OpenAI has secured billions of dollars in additional funding from Microsoft to continue its development of generative artificial intelligence tools such as Dall-E 2 and ChatGPT. A move that is likely to unlock similar investment from competitors — notably Google — and open the way to new or improved software tools for enterprises large and small.
Microsoft stands to benefit from its investment in three ways. As a licensee of OpenAI’s software, it will have access to new AI-based capabilities that it can resell or build into its own products. As OpenAI’s exclusive cloud provider it will see additional revenue for its Azure services, as one of OpenAI’s biggest costs is providing computing capacity to train and run its AI models. And as an investor it can expect some return on its capital, although it will be governed by OpenAI’s status as a limited-profit company as a nonprofit.
The deal, announced by OpenAI and Microsoft on January 23, 2023, is likely to shake up the market for AI-based enterprise services, said Rajesh Kandaswamy, Distinguished Analyst and Fellow at Gartner: “This is additional motivation for Google to revisit provides. Its roadmap. It’s the same for other competitors like AWS,” he said.
Ritu Jyoti, IDC’s Global AI Research Lead, sees more than just AI bragging here. “There’s a big battle going on between the three hyperscalers – Amazon, Google and Microsoft – and it’s not just about AI. It’s going to drive who’s going to reign supreme in the cloud because there’s going to be tons and tons of Calculations are required, and they’re all fighting each other. It’s going to get ugly,” she said.
Employees are already experiencing some of that ugliness: Since the beginning of the year, Microsoft, Amazon, and Google parent Alphabet have all announced mass layoffs as they seek to focus on growth markets and AI. want to invest in.
billion dollar brain
Rumors that Microsoft may invest $10 billion to develop its AI business surfaced in early January. The company has been a supporter of OpenAI’s quest to build an Artificial General Intelligence since its early days, beginning in 2016 with hosting OpenAI experiments on dedicated Azure servers. In July 2019, it became the exclusive cloud provider of OpenAI and invested $1 billion in the company. To support his quest to create “artificial general intelligence”. In 2020, Microsoft became the first country to license OpenAI’s Generative Pre-trained Transformer (GPT) AI software for inclusion in its own products and services. Up to that point, OpenAI had only allowed enterprises and academia access to the software through a limited API.
Enterprises already have access to some of that technology through Microsoft’s Azure OpenAI service, which provides pay-as-you-go API access to OpenAI tools, including text generator GPT 3, image generator Dall -E 2, and a special codec is included. Version of GPT that can translate between natural language and programming language. Microsoft is also offering Codex as a Service as GitHub Copilot, an AI-based pair programming tool that can generate code snippets from natural language prompts. And it will soon offer Microsoft 365 subscribers a new application combining PowerPoint’s features with OpenAI’s Dall-E 2 image generator. That app, Microsoft Designer, is currently in closed beta testing. And, sure, they can check out ChatGPT, the interactive text generator that’s been making waves since its November 2022 release.
GPT-3.5, the OpenAI model on which ChatGPT is based, is an example of Transformer, a deep learning technology developed by Google in 2017 to tackle problems in natural language processing. Others include BERT and PaLM from Google; and MT-NLG, which was co-developed by Microsoft and Nvidia.
Transformers improve upon the previous generation of deep learning techniques, recurrent neural networks, in their ability to process entire text at once, rather than treating them sequentially, one word after another. This allows them to infer the relationship between words in multiple sentences, something that is especially useful when interacting with humans who use pronouns to save time. ChatGPT is one of the first to be provided as an interactive tool rather than as an API.
robot in disguise
The text that ChatGPT generates reads like a pedantic and not always well-informed human, and part of the concern about it being used to fill the Internet with human-sounding but confusing or meaningless text is can go. The risk there – apart from making the internet useless for humans – is that it will pollute the resources needed to train better AI.
Interacting with ChatGPT is fun, but the beta version available today isn’t very useful for enterprise purposes. This is because it does not have access to new information or services on the internet – the dataset it was trained on was frozen in September 2021 – and although it can answer questions about the content of that dataset, it cannot reference its sources, raising doubts about the accuracy of its statements. To its credit, it regularly and repeatedly reminds users of these limits.
An enterprise version of ChatGPT, however, refined to deal with an industry-specific terminology and with access to up-to-date information from the ERP on product availability, or the latest updates to the company’s code repository, would be quite a lot. .
in your own words
ChatGPT itself prompted the question, “What use would a CIO have for a system like ChatGPT?” suggested that it could be used to automate customer service and support; analysis of data for the preparation of reports; and generate suggestions and recommendations based on data analysis to aid decision making.
Asked to describe its limitations, ChatGPT said, “Its performance may be affected by the quality and quantity of training data. Additionally, it may not always be able to correctly understand or respond to certain information.” Illustrating well its tendency to repeat the same point multiple ways, it went on: “It is also important to monitor the performance of the model and adjust the training data as needed to improve its accuracy and relevancy.” Is.”
Regarding Microsoft’s plans for OpenAI’s generative AI tools, IDC’s Jyoti said she expects some of the most visible changes to come on the desktop. “Microsoft will completely transform its entire suite of applications: Word, Outlook and PowerPoint,” he said, noting that the integration of OpenAI could introduce or enhance features such as image captioning, and text autocompletion and recommendation of next actions .
Gartner’s Kandaswamy said he expects Microsoft to add new OpenAI-based capabilities to properties like Dynamics and even LinkedIn or GitHub, in addition to updating its productivity suite.
It’s important for CIOs to adopt these tools for the incremental value they bring, he said, but cautioned: “Be very careful not to get blindsided by the disruption AI can produce over the long term.”
chief ai officer
Jyothi assigns some responsibility for the effects of AI on enterprises themselves. “People always blame technology suppliers, but enterprises have a responsibility too,” she said. “Businesses, right from the C-suite, need to put together their AI strategy and put in place the right guardrails.”
For now, AI tools like ChatGPT or DAL-E2 are used to augment human creativity or decision making, not replace it. “Put a human in the loop,” he advised.
It won’t just be the CIO’s decision because the questions about which tools should be used, and how, are ethical as well as technical. Eventually, the job will return to the IT department. “They can’t ignore it: they have to pilot it,” she said.
make, don’t buy
Jyoti said that with few generative AI tools available to buy off the shelf for now, there will be a rebalancing of the build versus buy equation, prompting forward-thinking CIOs to build in the short term. With coding help from tools like GitHub Copilot or OpenAI’s Codex, limited developer resources can get there sooner.
Later, as ISVs move forward and build domain-specific solutions using the generative AI tools provided by OpenAI, Microsoft and other hyperscalers, the pendulum could swing back to buy for enterprises, he said.
The initial swing to customization (rather than configuration) could spell big trouble for Oracle, SAP, and other large ERP developers, who these days rely on enterprises to conform to the best practices incorporated into their SaaS applications.
“They have hardened processes over the years, but today AI has become data-driven,” Jyoti said. […] And it will require a fundamental change in the way things work.