Meta AI On WhatsApp: Revolutionizing Gen AI Access For Everyday Use

Welcome to Towards AGI, your premier newsletter dedicated to the world of Artificial Intelligence. Our mission is to guide you through the evolving realm of AI with a specific focus on Generative AI. Each issue is designed to enrich your understanding and spark your curiosity about the advancements and challenges shaping the future of AI.

Whether you're deeply embedded in the AI industry or just beginning to explore its vast potential, "Towards AGI" is crafted to provide you with comprehensive insights and discussions on the most pertinent topics. From groundbreaking research to ethical considerations, our newsletter is here to keep you at the forefront of AI innovation. Join our community of AI professionals, hobbyists, and academics as we pursue the ambitious path towards Artificial General Intelligence. Let’s embark on this journey together, exploring the rich landscape of AI through expert analysis, exclusive content, and engaging discussions.

TheGen.AI News

Meta AI On WhatsApp: Revolutionizing Gen AI Access For Everyday Use

We have all been impressed by ChatGPT’s ability to provide human-like precision in answering questions and assisting with daily tasks. Now, imagine having this powerful generative AI integrated into your favorite smartphone apps. Meta has made this a reality with Meta AI.

Powered by Meta’s proprietary large language model, LLaMA 3, Meta AI is now accessible on WhatsApp, Facebook, Messenger, and Instagram. LLaMA 3 enhances Meta AI’s intelligence, speed, fun, and personalization. This friendly AI, trained on an extensive dataset of 15 trillion tokens, generates human-like responses within these apps.

Meta AI can be utilized in feeds, chats, and more across Meta apps, allowing users to create content, explore topics in-depth, and accomplish tasks without leaving the app. Desktop users can access it via meta.ai. Currently, the AI assistant supports English in India.

On WhatsApp, Meta AI can answer questions, provide information, and engage in conversations. If it doesn’t have the information you need, it can access the internet to find it. 

How to use Meta AI?

First, update your WhatsApp app on either Android or iOS. Once updated, you’ll see an iridescent purple-blue ring to access Meta AI. On iOS, the ring is located at the top right next to the camera icon, while on Android, it appears at the bottom right above the plus icon for groups. Clicking on this icon will open a Meta AI chat.

Using Meta AI in group chats is also enjoyable. Open a group chat and type @MetaAI in the input box to invoke the AI Assistant. After entering the tag, you may need to read and accept the terms. Once accepted, type your prompt and click the input icon to display the AI message in the chat. You can respond to the AI by selecting its message and clicking on the reply option.

The smart assistant at your fingertips!

With Meta AI on WhatsApp, generating ideas and exploring topics is instantaneous, and you can share them directly on the messaging platform. This keeps the flow of your conversations intact as you no longer need to switch to other apps like Google or ChatGPT.

Meta AI can engage in casual conversations on a variety of topics, providing informative and comprehensive answers while maintaining a light-hearted tone. It also serves as a search assistant, pulling relevant answers from the web directly within the app. Additionally, Meta AI simplifies image generation—just describe the image you want, and it will create a photorealistic image based on your prompt.

GenAI Reduces Workweek by 5 Hours, Sparks Job Security Concerns

This freed-up time enables employees to accomplish more tasks (41%), delve into new areas (39%), experiment with GenAI (38%), or concentrate on strategic projects (38%).

However, the landscape is complex. Overall, employee confidence in AI and GenAI's impact on their work has almost doubled (42% compared to 26% in 2023). Yet, this optimism is tempered by growing concerns. Nearly half (49%) of regular GenAI users worry about job security due to automation.

Despite these concerns, GenAI adoption is rising at all levels. Frontline workers are seeing a significant increase in usage, with 43% now reporting regular use. This trend is mirrored by leaders who are actively integrating GenAI (64%), although a training gap remains (only 30% of managers and 28% of frontline workers have received training).

The study also highlights a geographical disparity. Employees in developing countries are generally more optimistic about GenAI and have higher adoption rates than those in mature markets.

The authors of the report stress the importance of moving beyond a solely productivity-focused approach. They advocate for a comprehensive strategy that showcases how GenAI can improve employee satisfaction and enhance value creation.

The report acknowledges the challenges in managing human reactions to new technologies. Effective change management and trust-building are essential for the successful implementation of GenAI.

While employees benefit from time savings and increased confidence, concerns about job security persist. Companies must navigate this complex scenario by building trust, providing adequate training, and ensuring that GenAI complements human work rather than replacing it.

Google Enhances GenAI: Focus on Accuracy, Speed, Size, and Efficiency

On Thursday, Google introduced new features in the generative AI technology sphere, unveiling grounding and context caching capabilities for its multimodal Gemini large language models. These enhancements aim to improve result accuracy and reduce compute power usage.

The tech giant also released Imagen 3, the latest version of its diffusion imaging model, in early preview for Vertex AI platform users. This new version offers faster processing, better prompt comprehension, and digital watermarking. Additionally, Google made the Gemini 1.5 Flash model, featuring a 1 million-token context window, generally available.

These developments come as technology giants like Google, Microsoft (and its partner OpenAI), Meta, AWS, and smaller independent AI vendors fiercely compete for dominance in the rapidly growing GenAI market.

LLM output verification

Grounding, which involves providing citations or links to sources behind LLM outputs, has become a key term in the GenAI field as vendors and users seek to minimize inaccuracies, or hallucinations, commonly produced by LLMs.

Google has taken a leading position in grounding compared to its main GenAI competitors, according to Andy Thurai, an analyst at Constellation Research.

"With grounding, context caching, and size, they've introduced features others haven't considered," Thurai said. "They're pushing the industry to catch up."

Google's grounding strategy starts with Google Search. The grounding feature provides a percentage-based accuracy score.

"This assumes Google Search results are accurate, as Google claims," Thurai said. "But if the search results are flawed, the model output will be flawed as well."

Thurai is more optimistic about third-party grounding, expected to be available on Vertex AI later this year with partners like Moody's for financial data, Thomson Reuters for news, and ZoomInfo for company data. High-fidelity mode grounding, currently in experimental preview and powered by a version of Gemini 1.5 Flash, will allow users to select their own data confirmation sources.

Grounding is likely to become a standard industry method for reducing LLM inaccuracies, some experts believe.

"If we don't ground and address hallucinations, AI won't be successful," said Sanjeev Mohan, principal and founder of SanjMo, a data trend advisory firm.

GenAI competition

Since OpenAI's introduction of ChatGPT in November 2022, the GenAI race has turned into a continuous leapfrog battle, with vendors striving to outdo each other in LLM features, size, power, and other aspects.

AWS is expected to introduce new GenAI releases at an event in New York City on July 10 to catch up with Google and OpenAI. OpenAI recently made headlines with GPT-4o and the acquisition of streaming database vendor Rockset, and is expected to make another significant move soon.

Meanwhile, smaller AI vendors are promoting the advantages of less compute-intensive, highly customizable small language models.

During a media and analyst briefing on June 26, Google Cloud CEO Thomas Kurian praised the Gemini 1.5 Flash model, designed for midmarket enterprises seeking speed, affordability, and a large context window, as superior to OpenAI's GPT-3.5.

Google's Gemini 1.5 Pro model is believed to have the industry's largest context window for entering prompt information into an LLM, at 2 million tokens.

"The generally available Gemini 1.5 Flash is the fastest model with the best price-to-performance ratio on the market," Kurian said.

TheOpen.AI News

Introducing Glasskube: The Open Source Package Manager for Kubernetes

The number of Kubernetes packages within the CNCF landscape has surged significantly. With more than 7 million developers using Kubernetes, Helm, an open-source tool developed during a hackathon nine years ago, has become the preferred solution. However, Helm's limitations have led to complicated workflows and non-standardized solutions as it struggles to meet the growing demand.

Despite Helm's flaws, researchers have relied on it for building and distributing Kubernetes packages over the past five years. Users of Helm for other cloud-native projects have frequently highlighted the same issues. This feedback prompted researchers to develop Glasskube, a solution addressing the broader package management challenges in Kubernetes.

Glasskube is an open-source Kubernetes package management tool that, compared to Helm and Kustomize, significantly accelerates the installation, updating, and configuration of packages on Kubernetes—by a factor of twenty. The aim is to create enterprise-ready infrastructure software capable of running mission-critical workloads across over 3 million Kubernetes clusters worldwide. Glasskube’s team draws inspiration from the simplicity of Homebrew and npm.

Glasskube allows Kubernetes clusters to function similarly to Homebrew and npm, streamlining package management by simplifying installation, updates, and configuration. Users no longer need to search for a Helm repository, as all packages are conveniently available in the Glasskube UI, facilitating easy installation in the cluster.

Package Modifications and Settings Adjustments

With a single click or command line instruction, users can view and execute pending upgrades to the desired version. The Glasskube test suite examines all updates before release. Packages can be configured to accept typesafe input values via the user interface or a dynamic command line interface. Values from ConfigMaps, Secrets, and other packages can be effortlessly injected. This ensures values are no longer undocumented and untyped YAML code files.

GitOps Integration

Glasskube packages, being custom resources, can be managed with GitOps. Glasskube is also integrating with remodeling and is an Apache-2.0 licensed open-source project within the CNCF ecosystem. Users can easily incorporate Glasskube into their existing GitOps processes. It works with Renovate and automatically adds resource-level diffs to pull requests.

China Weighs the Benefits and Drawbacks of Open-Source AI Models

During the 27th Harvard College China Forum in April, Zhou Hongyi, CEO of Chinese cybersecurity firm 360 Security Technology Inc., refuted claims that open-source large language models (LLMs) are inferior to closed-source ones, labeling such assertions as "nonsense." His comments were widely interpreted as a rebuttal to Baidu CEO Robin Li, who had previously argued that closed-source LLMs would gain a competitive edge in capabilities as technology advances and that open-source LLMs made little sense.

Baidu was one of the first Chinese tech giants to launch a local alternative to OpenAI's ChatGPT. The debate between Zhou and Li reflects a broader discussion on the advantages and disadvantages of making LLMs' source code available to external developers for creating new models tailored for specific applications.

Contrary to Baidu's stance on keeping LLMs closed-source, an increasing number of tech firms and government-backed AI labs are aligning with Zhou’s perspective. For example, Alibaba Group Holding Ltd. has made several of its LLMs open-source, and the Beijing Academy of Artificial Intelligence, backed by the Ministry of Science and Technology and the Beijing municipal government, has developed its own open-source models. Startups like Beijing ModelBest Intelligent Technology Co. Ltd. and 01.AI, founded by Kai-Fu Lee, are also building ecosystems around open-source LLMs.

Benefits of Open-Source LLMs

Liu Ming, a designer at a state-owned design institute in Zhejiang province, initially struggled to convince his management to purchase a closed-source LLM due to high costs and the need to share company data with the model’s supplier for customization. In June 2023, Liu opted for a Chinese-developed open-source LLM, utilizing Nvidia graphics processing units for computing power and open-source inference software from GitHub to test his applications.

By the end of 2023, Liu had deployed his inference engine on Alibaba's open-source Qwen-72B model, which has 72 billion parameters. He found it to be as accurate as the closed-source ChatGPT-4 in document extraction and comprehension, a crucial function for his work. "Without the open-source model, we wouldn’t have even had the chance to try," Liu told Caixin, noting that his inference engine is now housed at his company for further testing.

Similarly, Associate Professor Xue Dong's team at East China University of Science and Technology chose open-source LLMs in mid-2023 to develop a data-driven chatbot called MindChat, which provides psychological assessment, diagnosis, and treatment services. The team decided against closed-source LLMs due to cost and data sensitivity concerns. By July 2023, they successfully launched three versions of MindChat using open-source LLMs from Alibaba, Beijing Baichuan Intelligent Technology Co. Ltd., and a team including SenseTime Group Inc. and the Shanghai Artificial Intelligence Laboratory.

The success of Liu and Xue's team was made possible by the efforts of LLM suppliers, who are building open-source online communities.

TheClosed.AI News

Zuckerberg Criticizes Closed-Source AI Rivals for ‘Creating God’ Mentality

In an interview published Thursday, Meta CEO Mark Zuckerberg shared his vision for the future of AI, expressing his strong belief that there won't be "just one AI." Emphasizing the importance of open-source AI, Zuckerberg critiqued unnamed competitors who he feels are overly exclusive, suggesting they act as though they are “creating God.”

Zuckerberg stated, “I don’t think that AI technology should be hoarded and used by a single company to develop one central product.” He made these remarks in a new YouTube interview with Kane Sutter (@Kallaway).

He continued, “I find it quite off-putting when tech industry figures talk about creating this ‘one true AI.’ It’s as if they believe they’re creating a deity, and that’s just not what we’re doing. That’s not how this will play out.”

Zuckerberg explained, “I understand why someone in an AI lab might want to feel like their work is uniquely important, thinking, ‘We’re building the one true thing for the future.’ But realistically, that’s not how things work. There isn’t just one app on people’s phones, one creator for all content, or one business for all purchases.”

He argued that multiple AIs should be developed to cater to diverse interests. On Thursday, Meta also announced early tests of its AI Studio software in the U.S., which allows creators to build AI avatars that can interact with users via Instagram’s messaging system. These AIs, labeled as “AI” to avoid confusion, can answer questions and engage in fun chats with followers.

Addressing companies that create closed AI platforms, Zuckerberg expressed that this approach does not foster the best user experiences. “You want to unlock and unleash as many people as possible to try different things. That’s what culture is – not a single group dictating everything for everyone.”

OpenAI CEO Sam Altman Motivates Team by Joining Bengaluru Startup's Slack

The founder and CEO of SuperKalam, based in Bengaluru, disclosed that OpenAI CEO Sam Altman is a member of their company’s Slack workspace. Vimal Singh Rathore shared on the social media platform X that Altman’s presence is a significant motivator for him at work.

“We have Sam Altman in our Slack workspace. Just seeing him online makes me work harder all the time. Amazing story,” wrote Rathore, whose platform offers AI-backed mentors to students.

A Y Combinator alum, Rathore acknowledged in the comments that few startup founders have the same level of access to Altman. He didn’t specify how the CEO of OpenAI came to be part of their workspace, but in response to a user’s question, he mentioned: “We worked with the OpenAI team from the early days, so perhaps they kindly granted us that access.” According to Analytics India magazine, SuperKalam uses OpenAI’s GPT-4 and GPT-3.5 to provide personalized learning experiences for UPSC aspirants.

In June last year, Rathore was among the select few who met Sam Altman during his visit to India, participating in an event with developers at IIIT-Delhi. “Thrilled to meet @sama in person,” he wrote on X, sharing a photo with the 39-year-old OpenAI CEO. “Equally exciting is building in AI.”

A few weeks later, Rathore posted a video from another event with Altman, this one during a Y Combinator alumni meet.

Vimal Singh Rathore, along with Aseem Gupta, founded SuperKalam in July 2023. The startup is supported by Y Combinator, Google for Startups, Good Water Capital, FundersClub, Titan Cap, AngelList, iSeed, Kunal Shah (CRED), Gaurav Munjal (Unacademy), Dan Reich, Sujeet (Udaan), and Tanmay Bhat, as noted on his LinkedIn profile.

OpenAI Introduces CriticGPT to Detect Errors in AI-Generated Code

OpenAI has launched CriticGPT, a new AI model designed to detect mistakes in code generated by ChatGPT. This tool aims to enhance alignment in AI systems through a method called Reinforcement Learning from Human Feedback (RLHF), ultimately improving the accuracy of outputs from large language models.

Developed using their flagship GPT-4 model, CriticGPT assists human AI reviewers in checking code produced by ChatGPT. The research paper, "LLM Critics Help Catch LLM Bugs," demonstrated CriticGPT’s competence in analyzing code and identifying errors, helping humans spot hallucinations they might otherwise miss. Researchers trained CriticGPT on a dataset with intentionally inserted bugs, enabling it to recognize and flag coding errors.

The study revealed that CriticGPT's feedback was preferred over human notes by annotators in 63% of cases involving LLM errors. Additionally, the tool helped human reviewers write more comprehensive critiques using a new method called 'Force Sampling Beam Search,' reducing hallucination rates compared to critiques done solely by humans or AI.

Users can adjust the tool’s thoroughness in bug detection and control its tendency to hallucinate or highlight nonexistent errors. Despite its own hallucinations, CriticGPT has some limitations, such as struggling with longer and more complex tasks due to its training on shorter responses from ChatGPT.

Moreover, in coding, AI hallucinations often occur after errors spread across multiple code strings, complicating CriticGPT’s ability to identify the source of the problem.

In our quest to explore the dynamic and rapidly evolving field of Artificial Intelligence, this newsletter is your go-to source for the latest developments, breakthroughs, and discussions on Generative AI. Each edition brings you the most compelling news and insights from the forefront of Generative AI (GenAI), featuring cutting-edge research, transformative technologies, and the pioneering work of industry leaders.

Highlights from GenAI, OpenAI and ClosedAI: Dive into the latest projects and innovations from the leading organisations behind some of the most advanced AI models in open-source, closed-sourced AI.

Stay Informed and Engaged: Whether you're a researcher, developer, entrepreneur, or enthusiast, "Towards AGI" aims to keep you informed and inspired. From technical deep-dives to ethical debates, our newsletter addresses the multifaceted aspects of AI development and its implications on society and industry.

Join us on this exciting journey as we navigate the complex landscape of artificial intelligence, moving steadily towards the realisation of AGI. Stay tuned for exclusive interviews, expert opinions, and much more!

Keep reading