Nvidia and Foxconn making AI factory, Project Primrose, and Deepfakes in Election | W@G #6
Week At Glance #6. [Formerly "On My Feeds"]
Welcome to this week’s Week at Glance. In this weekly segment, I share curated links and reports that will keep you updated on what happened over the week in the the AI, Tech and Science community.
For this week, I have curated links mostly from the AI domain. The credit goes to the original authors and sources, so do check them out the original and support their work.
As a note, all the images and videos are taken from the respective articles — unless mentioned otherwise.
Let’s Begin.
Here is the Top-5 news that happened over the week :
Foxconn and Nvidia to collaborate and make AI software for self-driving cars. (Maxwell Zeff, gizmodo.com)
Stanford researchers issue AI transparency report. All models scored "unimpressively". (Anna Tong, Reuters)
Adobe aims to revolutionize the fashion industry. Uses flexible displays that can create static or dynamic patterns on any surface, even clothes. (Jimmy Pezzone, TechSpot)
AI deepfakes create false information in Slovakian vote, turning the deepfake fear into reality. ( Olivia Solon, Bloomberg)
Million of Workers Training Million-dollar AI Models for Pennies. (Niamh Rowe, WIRED)
I. Foxconn and Nvidia to collaborate and make AI software for self-driving cars.
—Maxwell Zeff, gizmodo.com
This reveals that Nvidia — a dominant supplier of AI and graphics hardware and software, and Foxconn — the world's largest contract manufacturer of electronics, are teaming up to create AI factories for self-driving cars. This announcement, made at the Hon Hai Tech Day earlier this week, shows Nvidia’s plans and vision to dig deep into the AI industry — including that of Self-driving cars.
Nvidia CEO Jensen Huang and Foxconn CEO Young Liu shared their vision for the future of AI at the event. These AI factories would use Nvidia's GPU to process, train and transform large amounts of data into AI models. The infrastructure will be powered by Nvidia's GH200 Grace Hopper Superchip, which are designed for high-performance computing and AI applications.
GH200 Grace Hopper is a combination of CPU and GPU chips, with faster memory, that is designed for giant-scale AI and high-performance computing.
“There will be smart factories that build cars, there will be smart factories that build AIs, and these two factories will be compliments of each other,” said Huang in the event.
Huang believes that every industry will need AI factories in the future. Liu, on the other hand, said that Foxconn is transforming into a platform solution company that focuses on smart cities, smart manufacturing and smart EVs.
Matter of fact, Nvidia and Foxconn are both undergoing transitions in their businesses after the onset of AI.
Nvidia is moving from a graphics chip maker to a data center-scale computing company. Whereas, Foxconn is moving from a contract electronics maker to a platform solution provider.
The AI factory will develop software for cars like the Foxtron Model B, which Foxconn showcased behind Huang and Liu during the announcement.
II. Stanford researchers issue AI transparency report. All models scored "unimpressively".
—Anna Tong, Reuters
A team of researchers from Stanford University created the Foundation Model Transparency Index — a report that evaluates the transparency of popular AI foundation models from different companies.
The index graded 10 popular models on 100 different transparency indicators to measure how much information the model developers provide about the data, compute resources, human labour, and other aspects of their models.
The result: All AI models scored “unimpressively”.
The report found wide holes when it comes to transparency while building these AI models.
“It is clear over the last three years that transparency is on the decline while capability is going through the roof,” said Stanford professor Percy Liang, a researcher behind the Foundation Model Transparency Index
The report found that most models lack sufficient documentation and disclosure.
Many of the important aspects, such as data sources, data quality, model architecture, model training and evaluation, and model governance are not fully disclosed. This report, in some way, highlights the high need for regulations in AI governance.
These are the final positions of each model in terms of transparency and, as highlighted by the report, the results are not very encouraging.
The highest-scoring model, Llama 2 from Meta, only achieved a score of 53 out of 100. The lowest-scoring model, Titan from Amazon, scored 11 out of 100.
OpenAI's GPT-4 model — the ‘brain’ of ChatGPT and the most powerful and versatile AI foundation model ever created — scored 47 out of 100.
The researchers behind the report warn that the lack of transparency in AI foundation models poses serious challenges and risks for society.
The report also presents that transparency can benefit the model developers themselves, by fostering trust, collaboration, innovation, and feedback from the AI community and the public.
The researchers hope that their report will serve as a useful resource and a catalyst for change in the AI field.
III. Adobe aims to revolutionize the fashion industry, uses flexible displays that can create static or dynamic patterns on any surface, even clothes.
— Jimmy Pezzone, TechSpot
Adobe's latest innovation Project Primrose aims to merge technology and fashion.
Project Primrose, unveiled at the Adobe Max conference in Los Angeles, showed the potential of this technology with an interactive dress. In the event, Adobe research scientist Dr Christine Dierk demonstrated how the dress could change its appearance based on the user's input, such as voice commands, gestures, or touch.
Project Primrose is a digital dress that is made up of small scales or petals. These scales are programmed with Adobe software, which changes its design on user interaction and command. This futuristic "fabric" can be used for clothing, handbags, and other accessories
“Unlike traditional clothing, which is static, Primrose allows me to refresh my look in a moment,” says Dr. Dierkt. "Fashion doesn’t have to be static. It can be dynamic and even interactive. And we’re excited for a future where there’s more ways to express yourself,"
According to Adobe, user could choose from a variety of designs created with Adobe Stock, After Effects, Firefly, and Illustrator, or even create their own custom patterns.
Project Primrose is still in the early stages of development and is not yet available for consumers. However, the company hopes to inspire more artists and designers to explore the possibilities of this technology and create new forms of expression and communication
IV. AI deepfakes create false information in Slovakian vote, turning the deepfake fear into reality.
— Olivia Solon, Bloomberg
As Slovak parliamentary elections are approaching, some political actors are using AI-generated deepfake voices to spread false and misleading information about their opponents.
These videos are circulating on social media platforms such as Facebook, Instagram and Telegram, without any warning labels or fact-checking.
Touted as “the first swung by deepfakes” by the Sunday Times, a deepfake video showing a conversation between Michal Simecka, the leader of the Progressive party, and a journalist went viral on Meta; in which they allegedly discussed buying votes from the Roma minority.
The audio was synthesized by an AI tool that mimicked the speakers' voices.
“For three or four years everyone has been talking about the wave of deepfake manipulation.. but for whatever reason it didn’t happen. But there’s reason to think it could be different now,” said Rolf Fredheim, who conducted the research for Reset, who reported this deep-fake video.
Another example, according to the report, showed a video posted by Republika, a far-right political party, that features AI-generated voiceovers of Simecka and President Zuzana Caputova, urging voters not to follow "the progressive herd blindly".
These videos are still available on Facebook and Instagram as I am writing this.
Rolf Fredheim urged social media platforms and authorities to take action to prevent the spread of disinformation and protect the integrity of the election.
V. Millions of Workers Training Million-dollar AI Models for Pennies.
—Niamh Rowe, WIRED
AI is transforming the world, but at — literally — what cost?
This article by WIRED highlights what happens behind the scenes: millions of workers working tedious, for low-paid tasks to label data for training AI algorithms.
These workers are often exploited, overworked, and underpaid by platforms that profit from their labour.
The article begins by anchoring the story of Fuentes, a Venezuelan woman who fled her country's economic and political crisis. She works for Appen, an Australian company that provides data services for AI clients.
Appen pays Fuentes between 2.2 cents to 50 cents per task, depending on the complexity and availability of the work. Sometimes, she barely make $1 a day.
“Appen’s clients have included Amazon, Facebook, Google, and Microsoft, and the company’s 1 million contributors are just a part of a vast, hidden industry. The global data collection and labeling market was valued at $2.22 billion in 2022 and is expected to grow to $17.1 billion by 2030, according to consulting firm Grand View Research.” highlighted the article
The report further delves into the insights by the World Economic Forum, which says there are an estimated 260 million online workers worldwide, many of whom are doing data labelling for AI.
These workers are mostly located in developing countries — the likes of Venezuela, India, and the Philippines — where labour costs are low and regulations are weak. They face poor working conditions, lack of benefits, and constant uncertainty.
The article further highlighted:
“Mutmain, 18, from Pakistan, who asked not to use his surname, …says he joined Appen at 15, using a family member’s ID, and works from 8 am to 6 pm, and another shift from 2 am to 6 am. “I need to stick to these platforms at all times, so that I don't lose work,” he says, but he struggles to earn more than $50 a month.”
Some researchers have criticized this system as a form of data colonialism, where rich countries and corporations extract data and value from poor countries and workers.
They argue that this creates a digital divide and reinforces existing inequalities. They also question the ethics and quality of the data produced by such a system.
That’s it for this week. Hope you enjoyed reading them as much as I did. If you have any feedback, suggestions, or questions, feel free to hit reply and let me know.
If you find this newsletter helpful, share this newsletter with others.
Thanks for reading and see you in the next post
