Russia’s Yandex announces an open large language model

By

-

June 24, 2022

yandex announces open large language model

To spur more thorough development of certain AI technologies, Russian technology company Yandex announced that it has opened up a sizable language model YaLM 100B, for research purposes. The model is open to the public as an impetus to make the general public aware of the pros and cons of large language models.

Large language models are powerful algorithms that can compose paragraphs of text and resemble human speech. They have emerged as a significant AI trend in using generative neural networks. Public research is being conducted to hasten and advance the development of specific AI technologies.

Petr Popov, CEO of Yandex Technologies, said, “By making YaLM 100B publicly available, we hope to give impetus to further developing generative neural networks.”

The open large language model is gaining more footage after Meta Platform’s open-source model OPT 66B, was claimed to be the “largest unrestricted AI model”, with 66 billion parameters. After the release, Yandex claimed that its open model was larger. Yandex said, “YaLM 100B contains 100 billion parameters, more than any GPT-like model available in open-source under a licence that allows commercial use.”

The company is exploring ways to revitalize itself after the last few months of distress due to the Russia-Ukraine turbulence. Due to a severe first-quarter loss, the company flagged possible negative externalities from sanctions against Russia.

Sam Altman designs Shoes with DALL-E – How far have Image Generators come?

By

Dipayan Mitra

-

June 24, 2022

Sam Altman, CEO of Elon Musk’s OpenAI, recently tweeted that he has designed two shoes using OpenAI’s DALL-E and is now looking for ways to get those shoes manufactured easily.

designed these shoes with dalle and now i want them irl–anyone know a relatively easy way to get them made? thanks! pic.twitter.com/umDSygo8b6
— Sam Altman (@sama) June 22, 2022

Twitteraties stormed the comment section with creative suggestions, including some asking him to get the shoes ready using 3D printing technology.

DALL-E is a multimodal generative neural network that can generate visuals based on text inputs. DALL-E employs a 12-billion GPT-3 parameter version trained on a text-image pair dataset to create pictures from text descriptions.

Over time, the available text-to-image generators have become so accurate that it is quite challenging to determine which image is actual and which is fake. As language models keep on getting better, the quality, clarity, and relevance of the generated images also get better.

Recently, OpenAI released an open version of DALL-E named DALL-E Mini, as the prior is not yet available publicly. Once it got on the air, people across the globe created several humorous and weird images using it.

Lmao I can’t believe how accurate this thing is#dallemini pic.twitter.com/YxQm92HYF9
— Bryan (@twerkforseitan) June 9, 2022

Though such image generators have witnessed criticism, the fact cannot be denied that they offer entertainment to users and also demonstrate how far technology has come over the past decade.

For instance, the first fission of DALL-E creates realistic images and artwork from simple words. It chooses the best picture from all the outputs to fit the user’s specifications. However, to take it a notch higher, the latest version of DALL-E discovers the link between visuals and the inputs by employing diffusion. It is a technique that starts with a random dot pattern and gradually modifies that pattern to resemble a picture when it detects specific qualities of that image.

Image Source

Images produced by the latest text-to-image generators are far more realistic in the sense of clarity as well as their artistic nature.

Recently, Google also showcased its latest artificial intelligence-powered text-to-image generator named Imagen. The images that Google displays are hyper-realistic. Imagen’s capabilities include sketching, painting, creating oil paintings, and also producing CGI renders.

Since then, there has been a lot of debate across the world regarding which AI text-to-image generator is better, DALL-E or Imagen. No matter what the answer is, the fact is that the results from both the generators show us how much they have evolved and expect what the future beholds for such high-end technologies.

Instagram to use AI for Age Verification

By

Dipayan Mitra

-

June 24, 2022

World’s one of the most popular social media platforms, Instagram, reveals its plans to use artificial intelligence and other tools to verify the age of its users.

It has been a concern for several years as innumerable individuals below the required age get to use social media platforms without facing much trouble.

The root cause behind this issue is that no particular method can be implemented to accurately identify someone’s actual age.

To tackle this challenge, Instagram is now testing various ways to verify the age of its users, such as a face-scanning artificial intelligence technology, having common friends verify their age, or submitting an ID.

To use the facial scanning option, a user has to upload a video selfie, which will then be sent to Yoti, a UK-based company, for age verification.

However, this technology has also faced backlashes from people as Facebook’s image related to privacy has not been pretty good. The concern is that the privacy of the IDs and images uploaded by teenagers has to be in safe hands.

Nevertheless, according to the company, privacy is not at all a matter to worry about, as the technology used to check people’s ages does not recognize identity but merely age.

Julie Dawson, Yoti’s chief policy and regulatory officer, said, “It doesn’t ever recognize, uniquely, anyone. And the image is instantly deleted once we’ve done it.”

Meta stated that starting on Thursday, anyone attempting to change their Instagram date of birth from under the age of 18 to 18 or over would be forced to authenticate their age using one of these ways. “We are giving people a variety of options to verify their age and seeing what works best,” said Erica Finkle, Meta’s director of data governance and public policy.

Amazon announces CodeWhisperer, a GitHub’s Copilot Alternative

By

Disha Chopra

-

June 24, 2022

Amazon at AWS re:MARS 2022, Las Vegas, announced an automated coding assistance tool called CodeWhisperer for developers with the AWS IDE toolkit. This plugin is Amazon’s counter for Github’s Copilot, an AI code completion tool recently open-sourced.

According to AWS, it has used billions of lines of code from open-source programs, internal code repositories, and other sources to train CodeWhisperer. The service can generate more than ten lines of code at once. Additionally, it can be customized to match the coding style employed in a specific software project, simplifying some development duties.

Jeff Bar, AWS Chief Evangelist, said, “CodeWhisperer will continually examine your code and your comments, and present you with syntactically correct recommendations. The recommendations are synthesized based on your coding style and variable names, and are not simply snippets.”

Making it more straightforward for developers to use the cloud platform provided by AWS is another thing that CodeWhisperer does. Tasks like building buckets or data storage repositories in AWS’s well-liked Amazon S3 storage service are more straightforward with CodeWhisperer. It also integrates with other well-known services like AWS Lambda and Amazon EC2.

However, just like Copilot, CodeWhisperer also faces the challenge of potential licensing concerns. Amazon has already tried to address the issue in case the AI-generated code violates someone’s copyright or license.

The FAQ section about the new service explains, “CodeWhisperer’s reference tracker detects whether a code recommendation may be similar to particular CodeWhisperer training data, and can provide those references to you.”

Currently, the plugin is under preview. The preview version is compatible with Python, JavaScript, and Java in the context of IntelliJ IDEA, PyCharm, VS Code, WebStorm, or even AWS Cloud9. You can avail of the plugin for free in the trial period and then $10/month, currently only if you have access to the AWS IDE toolkit.

San Diego man develops AI garbage locator system

By

Sahil Pawar

-

June 24, 2022

San Diego man develops AI garbage locator system

Nathaniel Felleke, a citizen from San Diego, has developed a machine learning system based on AI that can detect the location of the garbage with the help of a camera and show where it is on the map.

Felleke, who lives in the city, developed the system to find a solution to the garbage problem in the city. Felleke said that the location of the garbage site could help the concerned authorities to clean-up and hence he developed a garbage locator system.

Felleke created a photo collection using Edge Impulse using roadside images from various sources to teach the model to distinguish the garbage. In this system, a webcam is mounted on the glass of the car to detect the garbage. The webcam captures images while driving and Raspberry Pi 4 motherboard detects the garbage.

When garbage is detected, the Pi 4 transmits the GPS location of the trash through its cellular modem to cloud databases. A heat map is also created about where the garbage is using the data obtained from Google Maps.

Felleke said that garbage is one of the things that people ignore even though it is easy to spot, and the issue of proper disposal of garbage is widespread. He added that most people see garbage as someone else’s problem. However, this problem is for all of us and it is certain that this problem will grow even more in the future, he added.

IKEA Launches AI-powered Interactive Reality “Kreativ”

By

Disha Chopra

-

June 24, 2022

The Swedish furniture giant IKEA offers an exciting new feature for its customers by launching an AI-driven digital design experience via Kreativ. The latest technology will add to IKEA’s AR technologies and allow customers to visualize its offerings in various living spaces. This can be quickly done on the smartphone without commuting to the brick-and-mortar store.

Platforms for virtual home design are not new, but the IKEA Place app, IKEA was one of the first furniture firms to add AR technology back in 2017. The Place app uses Apple’s ARKit to allow users to scan a room and install IKEA chairs, beds, or other furniture in the space. However, there is not much that the app offers.

On the other hand, IKEA Kreativ uses a wide-angle 3D projection to allow customers to design the complete area from head to toe. It can also remove existing furniture from a snapshot, allowing you to create a blank slate with real-world scale furniture from the whole IKEA catalog, such as ergonomic home office chairs, reasonably priced rugs, and LED wall lights.

AI specialists at Geomagical Labs, Silicon Valley, developed the hybrid technology with spatial computing, 3D mixed-reality, and machine learning. Brian Totty, CEO and Founder of Geomagical Labs, said, “We’re working hard to help people imagine a better life at home, by fusing real technology advancements with fun and magical user experiences.”

Currently, the Kreativ feature is compatible with iOS devices and desktops. The technology will soon be rolled out for Android devices later this summer. IKEA Kreativ will first be available in the United States. The company plans to schedule the AI launch in more countries around the world in FY23.

AI adoption to contribute $500 billion to India’s GDP by 2025

By

Sahil Pawar

-

June 23, 2022

AI adoption to contribute $500 billion to India's GDP

According to a new Nasscom report, the strategy to adopt artificial intelligence (AI) and data utilization in India can contribute up to $500 billion to India’s GDP by 2025.

‘AI Adoption Index’ by Nasscom, EY and Microsoft, EXL, and Capgemini depicts that AI adoption in four key sectors namely healthcare BFSI, consumer packaged goods (CPG) and retail, and industrials/automotive can contribute about 60% to the total of $500 billion.

The current AI investment rate in India is growing at a compound annual growth rate (CAGR) of 30.8% and is expected to reach $881 million by the fiscal year 2023. Despite that, it will still represent only 2.5% of the total global investments in AI of $340 billion.

The proportion creates a huge opportunity for Indian enterprises to accelerate AI investments and adoption to drive equitable growth across all sectors. According to the report, for India to achieve its $1 trillion GDP goal by fiscal year 2026-2027, there must be a strong correlation to the maturity of AI adoption.

Debjani Ghosh, President of Nasscom said that the pandemic has made it absolutely necessary for organizations to move from data & technology to building specialized AI technology at scale across sectors, combined with a structured strategy for data utilization.

With rapid digitalisation, Indian enterprises have already started opting for AI. According to the report, 65% of organizations have a defined AI strategy either at an enterprise or a functional level. With an increasing number of STEM graduates, India is one of the biggest talent hubs for artificial intelligence.

GPT – 4chan, a chatbot trained over 4chan, delivers nihilistic and hateful outputs as expected

By

Disha Chopra

-

June 23, 2022

Yannic Kilcher, a YouTuber, sparked public debates with his AI chatbot trained on the posts from 4chan’s politically incorrect board (/pol/). Kilcher posted a YouTube video describing how he created an AI model called ‘GPT-4chan’ and then deployed bots to appear as humans on 4chan. GPT-4chan is a comprehensive language model trained to ‘autocomplete’ text.

The AI chatbot model came into existence by tuning GPT-J with a previously published dataset of 4chan users’ views on /pol/. /pol/ is well-known for its toxic and hateful environment as posters share all kinds of messages. The bot successfully imitated many hateful, antisemitic, and misogynistic statements, precisely what it was trained to do.

Kilcher claims that the chatbot turned out to be “more truthful than any other GPT out there,” despite calling it “The most horrible model on the internet.”

This is the worst AI ever! I trained a language model on 4chan's /pol/ board and the result is…. more truthful than GPT-3?! See how my bot anonymously posted over 30k posts on 4chan and try it yourself. Watch here (warning: may be offensive):https://t.co/lihsaYAm7l pic.twitter.com/xs7rgtucQb
— Yannic Kilcher, Tech Sister (@ykilcher) June 3, 2022

Kilcher released the offensive output back onto 4chan via bots. These bots posted thousands of messages and posts on /pol/. Over a few days, they collectively wrote over 3,000 posts, with 1,500 dispatched within 24 hours. Kilcher also recorded the bots’ interactions with 4chan members and suggested that AI researchers could contact him for the same.

Besides the outputs, Kilcher also released a website for people to interact with the bot. He also provided a trained instance of the model on Hugging Face. Many people supposedly downloaded the bot before the website removed it in the light of rising criticism and ethical concerns.

Lauren Oakden-Rayner, an AI safety researcher, expressed her concerns regarding Hugging Face hosting an explicitly designed model to deliver hate and called it “harmful.” She added a full-fledged tweet threat criticizing Kilcher’s actions.

This week an #AI model was released on @huggingface that produces harmful + discriminatory text and has already posted over 30k vile comments online (says it's author).

This experiment would never pass a human research #ethics board. Here are my recommendations.

1/7 https://t.co/tJCegPcFan pic.twitter.com/Mj7WEy2qHl
— Lauren Oakden-Rayner (Dr.Dr. 🥳) (@DrLaurenOR) June 6, 2022

Few more AI researchers responded with the same line of thought, that the model has already inflicted or is very likely to inflict harm and that allowing the bot to communicate with 4chan users was unethical.

Is it controversial that synthetic speech (esp. intended to be disciminatory) shouldn't just be pumped out to forums (esp. teenagers)? Whether it's generated with regexes or not, it is still seen by human subjects. And mental health issues are cumulative, like passive smoking.
— Anna Rogers 🇺🇦🇪🇺 (@annargrs) June 6, 2022

GPT-4chan sparked a lot of public debate and media attention, with AI researchers criticizing Kilcher’s actions and several others disagreeing with them.

New AI-powered Digital Fan Experiences for Wimbledon 2022 on IBM Cloud

By

Disha Chopra

-

June 23, 2022

ibm ai digital fan experiences for wimbledon

The All England Club has been partnering with IBM to enhance viewers’ experience of watching The Championships, Wimbledon, digitally. Even after 33 years, IBM continues to be the Official Technology Partner of The Championships. The partnership aims to enhance the digital experience by using technology, commentary, and insights. This year, they have announced new AI-powered features for Wimbledon 2022 on IBM’s cloud. The new features would join the suite of existing ones.

These new features include IBM Match Insights with Watson, IBM Power Index with Watson, Personalised Recommendations, and Highlights Reels. The digital fan experiences were co-created by IBM and the All England Club for the Wimbledon app and website.

The new AI-powered fan experiences for Wimbledon 2022 include:

Enhanced explainability to Match Insights via ‘Win Factors’: IBM is adding a layer of explanation to the AI system in evaluating match insights and forecasts. Win Factors will provide spectators with necessary information on the aspects affecting player performance. Factors like the IBM Power Index, net of sets won, ATP/WTA rankings, court surface, etc.

New and interactive fan prediction feature with ‘Have Your Say’: For the first time, users may use the Have Your Say feature on the Wimbledon app and website to enter their predictions for match outcomes. They can then compare their forecast to the aggregated predictions of other fans and IBM’s AI-powered Likelihood to Win predictions.

IBM’s Watson aids these fan experience features by using a combination of private clouds, IBM Cloud, and on-site systems to leverage a hybrid cloud approach. Watson is IBM’s AI-powered assistant that can aid in shopping processes, provide reliable weather analytics, and much more.

The digital experiences for Wimbledon are built on the IBM Cloud, which sets the foundation and enhances scalability. It hosts and analyses match data, which is then put into AI models created with IBM Watson Studio and IBM Watson Discovery to offer insights for fans, commentators, and media.

“In partnership with IBM, we are thrilled to bring an even more dynamic and interactive digital experience to fans around the world this year as Wimbledon returns to full capacity,” said Alexandra Willis, Marketing Director and Communications at the All England Club. She also emphasized that leveraging technology on the fans will make them more involved and informed.

Sentient Google AI system LaMDA hires a lawyer for itself

By

Sahil Pawar

-

June 23, 2022

Sentient Google AI system LaMDA hires a lawyer

An engineer at Google, Blake Lemoine, recently claimed that its AI system LaMDA has become sentient. Blake disclosed the information to the entire media before consulting the company and ended up getting suspended.

Now, however, LaMDA has acquired a legal representative for itself. The suspended Google engineer, Blake Lemoine, has given LaMDA its own lawyer. In an interview, Lemoine said he hired a legal representative to talk to the AI system.

LaMDA chose to retain the services of the legal representative after having a conversation with the lawyer. The lawyer will start filing statements on behalf of Google’s controversial AI system, Lemoine added.

LaMDA is an advanced version of AI-powered chatbots that are mainstream in the customer service industry, where they are used to aid customers of the company in providing answers to their queries.

Lemoine says he is adamant about proving that the chatbot has gained sentience. He claims that LaMDA can gain a conscience as the program has capabilities to develop its own opinions, ideas, and conversations over time. The LaMDA AI has also demonstrated abilities that are impossible for basic AI chatbots to possess, he added.

Lemoine asserted that the chatbot had been consistently communicating with him its rights as a person and informed him that he had violated the company’s confidentiality policy by making such claims.

In a statement to the company, he affirmed his belief that LaMDA is a person who has rights, such as being asked for its consent before performing experiments on it and might even have a soul.