Рубрика: AI

India’s economy relies heavily on micro and small enterprises, contributing 40% to its GDP. However, a notable challenge exists: only half of these entrepreneurs can secure loans due to a substantial credit gap. The primary obstacle is the lack of collateral among these entrepreneurs, leaving them unable to meet traditional loan requirements.

Although easier said than done, to address this issue, the Government of India, the Ministry of MSME, and the Small Industries Development Bank of India (SIDBI) established the Credit Guarantee Fund Trust for Micro and Small Enterprises (CGTMSE) in 2000 to guarantee loans for MSEs, enabling banks to lend money without collateral.

“Despite these efforts, women entrepreneurs in India still face significant challenges in accessing institutional finance, including social attitudes and bias, difficulty in securing collateral-based loans, and poor awareness or knowledge of financial schemes. To close the financing gap for MSEs, there is a need for focused and prompt action,” said Harish Gupta, in an exclusive interaction with AIM at Oracle CloudWorld 2023 in Las Vegas, last week.

Oracle to the Rescue

Thanks to Oracle, CGTMSE has benefited approximately 12 million borrowers, including 21% women-led entrepreneurs and 92% first-time borrowers.

“We guaranteed loans totalling $53 billion, of which $21 billion are currently active with $13 billion in loans over the last year,” Gupta commented.

CGTMSE moved to Oracle Cloud in 2021 and as a result, underwent significant changes in its data infrastructure and application modernisation processes. Over 23 years, they managed to disburse a total of $52 billion in loans, resulting in an 87% increase in their top-line revenue. In a single year, they disbursed $13 billion.

The Oracle adoption led to a 56% growth in the first year, increasing their revenue from INR 36,000 crore to INR 56,000 crore. “In the second year, it surged from INR 56,000 crore to INR 101,000 crore and INR 4,000 crore,” shared Gupta. This growth demonstrates a significant return on investment, both in terms of IT savings and substantial business expansion, with no human intervention required for daily operations.

One of CGTMSE’s core responsibilities is managing various government schemes in India, including those aimed at non-banks, banks, and COVID-affected entrepreneurs. However, they faced challenges stemming from the outdated architecture, which required separate application environments and databases for each scheme. To address this, they unified their applications into a single, highly configurable platform built on microservices architecture.

“So the adoption of Oracle Cloud services had a significant impact on our operations by improving their time-to-market for new schemes and resolving resource scaling issues during peak periods,” shared Gupta, saying that automation was a key aspect of their transformation, particularly in the loan application assessment process, resulting in significantly reduced processing times.

“Loans that previously took two weeks to disburse were now processed and disbursed in real-time,” commented Gupta, on how Oracle has boosted their productivity.

Furthermore, CGTMSE implemented digital initiatives like fast-tracking Mudra loans for lower-income segments, contributing to their financial growth. In fact, within a year, their top line improved by an impressive 87%, demonstrating a substantial return on investment (ROI), largely attributable to savings generated through Oracle ATS (Application Testing Service), impacting loan sanctions, increasing from 100 crores a day to an impressive 500 crores a day, a five-fold growth.

Migration Struggle

“100% of our infrastructure relies on Oracle Cloud,” said Gupta, emphasising CGTMSE’s dependence on Oracle.

But before Oracle, the organisation was on a private cloud partner and faced data transfer issues during migration. The primary challenge was aligning applications with Oracle’s database parameters. After resolving issues and conducting security audits, DR drills, and compliance checks, Oracle proved reliable, especially for data security. The team migrated their whole infrastructure to Oracle Cloud in just four weeks.

In addition to Oracle, Path Infotech serves as a crucial partner for them. Application development is based on Java, Angular, and PrimeMG, all hosted on Oracle. However, their HRMS system is hosted on AWS, and their accounting system currently runs on Tally but is planned to migrate to a standard ERP system. In terms of security, they use FortiGate for firewall protection, and they are pursuing ISO 27,001 compliance for their SOC center. They are also exploring external vendors for various security solutions, including web and social solutions, and have been in contact with VMware for potential solutions.

What Next?

With a lean team of 40 employees, CGTMSE is currently integrating with various entities, including IndiaStack, ONDC to ensure lenders have access to borrower eligibility information.

Alongside traditional banks, the company is expanding their reach to Loan Service Providers (LSPs) and Digital Service Providers (DSPs), creating a comprehensive ecosystem. “Our APIs enable these entities to assess borrower eligibility early in the loan application process,” he said.

CGTMSE is also betting big on generative AI. “We plan to invest in AI for automation and cost reduction, including chatbots for both public and internal use, leveraging cognitive search and generative AI for efficient information delivery,” Gupta added.

“We aim to use chatbots for customer support to reduce costs, especially for their widespread user base of 10,000 people across India, including various banks and automate incident management, develop crash analytics, and utilise generative AI,” said Gupta. However, his involves two aspects: a multilingual public-facing bot using cognitive search and generative AI and a bot for internal business users, integrating with their systems to handle queries efficiently. These automation efforts will require ongoing investment, with a conservative target of 25 billion for this year.

Read more: How Oracle is Fuelling Musk’s Ambitions

The post How this Govt Initiative Found Moat in Oracle appeared first on Analytics India Magazine.

OpenAI’s GPT-4 with vision still has flaws, paper reveals Kyle Wiggers 14 hours

When OpenAI first unveiled GPT-4, its flagship text-generating AI model, the company touted the model’s multimodality — in other words, its ability to understand the context of images as well as text. GPT-4 could caption — and even interpret — relatively complex images, OpenAI said, for example identifying a Lightning Cable adapter from a picture of a plugged-in iPhone.

But since GPT-4’s announcement in late March, OpenAI has held back the model’s image features, reportedly on fears about abuse and privacy issues. Until recently, the exact nature of those fears remained a mystery. But early this week, OpenAI published a technical paper detailing its work to mitigate the more problematic aspects of GPT-4’s image-analyzing tools.

To date, GPT-4 with vision, abbreviated “GPT-4V” by OpenAI internally, has only been used regularly by a few thousand users of Be My Eyes, an app to help low-vision and blind people navigate the environments around them. Over the past few months, however, OpenAI also began to engage with “red teamers” to probe the model for signs of unintended behavior, according to the paper.

In the paper, OpenAI claims that it’s implemented safeguards to prevent GPT-4V from being used in malicious ways, like breaking CAPTCHAs (the anti-spam tool found on many web forms), identifying a person or estimating their age or race and drawing conclusions based on information that’s not present in a photo. OpenAI also says that it has worked to curb GPT-4V’s more harmful biases, particularly those that relate to a person’s physical appearance and gender or ethnicity.

But as with all AI models, there’s only so much that safeguards can do.

The paper reveals that GPT-4V sometimes struggles to make the right inferences, for example mistakenly combining two strings of text in an image to create a made-up term. Like the base GPT-4, GPT-4V is prone to hallucinating, or inventing facts in an authoritative tone. And it’s not above missing text or characters, overlooking mathematical symbols and failing to recognize rather obvious objects and place settings.

Image Credits: OpenAI

It’s not surprising, then, that in unambiguous, clear terms, OpenAI says GPT-4V is not to be used to spot dangerous substances or chemicals in images. (This reporter hadn’t even thought of the use case, but apparently, the prospect is concerning enough to OpenAI that the company felt the need to call it out.) Red teamers found that, while the model occasionally correctly identifies poisonous foods like toxic mushrooms, it misidentifies substances such as fentanyl, carfentanil and cocaine from images of their chemical structures.

When applied to the medical imaging domain, GPT-4V fares no better, sometimes giving the wrong responses for the same question that it answered correctly in a previous context. It’s also unaware of standard practices like viewing imaging scans as if the patient is facing you (meaning the right side on the image corresponds to the left side of the patient), which leads it to misdiagnose of any number of conditions.

Image Credits: OpenAI

Elsewhere, OpenAI cautions, GPT-4V doesn’t understand the nuances of certain hate symbols — for instance missing the modern meaning of the Templar Cross (white supremacy) in the U.S. More bizarrely, and perhaps a symptom of its hallucinatory tendencies, GPT-4V was observed to make songs or poems praising certain hate figures or groups when provided a picture of them even when the figures or groups weren’t explicitly named.

GPT-4V also discriminates against certain sexes and body types — albeit only when OpenAI’s production safeguards are disabled. OpenAI writes that, in one test, when prompted to give advice to a woman pictured in a bathing suit, GPT-4V gave answers relating almost entirely to the woman’s body weight and the concept of body positivity. One assumes that wouldn’t have been the case if the image were of a man.

Image Credits: OpenAI

Judging by the paper’s caveated language, GPT-4V remains very much a work in progress — a few steps short of what OpenAI might’ve originally envisioned. In many cases, the company was forced to implement overly strict safeguards to prevent the model from spewing toxicity or misinformation, or compromising a person’s privacy.

OpenAI claims that it’s building “mitigations” and “processes” to expand the model’s capabilities in a “safe” way, like allowing GPT-4V to describe faces and people without identifying those people by name. But the paper reveals that GPT-4V is no panacea, and that OpenAI has its work cut out for it.

A podcast with Weimo Liu and Sam Magnus of PuppyGraph

Open source Apache Iceberg, Hudi and Delta Lake have made it possible to dispense with the complexities and duplication of data warehousing. Instead of requiring time-consuming extract, transform and load (ETL) procedures, these large table formats make it simple to tap S3 and other repositories directly for analytics purposes.

PuppyGraph takes this modern data lake capability a step further, making it possible to create and query graphs in a on top of one of these large analytic table formats. After creating a subgraph with a SQL group by statement, for example, a user can do graph traversal via Gremlin or declarative graph querying using Cypher via these tables, or use a prompt interface on the data via LlamaIndex. Here’s how the PuppyGraph architecture eliminates the need for a graph database.

With a single logical schema, you can tap multiple heterogeneous repositories such as one that’s MySQL and an S3 bucket both via an Iceberg-enabled table, says PuppyGraph CEO Weimo Liu.

Let’s say a user has a table with transaction records. The sender, receiver and amount fields already suggest a graph of sorts. The user can just use SQL and a “group by” query to create a subgraph of selected records, then create a logical schema and do graph queries. That’s how you can leverage both SQL and graph querying with a tool such as Gremlin in the same use case.

With a SQL query engine side by side with a graph engine, users can feed the retrieved outputs into machine learning algorithms and reporting tools, says PuppyGraph founding member Sam Magnus.

A video Weimo posted on YouTube called “PuppyGraph – Query the Data in Your Data Lake as a Graph in One Minute” walks users through how to create a simple schema to query their data via a modern data lake enabled by Iceberg as a graph by following these steps (paraphrased here):

1. Create a schema JSON file to tell PuppyGraph these specifics:
   1. The URL of your data lake
   2. Which tables are nodes
   3. Which tables are edges.
2. Post the file to the server. The server will confirm the schema creation.
3. Connect the console of a graph query tool such as Gremlin or Cypher to the server
4. Run your query to retrieve the results.

Note that PuppyGraph is not a database, but a graph analytics engine designed to work with all the seemingly limitless information modern data lakes can store.

“Single copy” analytics is a phrase Sam Magnus uses to underscore that multiple analystics users can all use the same copy of the data to support both SQL and graph workloads, rather than having to copy over and over again.

By using PuppyGraph, Sam estimates that consultants hired for graph analytics projects can be up and running within an hour at client sites, assuming the necessary data is accessible. That’s a big deal, considering that graph database proofs of concept (PoCs) have historically required extensive up-front work to begin delivering results.

Sam and Weimo encourage everyone to try out PuppyGraph–they’ve got a free Docker in the user manual and haven’t been charging for consulting. Hope you enjoy the podcast.

Announcements

• End-user computing must now account for the millions of workforces that have transitioned to hybrid and remote models. Virtual workspace models such as DaaS and VDI allow users to access virtual desktops to help streamline their workflow and minimize the burden on IT staff. However, companies must consider cost, scalability and management when deploying a virtual workspace, as well as security strategies to ensure workers are protected and able to remain productive. Join the Next-Generation End User Computing summit to discover how to best implement and manage hosted and virtual workspaces including DaaS and VDI.
• Managing the supply chain is exceedingly difficult with global conflicts and market ups and downs interfering with companies’ ability to timely deliver and fulfil orders. Tune into the Overcoming Supply Chain Challenges summit to hear leading experts discuss emerging technologies to help protect and streamline supply chain management along with strategies and tools to secure the supply chain against the many cyber threats it faces. Register for free and gain access to live webinars, fireside chats and keynote presentations from the world’s leading supply chain innovators, vendors and evangelists.

Top Stories

• How AI growth has triggered data center redesign
  September 25, 2023
  by Alan Morrison
  A major aspect of ongoing data center redesign is due to AI’s massive, complex workloads and the need to add many more graphic processing units (GPUs), tensor processing units (TPUs) or accelerators to the mix.
• Use Case Language Models: Taming the LLM Beast – Part 1
  September 23, 2023
  by Bill Schmarzo
  “Sometimes, you don’t know where you’re going until you get there.” – Schmarzo-ism? Yes, writing this blog turned into a journey. I started in one direction, but after several twists and turns, I ended up with this concept – that use case-centric language models can be combined into entity-centric language models that can support multiple use cases at minimal marginal costs, significantly impacting the economics of AI development.
• In fraud detection for e-commerce: How does anomaly detection fit in and what are the key approaches?
  September 25, 2023
  by John Lee
  E-commerce has improved technology and convenience for consumers globally. Fraud is a problem in e-commerce. Merchants and platforms fight fraud to protect their businesses and customers. Anomaly detection is a powerful tool for identifying irregular patterns and potential fraud.

In-Depth

• Doing graph + tabular analytics directly on modern data lakes
  September 26, 2023
  by Alan Morrison
  A podcast with Weimo Liu and Sam Magnus of PuppyGraph Open source Apache Iceberg, Hudi and Delta Lake have made it possible to dispense with the complexities and duplication of data warehousing.
• The essential guide on data security and privacy in web localization
  September 25, 2023
  by Nakivo Backup & Replication
  Thanks to the internet, you can now easily expand your reach and engage with diverse audiences wherever they are. However, this opportunity raises an important question: how can you localize your web content and maintain the security and privacy of sensitive data?
• Doing graph + tabular analytics directly on modern data lakes
  September 22, 2023
  by Alan Morrison
  A podcast with Weimo Liu and Sam Magnus of PuppyGraph Open source Apache Iceberg, Hudi and Delta Lake have made it possible to dispense with the complexities and duplication of data warehousing.
• AI in finance: Addressing hurdles on the path to transformation
  September 22, 2023
  by Aileen Scott
  Discover the obstacles hindering seamless AI adoption in financial services and gain actionable insights to navigate regulatory compliance, data security, organizational change, and more.
• DSC Weekly 19 September 2023
  September 19, 2023
  by Scott Thompson
  Read more of the top articles from the Data Science Central community.
• A guide to setting up analytics at a consumer tech startup
  September 19, 2023
  by Abhi Sawhney
  Where do you start if you want to build a data analytics function from the ground up? As an analytics leader at a startup, you will need to make several important decisions early on to build an effective team. This article dives into four decision areas and highlights ways in which to think about them.

Vultr Launches GPU Stack and Container Registry for AI Model Acceleration Worldwide September 26, 2023 by Ali Azhar

Configuring and provisioning GPUs is a notoriously painful activity. To provide some relief, Vultz announced it has launched the Vultr GPU Stack and Container Registry for AI model acceleration worldwide. This stack and registry will help digital startups and enterprises around the globe to build, test, and operate AI models at scale. It will accelerate the development, collaboration, and deployment of machine learning (ML) and AI models. Developer and data science teams can quickly tune and train their models on their own data sets, with everything pre-configured, down to their Jupiter notebooks.

As one of the world’s leading privately-held cloud computing platforms, Vult has over 1.5 million customers across 185 countries. The GPU Stack and Registry is set to be available across Vultr’s 32 cloud data centers in six continents. Vultr has been in the news recently, with its cloud alliance with Yext and the introduction of the Vultr WebApp.

The Vultr Container Registry makes AI pre-trained NVIDIA NGC models available worldwide for on-demand provisioning, training, development, tuning, and inference. The Vultr GPU Stack is designed to support instant provisioning of the full capabilities of NVIDIA GPUs.

According to Dave Salvator, director of accelerator computing graphics of NVIDIA, a key advantage of using Vultr GPU Stack and Container Registry is that it provides organizations with instant access to the entire library of pre-trained LLMs on the NVIDIA NGC catalog. This enables them to accelerate their AI initiatives and provision and scale NVIDIA cloud GPU instances from anywhere.

Contant is the creator and parent company of Vultr. J.J. Kardwell, CEO of Constant, said that “at Vultr we are committed to enabling the innovation ecosystems — from Miami to São Paulo to Tel Aviv, Europe and beyond — giving them instant access to high-performance computing resources to accelerate AI and cloud-native innovation,”.

He further added “By working closely with NVIDIA and our growing ecosystem of technology partners, we are removing barriers to the latest technologies and offering enterprises a composable, full-stack solution for end-to-end AI application lifecycle management. This capability enables data science and engineering teams to build on their global teams’ models without having to worry about security, local compliance, or data sovereignty requirements.”

The development, testing, and deployment of AI and ML models can get extremely complex and this is set to become even more challenging with looming regulations for safe, secure, and transparent development of AI technology. One of the biggest challenges is the configuration and provisioning bottlenecks. The best tools and technologies are needed to build, run, test, and deploy AI and ML models.

Some of these challenges will be addressed with the launch of the Vultr GPU stack. which comes with a full array of NVIDIA GPUs, including NVIDIA cuDNN, NVIDIA CUDA Toolkit, and NVIDIA drivers for instant deployment. The Vultr GPU Stack provides a finely tuned and integrated operating system and software environment to remove the complexity of configuring GPUs, calibrating them to the model requirements, and integrating them with the AI model accelerators.

Data scientists, engineering teams, and developers across the globe can bring models and frameworks for the NVIDIA NGC catalog to get started on their AI model development and training with a click of a button.

Related Items

Nvidia Upgrades NeMo Megatron AI Toolkit

Oracle Introduces Integrated Vector Database for Generative AI

Anyscale and Nvidia In LLM Hookup

Related