Microsoft has announced a significant leadership change as Pavan Davuluri takes the helm as Corporate Product Chief, succeeding Panos Panay. This transition places Davuluri in a pivotal position as he will be reporting directly to Microsoft’s CEO, Satya Nadella.
In parallel, Panay is set to join Amazon to head the unit responsible for the company’s Alexa and Echo products, stepping into the role previously held by David Limp, who plans to retire this year. Panos Panay, after nearly two decades at Microsoft, during which he played a crucial role in creating the Surface line of computers and overseeing the launch of Windows 11, steps down.
This leadership shift comes as Microsoft holds its product launch event, where new additions to the Surface lineup are anticipated, alongside the unveiling of the company’s latest AI innovations.
Pavan Davuluri, an alumnus of the Indian Institute of Technology (IIT) Madras and the University of Maryland, has been an integral part of Microsoft since the early stages of his career, starting in 2001 as a Reliability Component Manager.
His most recent role before this promotion was as Corporate Vice President, overseeing Windows Silicon and Systems Integration. In his new role, Davuluri is tasked with leading a team focused on the development of silicon and systems.
Davuluri will play a key role in Microsoft’s emphasis on integrating generative AI into its wide array of products, spanning cloud services, search, and productivity software. This highlights the company’s commitment to advancing technology in an increasingly AI-driven world.
GitHub has announced that it has expanded access to its programming-centric chatbot, Copilot Chat, to all current GitHub Copilot for Individual subscribers using Visual Studio and VS Code. Previously, Copilot Chat was exclusively available to organizations with a Copilot for Business subscription. This move allows individual users to access the chatbot’s features at no additional cost, provided they have the $10/month Copilot for Individual subscription.
Copilot Chat resides in the integrated development environment (IDE) sidebar, facilitating multi turn conversations about coding, including discussions about the code currently being worked on. GitHub emphasizes that Copilot Chat’s ability to understand and provide context-specific assistance sets it apart from general-purpose chat assistants.
Shuyin Zhao, VP of Product Management at GitHub, believes that the integration of Copilot Chat with GitHub Copilot creates a powerful AI assistant, capable of helping developers work efficiently in their preferred natural language. This, in turn, reduces repetitive tasks and positions natural language as a universal programming language.
Common use cases for Copilot Chat include real-time guidance on best practices, tailored code-related tips and solutions, and assistance with code analysis and security issue resolution, all within the IDE, eliminating the need to switch between tools.
GitHub’s goal is to promote “natural language as a new universal programming language” to democratize software development. This aligns with the company’s recent emphasis on enabling developers to work more efficiently and effectively.
OpenAI has unveiled the third iteration of its generative AI visual art platform, DALL-E, with enhanced features and a focus on safety. DALL-E 3, known for converting text prompts into images, has made significant strides in understanding context compared to its predecessor.
One notable addition to DALL-E 3 is its integration with ChatGPT, allowing users to generate prompts more easily. Users can request ChatGPT to provide a detailed prompt, and the chatbot will craft a paragraph that DALL-E 3 can work with, simplifying the creative process.
The new version of DALL-E will first be available to ChatGPT Plus and ChatGPT Enterprise users in October, followed by research labs and its API service in the fall. OpenAI has not provided a specific timeline for a free public release.
DALL-E’s journey began in January 2021, preceding other text-to-image generative AI art platforms. However, with the release of DALL-E 2 in 2022, OpenAI faced criticism for issues like generating explicit images and displaying bias. The platform was initially put on a waitlist but later made accessible to the public.
OpenAI has emphasized robust safety measures in DALL-E 3 to prevent the creation of inappropriate or harmful images. External red team testing and input classifiers have been employed to enhance safety. Additionally, DALL-E 3 will not generate images of public figures unless explicitly mentioned in the prompt.
While OpenAI expresses confidence in the safety measures, it acknowledges that the model continues to improve and is not flawless. DALL-E 3 has also been trained not to mimic the styles of living artists, unlike DALL-E 2, which could replicate certain artistic styles when prompted.
Microsoft has taken immediate action to address a significant security incident that led to the exposure of a staggering 38 terabytes of private data. The breach was identified within the company’s AI GitHub repository and is believed to have occurred inadvertently during the publication of open-source training data, according to Wiz, a cybersecurity research team.
This breach included a backup from the workstations of two former employees, containing sensitive information like secrets, keys, passwords, and over 30,000 internal Teams messages.
The repository, named “robust-models-transfer,” has been made inaccessible. Before its takedown, it housed source code and machine learning models related to a 2020 research paper titled “Do Adversarially Robust ImageNet Models Transfer Better?”
Wiz’s report revealed that the breach resulted from an overly permissive Shared Access Signature (SAS) token, an Azure feature that facilitates data sharing in a challenging-to-track and revoke manner. Specifically, the repository’s README.md file inadvertently allowed developers to download models from an Azure Storage URL that also granted access to the entire storage account, exposing additional private data.
To address this issue, Microsoft promptly revoked the SAS token and blocked external access to the storage account. The company’s investigation found no unauthorized exposure of customer data and confirmed that no other internal services were compromised.
The company also identified a bug in its scanning system that led to the false flagging of the specific SAS URL in the repository. To enhance future security measures, Microsoft has expanded its secret scanning service to include SAS tokens with overly permissive settings.
Conversation Intelligence Systems, a vital component of the modern tech landscape, play a pivotal role in deciphering and enhancing human-machine interactions. These systems employ advanced AI and natural language processing to comprehend the nuances of conversations, whether they occur between humans or involve machines. With the rising importance of personalized user experiences, startups are increasingly investing in Conversation Intelligence to refine customer support, automate tasks, and bridge the communication gap between users and technology. One such pioneering startup is Mihup.ai.
Mihup.ai
Mihup.ai is a pioneering startup at the forefront of natural language processing and artificial intelligence, poised to revolutionize the way we interact with technology. It is a Conversation Intelligence platform that helps businesses boost their contact center performance. It is built on proprietary ASR technology that offers the best blend of accuracy, speed, and cost-effectiveness. Founded on the principle that human-machine communication should be intuitive and seamless, Mihup.ai has harnessed cutting-edge AI technologies to create a dynamic platform that understands and responds to natural language with unprecedented accuracy and efficiency. The company’s products and services are used by businesses of all sizes, across a variety of industries, including telecommunications, financial services, healthcare, and retail.
Vision to Democratize AI
Mihup.ai was founded in 2016 by Tapan Barman and Biplab Chakraborty. Currently, Barman and Chakraborty are the CEO and COO of the company, respectively.
The initial idea and foundation of Mihup.ai centered around building the most accurate voice interface for the next billion users, particularly in regions where technology adoption was on the rise but faced barriers due to the complexity of the man-machine interface. In 2016, amidst a less prevalent AI landscape, the founders recognized the untapped potential of reaching users in these regions. They identified a significant challenge: despite the growth in technology, user experience had not evolved in line with it.
Priyanka Kamdar, Head of Growth at Mihup.ai said “People, especially in India, had reservations about interacting with machines due to language proficiency concerns and a fear of judgment. Mihup’s vision was to address these barriers by offering a display-less interaction mode for various types of communication, whether between humans and machines, machines and machines, or humans and humans. The company aimed to establish a protocol for adaptive conversations, bridging language and dialect gaps to cater to diverse user preferences.”
In nutshell, Mihup sought to create a platform where developers could leverage their technological prowess to build products tailored to the specific requirements of internet users worldwide, transcending language and cultural differences.
Priyanka Kamdar, a representative of Mihup.ai, explained the company’s three main products and services as follows:
Mihup Interaction Analytics
According to Kamdar, this product is designed for the contact center industry. It offers post-call analytics, addressing the challenges of manual quality assurance processes in call centers. While traditional methods could only sample a small percentage of calls, Mihup Interaction Analytics enables the analysis of 100% of interactions between customers and agents. It captures the voice of the customer, monitors agent performance, and generates reports for actionable insights. This ensures that valuable business insights are derived from all interactions, improving customer service and decision-making.
Mihup Agent Assist
Also serving the contact center industry, Mihup Agent Assist is a real-time assistant for call center agents. It operates during live customer-agent interactions, providing agents with guidance based on customer cues. For example, it can help agents adjust their tone when a customer is angry or suggest appropriate solutions. “This intelligent assistant enhances the agent’s performance and ensures a successful call outcome by bridging the gap between human-to-human and human-to-machine conversations,” said Kamdar.
Mihup AVA
Mihup’s third product is Mihup Automated Voice Assistant (AVA), a voice bot tailored for the automotive industry. It resides within a vehicle’s infotainment system and operates offline as well as online, making it suitable for use with and without an internet connection. This voice bot enhances the driving experience by performing in-vehicle functions and providing assistance to users. It is designed to work in multiple languages, including Hindi, English, Bengali, Tamil, Marathi, Haryanvi, and Malwari, making it versatile and user-friendly for a wide range of scenarios and users.
The Future of Customer Service with AI
Mihup.ai utilizes artificial intelligence extensively in its products and services. The company’s AI technology encompasses a wide range of components that contribute to the functionality of their offerings. For instance, in their voice bot product, AI components such as speech recognition, speech-to-text conversion, dialogue management, advanced natural language processing (NLP), and text-to-speech conversion are crucial. Mihup’s Interaction Analytics and Agent Assist uses speech recognition technology to transcribe customer conversations. This allows businesses to analyze customer interactions and identify areas for improvement. Mihup.ai uses natural language processing (NLP) to understand the meaning of customer conversations. The platform also offers a chatbot platform that businesses can use to automate customer interactions.
“Importantly, Mihup.ai has developed each of these components in-house, without relying on third-party solutions. This approach allows them to have complete control over their AI models and technology stack, ensuring that their products are tailored to their specific needs and requirements,” said Kamdar.
Unique Approach to Conversational AI
Mihup.ai differentiates itself from other startups in the conversational AI space through several key aspects. Firstly, their proprietary tech stack is entirely developed in-house, granting them full control over their product direction, security, and scalability. “Unlike language-based models, Mihup.ai’s approach is vocabulary-based and even goes beyond words to consider phonetic imprints and modulation, allowing them to detect not just keywords but also the context of interactions,” explained Kamdar.
The company’s focus is on speech-to-intent and meaning, addressing the challenge where machines are getting smarter but users struggle to use them effectively. “Mihup.ai aims to provide a horizontal platform that spans various industries, including contact centers, automotive, and government services, enabling developers to create tailored solutions,” she said. Their unique approach and adaptability position them as pioneers in the field of conversational AI, according to her.
Mihup.ai’s Funding Journey
“Mihup.ai’s funding journey began in 2016 when the company was formally established. However, securing initial funding for a deep tech startup like Mihup was challenging, as the concept was ahead of its time in India. Despite facing multiple rejections, Mihup.ai found its early investors in Accel and IdeaSpring Capital,” said Kamdar.
Mihup.ai has raised a total of $5 million in funding over 6 rounds. The funding has been used to expand Mihup.ai’s product offerings, grow its team, and expand into new markets, according to Kamdar.
Vision for the Future
“Mihup.ai’s future plans revolve around their larger vision, which is to become the interpreter for all types of conversations in the man-machine world,” says Kamdar. While they have short-term tactical goals related to product advancement and upgradation, their ultimate objective is to develop an interface that can handle verbal, non-verbal, written, or visual interactions between humans and machines.They foresee Mihup.ai playing a significant role in the IoT world of tomorrow, potentially becoming a key player in chatty operating systems, aligning with their vision of bridging the gap between human and machine communication.
Researchers at Tsinghua University, TAL AI Lab, and Zhipu.AI have introduced MathGLM, a robust mathematical model designed to handle a wide range of complex arithmetic operations. MathGLM’s performance rivals industry-leading models like GPT-4, excelling in tasks such as addition, subtraction, multiplication, division, and exponentiation.
What sets MathGLM apart is its versatility, as it effortlessly manages various number types, including decimals, fractions, percentages, and negative numbers. To train MathGLM, the team utilized the Ape210K dataset, a comprehensive collection of math word problems from across the internet.
The Ape210K dataset compiles math word problems sourced from the internet, serving as a rich repository of diverse mathematical challenges. It proves invaluable for training MathGLM due to its wide array of problem types and complexities.
Unlike traditional datasets, Ape210K contains explicitly calculated answers. However, to address potential limitations, the researchers employed a step-by-step approach to reconstruct the dataset, enhancing MathGLM’s ability to solve math word problems.
MathGLM’s unique strength lies in its ability to break down complex arithmetic calculations into sequential phases. This method significantly improved accuracy, with MathGLM outperforming GPT-4 by an impressive 42.29% when fine-tuned on the original dataset.
By dissecting arithmetic word problems into manageable steps, MathGLM demonstrates superior mathematical reasoning, learning underlying calculation principles, and delivering more dependable results. These discoveries profoundly challenge the traditional belief that LLMs are incapable of tackling complex arithmetic tasks, highlighting their remarkable capacity for advanced mathematical reasoning.
IBM has unveiled an ambitious initiative aimed at training a staggering two million learners in artificial intelligence (AI) by the end of 2026. This initiative is driven by the goal of addressing the global AI skills gap, with a specific focus on underrepresented communities.
As part of this initiative, IBM has introduced a new generative AI coursework accessible through IBM SkillsBuild, designed to provide AI education developed by IBM experts to learners worldwide. The curriculum covers a wide range of topics, including Prompt-Writing, Getting Started with Machine Learning, Improving Customer Service with AI, and Generative AI in Action. These courses are offered free of charge.
IBM is enhancing the learning experience with AI-enabled features, including improved chatbots to support learners and personalized learning paths tailored to individual preferences and experiences. Upon course completion, participants can earn IBM-branded digital credentials recognized by potential employers.
IBM is collaborating with universities worldwide to bolster AI capacity, providing faculty with access to IBM-led training, courseware, and immersive skill experiences. Additionally, students will benefit from flexible and adaptable resources, including free online courses on generative AI and Red Hat open-source technologies.
This initiative aligns with IBM’s broader commitment to upskill 30 million individuals by 2030 and addresses the pressing skills gap in the application of AI and digitalization across various industries. IBM’s endeavor is in line with similar initiatives by tech giants like MongoDB and Infosys, who are also offering complimentary courses to empower individuals through self-learning, opening doors to employment opportunities based on performance in tasks and certifications.
Google DeepMind has introduced an AI program, AlphaMissense, capable of discerning whether millions of genetic mutations pose harm or are benign, aiming to expedite research and diagnosis of rare disorders. AlphaMissense focuses on missense mutations, where a single letter in the DNA code is altered, potentially disrupting protein function and causing diseases like cystic fibrosis, sickle-cell anemia, cancer, and brain developmental issues.
The researchers utilized AlphaMissense to evaluate 71 million single-letter mutations affecting human proteins. With a 90% precision setting, it identified 57% of missense mutations as likely harmless, 32% as probably harmful, leaving uncertainty regarding the rest. To benefit geneticists and clinicians, they’ve shared the predictions through a free online catalog.
While the human genome typically contains 9,000 missense mutations, less than 2% of over 4 million observed in humans are classified as benign or pathogenic. Existing computer programs for predicting disease-driving mutations lack precision.
AlphaMissense, inspired by DeepMind‘s AlphaFold, which predicts 3D protein structures, outperforms current “variant effect predictor” programs, aiding experts in swiftly identifying disease-driving mutations. It may also uncover novel links between mutations and specific disorders, offering guidance for improved treatments.
AlphaMissense was trained on human and primate DNA data to distinguish common, likely benign missense mutations from rare, potentially harmful ones. It also learned the protein “language” by analyzing millions of protein sequences to recognize a “healthy” protein structure.
Although AlphaMissense generates risk scores for mutations, it cannot elucidate the precise mechanisms of harm. Dr. Jun Cheng likened its operation to human language, where word substitutions can change sentence meaning. Professor Joe Marsh, a computational biologist at Edinburgh University, uninvolved in the project, recognizes AlphaMissense’s promising potential.
The UK government has unveiled plans to build a cutting-edge supercomputer with a budget of about £900 million, which is equivalent to US$1.1 billion. This high-performance computing powerhouse, known as Isambard-3, pays homage to the renowned 19th-century British engineer Isambard Kingdom Brunel.
The supercomputer is slated for installation at the National Composites Centre in Bristol later this year. Notably, the University of Bristol, which houses the UKRI Centre for Doctoral Training in Interactive Artificial Intelligence and is a member of the GW4 university consortium alongside Bath, Cardiff, and Exeter, will play a pivotal role in this endeavor.
Bristol University will also serve as the host institution for the new AI Research Resource, also referred to as AIRR or Isambard-AI. AIRR is envisioned as a national facility dedicated to supporting AI research and ensuring the responsible and safe use of this transformative technology. Both the supercomputer and AIRR are funded through the government’s AI investment initiative, which was initially announced in March.
The supercomputer is expected to be a formidable computing entity, featuring thousands of cutting-edge graphics processing units (GPUs). This configuration is anticipated to position it as one of Europe’s most powerful supercomputers. These developments were confirmed in a statement released by the Department for Science, Innovation, and Technology (DSIT).
Michelle Donelan, the Secretary of State for Science, Innovation, and Technology, expressed enthusiasm about this ambitious project, stating, “We are backing the future of British innovation, investing in a world-leading AI Research Resource in Bristol that will catalyze scientific discovery and keep the UK at the forefront of AI development.”
“The Isambard-AI cluster will be one of the most powerful supercomputers in Europe, and will help industry experts and researchers harness the game-changing potential of AI, including through the mission-critical work of our Frontier AI Taskforce,” she added.
SoftBank, in the wake of the successful listing of UK chip designer Arm, is actively pursuing investment opportunities in the realm of artificial intelligence (AI). Masayoshi Son, the founder and CEO of the Japanese conglomerate, is eyeing substantial investments in OpenAI, potentially in the tens of billions, dedicated to the field of AI following Arm’s recent IPO.
According to sources familiar with Son’s intentions, SoftBank is exploring various avenues, and one prominent option is an investment in OpenAI, a company backed by Microsoft. SoftBank is also considering the possibility of forging a comprehensive strategic partnership with OpenAI, the creator of ChatGPT.
Furthermore, SoftBank is not limiting its AI aspirations to OpenAI alone. The company is contemplating significant investments in competitors of ChatGPT and has even approached Graphcore, a UK-based AI chipmaker, with preliminary acquisition interest, as disclosed by insiders.
In response to inquiries, SoftBank provided a standard statement, stating, “We do not comment on rumors.” OpenAI declined to offer any comment on the matter, while Graphcore unequivocally denied receiving an offer from SoftBank.
The recent Arm IPO, which generated nearly $5 billion in proceeds, has substantially augmented SoftBank’s financial resources, potentially increasing its war chest to a staggering $65 billion. This total encompasses both the company’s liquid assets and its 90 percent stake in Arm, which may be utilized as collateral for loans.
Of particular note is Masayoshi Son’s personal affinity for ChatGPT, as he has openly declared himself a “heavy user” of the technology. His close rapport with Sam Altman, the CEO of OpenAI, is also well-documented, with Son describing Altman as “one of the key people on Earth” and noting frequent daily interactions between the two.
