Robots That Write Their Own Code | The Weekly Roundup

Hi there everyone. Welcome to another episode of the weekly roundup your go to for the latest and greatest in data science. This week we're talking about robots that can write their own code, the wave of tech layoffs, Italy's tax AI and more. So stay tuned to stay in the loop and make sure that you're subscribed so that you never miss another update. In our first story this week, we'll take a look at an announcement from Google's AI event in New York City in which the company demonstrates robots writing their own code based on natural language prompts. The research on display at the event has been called Code as Policies or CAP for short, and aims to demonstrate that by allowing language models to write code, they can perform precise arithmetic, spatial geometric reasoning and express a degree of behavioral common sense. Google deployed CAP on various robot platforms and showed that it can perform diverse tasks from tabletop Object manipulation to 2D drawing or without any additional model training. Large language models have shown impressive capabilities not just in natural language understanding, but also in reasoning tasks, from reading comprehension to answering difficult math questions. Essentially, Google wants to build a system that takes as input high level natural language instructions of a task, for example cleaning up a coffee spill, and outputs actions that a robot can execute. As part of the announcement, Google is releasing open source versions of the code, which we'll link in the description below for those of you who want to take a closer look. Moving on to our next story, A German high school allowed learners to use AI tools in exams and found that most students didn't blindly rely on just AI to do their work for them. At the Evangelische Stifte Schickmnasium In Gutersloch, Germany, GPT3 and other AI software heavyweights are being tested in lessons and are even required for class tests. Since the pandemic began, the school has been experimenting with digital exam formats, and a big part of this new culture is that students can do all or some of their work with the assistance of text AI. In lessons. Leading up to the exams, the students had learned that it's unwise to simply adopt AI generated text, and they covered topics such as AI text bias, the outdatedness of some models, the stylistic shortcomings of German writing AI programs, and the gaps in reasoning that AI texts can leave during the exam. The school noticed that many students do additional research on the Internet to verify the information in the AI texts or to sift through additional evidence, expert opinions and studies. Often they combined multiple AI tools to achieve better results. We love to see the integration of new technology in places of learning and can't wait to see the impact this will have in the future. Our next story takes a look at the huge wave of layoffs in the tech industry as Meta, Twitter and Salesforce all announced big job cuts. Mark Zuckerberg blamed the meta layoffs on the pandemic, claiming that he thought the boom in e commerce brought about by people staying at home during lockdowns would last indefinitely. But the company plans to let go of around 11,000 employees in the coming weeks, echoing moves made across the industry. Twitter has also let go of thousands of employees over the past week, according to Twitter employees who are now slated to leave. The company's entire machine Learning, Ethics, transparency and accountability team is to be disbanded. The team was researching and pushing for algorithmic transparency and algorithmic choice, along with inventing and building ethical AI tooling and methodologies. Salesforce's layoffs number in the hundreds and come alongside an announcement of a hiring slowdown as the company looks to significantly cut costs in the short term. This is certainly a difficult time for many folks in tech around the world, and our thoughts are with everyone who is now on the hunt for a new job in this tough economic climate. In our final story, Italy's anti tax evasion AI, which has prevented more than 6.8 billion euros worth of tax evasion, is running into data privacy concerns. The VERA algorithm compares tax filings, earnings, property records, bank accounts and electronic payments, looking for discrepancies. High risk taxpayers then receive a letter asking them to explain the differences. The more data VERA processes, the smarter it becomes. Italy's traditionally high income taxes, a cash economy and a culture of impunity have been the major drivers of this crisis, which sees Italy lose more than 99 billion euros in tax each year. More than 80% of payments made in Italy are made in cash, enabling VAT evasion, according to the bank of Italy. While many governments around the world use sophisticated tax systems to claw back money, few do so as aggressively and with as much data to draw on as Italy. Alessandro Santoro, an advisor to the Ministry of Economy and Finance who heads its anti tax evasion committee, says that the implementation of data protection decrees won't be immediate, further saying they'll need to do testing and experimental procedures before changing the way the algorithm works. We can't wait to see how Italy manages to strike a balance between compliance and innovation in the coming years. That's all for this week's roundup. Thanks for watching. We hope you have an amazing week. And don't forget to let us know in the comments what your favourite story of the week is.