 6,700 are Indigenous languages, many of which are at risk of becoming endangered, especially Native languages.
Woolaroo’s first featured language for translation was Yugambeh, an Indigenous Australian language of southeast Queensland.
Image Source: Google’s The Keyword
Google Arts & Culture first collaborated with the Yugambeh Museum team to use three decades of cultural and local language research into the database provided for the application, after acknowledging the difficulties for youth to learn the language to describe contemporary subjects.
Lost in translation is something that is more prone to occur for communities whose language may have no direct translation. For example, in the Yugambeh language, there is no translation for the word “shoe.”
For Allan Lena, a frontline advocate working to teach children Yugambeh, Woolaroo presents itself as a tool to reach and share the language to younger generations.
In 2021, Rory O’Connor, CEO of the Yugambeh Museum wrote for Google’s The Keyword about what Woolaroo meant for their community.
“Crucial to Indigenous communities is that Woolaroo puts the power to add, edit and delete entries completely in their hands,” said O’Connor.
Given the app’s open-source feature, family members and relatives from Indigenous communities are able to make contributions when they remember words or phrases.
Woolaroo comes from the word “shadow" in the Yugambeh language. The app’s developers described the reasoning behind the app's name as an appreciation of perspectives from the Yugambeh community that see pictures like shadows.
Launching Woolaroo for the first time?
When launching the app in your browser, one will be asked to take a photo or to choose a photo from their device. Once a photo has been taken or uploaded, the platform will generate words of subjects it identifies in the photo along with a translation and pronunciation guide according to the languages a user has selected.
Looking Ahead
While UNESCO predicts that as many as 3,000 Indigenous languages may be lost by the end of the century, Woolaroo’s goal is to help in combating this loss through the ability to snap or upload a photo and get a real-time translation, allowing individuals to interact with their environment and retain information more easily, so that it is easier to apply in day-to-day conversations.
“Our language is the vehicle to understanding the Māori world,” said Tania M. Ka’ai, a professor at Auckland University of Technology in New Zealand.
Additionally, language revitalization takes collective efforts including financial resources. Upcoming platforms, like Woolaroo, with a goal of serving communities should also continue to make space to include community input and feedback in order to build transparency around the ethical uses of new technologies.
As Indigenous communities navigate opportunities to be in touch with their relatives, the potential extension of Woolaroo for more Native languages may be of great support for these communities to preserve their languages too.
Is Social Media the New Risk to the Financial Sector?
Image Source:
The 2008 bank run on Washington Mutual, the sixth-largest bank in the United States at the time, lasted nine days. $16.7 billion in deposits were withdrawn, which amounted to 9% of the deposits it had held three months earlier. In March this year, Silicon Valley Bank depositors withdrew $42 billion, a quarter of its deposits, in a few hours, and the bank collapsed in two days. The speed at which this occurred left commentators calling it a “bank sprint”, one fueled by social media and enabled by digital banking.
Social media bank runs
A bank run occurs when a large number of customers lose confidence in a bank's ability to protect their money and withdraw their deposits all at once. As more depositors withdraw their funds, banks become less likely to meet demand, causing more worry and fueling the run.
Before the run, SVB was already in a vulnerable state due to having invested depositors’ money in long-term US government bonds, which depreciated in value when interest rates increased. To satisfy customer withdrawals, the bank had to sell the bonds at a huge loss. The day before the bank run started, they’d announced this loss amounted to $1.8 billion.
The trigger of the bank run is said to be recommendations from venture capital firms to companies in their portfolio to move their money out of SVB. The news spread over Slack, Twitter and WhatsApp and speculation about the bank's insolvency grew. With the ease of digital banking that allows customers to move their money in a few taps and the catalyst of fear that is social media, the rumors culminated in an actual crisis.
However, it could be argued that SVB’s characteristic of having a customer base highly concentrated in technology and venture capital entrepreneurs made them uniquely vulnerable to a social media-driven bank run. Lindsey Johnson, president of the Consumer Bankers Association, also cautioned that this wasn’t a sign of a long-term risk to the financial sector:
“The last several days represent a unique incident fueled by misinformation on social media and are not indicative of the health of our industry”
Credit Suisse and Twitter rumors
SVB might be the first social media bank run, but it wasn’t the first bank to be destabilized by Twitter rumors. Last October, Australian ABC business reporter David Taylor tweeted:
The tweet went viral as other reporters spread the news and rumors began to circulate that the unnamed bank in the tweet was Credit Suisse, the second-largest bank in Switzerland that had been plagued with scandals for years.
The tweet was posted on a Saturday and by Monday, Taylor had deleted the tweet. When the stock market opened that day, Credit Suisse shares plunged almost 12%. Despite guarantees that it had sufficient liquidity and capital, Credit Suisse failed to recover as clients continued to withdraw cash and sell shares.
On March 19, UBS took over Credit Suisse in a deal brokered by the Swiss government to prevent a banking crisis. Marlene Amstad, chief of Swiss financial regulator Finma, confirmed that the cause of the withdrawals and disinvestments were “rumors since the autumn”, a reference to Taylor’s tweet.
Conclusion
Now, information travels fast. The efficiency of digital banking enables people to act fast. Gone are the days where people had to physically line up outside a bank to withdraw their money (yes, I’m thinking of It’s a Wonderful Life).
This new digital age is exposing vulnerabilities in the financial sector. We saw a glimpse of this from the short squeeze on GameStop mobilized on the subreddit r/wallstreetbets in 2021 and the subsequent rise of meme stocks. With First Republic Bank being the latest bank to fail, it remains to be seen if officials can assure the stability of the financial system. One thing is certain, even traditional industries like the finance sector cannot escape the wrecking ball that is social media.
AI: Your Next Recipe Writer?
Image Source: Mark Fortes
If you’re reading this article, chances are you know by now just how powerful ChatGPT can be. Today, however, the purpose of my ChatGPT consultation is a little less ambitious. As a cooking enthusiast, I thought it would be an interesting experiment to ask ChatGPT to create and walk me through a recipe. Here’s how it went.
The experiment
As a Filipino-American in Kaibigan (the Philippine-American club on campus), with our annual cultural show just around the corner, I was feeling particularly cultural and thus opted to challenge ChatGPT with an ethnic dish. After briefly introducing myself, just to keep things polite since I was hoping for a great recipe, I popped the question:
“Create and walk me through a Filipino recipe.”
To which I was answered with,
“Sure, here's a recipe for a classic Filipino dish, Chicken Adobo.”
Nice! One of my favorites… Off to a good start! Let’s see if it can keep it up. Adobo is a Filipino staple, and a traditional recipe should culminate in a savory and tangy dish made with slowly simmered chicken in a flavorful sauce of soy sauce, vinegar, garlic, onion, bay leaves, and peppercorns. The sauce should provide a rich umami undertone with a deep brown and glossy sheen, with the chicken tender and moist. Can ChatGPT help me make my ancestors proud?
Trust the process
Here were the ingredients I was provided:
1.5 lbs chicken thighs or drumsticks, bone-in and skin-on
1/2 cup soy sauce
1/2 cup white vinegar
1/2 cup water
1/4 cup brown sugar
8 cloves garlic, minced
1 onion, chopped
2 bay leaves
1 tsp black peppercorns
2 tbsp cooking oil
chopped scallions or cilantro for garnish
After a quick consultation with my Philippines-born-and-raised parents– it looks fairly authentic! Besides the cilantro garnish, which is more of a modern, fusion take. I told ChatGPT it was time to “let him cook”, gathered my ingredients, and began.
ChatGPT’s instructions were as follows:
In a bowl, combine the soy sauce, vinegar, water, brown sugar, garlic, onion, bay leaves, and black peppercorns. Mix well.
In a large pot or Dutch oven, heat the cooking oil over medium heat. Add the chicken and cook until lightly browned on both sides, about 3-4 minutes per side.
Pour the sauce mixture into the pot with the chicken. Bring to a boil, then lower the heat to a simmer and cover the pot.
Cook the chicken in the sauce for about 30-40 minutes, or until the chicken is cooked through and tender, and the sauce has thickened.
Remove the bay leaves and discard.
Serve hot with steamed rice, garnished with chopped scallions or cilantro.
The result…
The result? Take a look at the cover photo!
If I had to rate ChatGPT’s recipe, I would give it a 6/10. I like that in addition to including an estimated cooking time frame, it also included how the chicken or sauce should look or feel. As a result, it was helpful to be able to use both criteria to help reach my goal. I felt that the instructions had just enough detail to set me up for success while also leaving room for variations in things like stove temperature. It was interesting that ChatGPT chose to omit some common initial steps in online recipes like “Mince the garlic and chop the onions,” instead opting to specify them in the ingredients list.
However, ChatGPT’s recipe was far from perfect. Despite including the quantity or measurements of each ingredient, the number of servings they would provide was not mentioned. Prep and cook time estimates were also absent, which I found problematic, as these three figures are very important factors in crafting and following a recipe. Furthermore, ChatGPT is unable to show pictures in its recipe, which I usually find to be helpful and motivating.
Overall, though, I’d say things still worked out nicely! It was a fun time trying out a new recipe from a robot, and although not the most beginner-friendly method, I would recommend this experiment to anyone who is interested in learning to cook a dish in a new, unconventional way. If necessary, you can ask ChatGPT to give more detailed instructions, and even receive tips on things like knowing when meat is fully cooked.
Although, ChatGPT could learn to be a little more humble…
Me: how could you have made these instructions better?
ChatGPT: To make the instructions even better…
Love HTC?❤️
Follow RAISO (our parent org) on social media for more updates, discussions, and events!
Instagram: @nu_raiso
Twitter: @raisotweets
RAISO Website: https://www.raiso.org
 | A guest post by
|
 | A guest post by
|