This week marked the unveiling of GPT-4, a generative AI large language model that can respond to both text and images and perform better than humans on many standardized tests, including college entrance and bar exams. If pundits are right the technology will become an important enterprise productivity tool, impacting the future of work.
“We think 2023 will be the year that generative AI will become prevalent and a key component of modern office work,” Charles Lamanna, Microsoft’s corporate vice president of business apps and platforms said in an interview with The Wall Street Journal.
Both Microsoft and Google made announcements this week aimed at doing exactly that.
On March 16 Microsoft, previewed a new GPT-4-powered “copilot” for Microsoft 365, its product suite that includes Word documents, Excel spreadsheets, PowerPoint presentations and Outlook emails. Microsoft said AI can offer a first draft in these applications, speeding up content creation and freeing up workers’ time.
The company showcased a new “business chat” experience that can pull data and perform tasks across Microsoft’s applications simply on a user’s written command. It also demonstrated how AI can democratize the computational powers of its Excel spreadsheet software to any person able to describe a calculation they would like in plain text and how its AI can summarize email threads and even virtual meetings as they occur live in its Teams collaboration software.
“We believe this next generation of AI will unlock a new wave of productivity growth,” Satya Nadella, Microsoft’s chief executive, said in a livestreamed presentation.
The news follows Microsoft’s unveiling last week of new AI-powered features integrated into several of its core business apps—including enterprise resource planning and customer relationship management systems—that tap natural-language models and generative AI technology developed by ChatGPT maker OpenAI. Microsoft has already rolled out an AI-enhanced version of its Bing search engine, built on top of OpenAI’s GPT technology, which is being used by a limited pool of users.
Not to be outdone Google this week announced its own AI enhancements to Google Workspace — which includes Google Docs, Gmail, and Excel-rival Sheets — showing how its tools can compose emails and create marketing materials. And, Anthropic, an artificial intelligence company backed by Google owner Alphabet, this week released a large language model that competes directly with offerings from Microsoft Corp-backed OpenAI, the creator of ChatGPT. Claude, as Anthropic’s model is known, is built to carry out similar tasks to ChatGPT by responding to prompts with human-like text output, whether that is in the form of editing legal contracts or writing computer code.
Meanwhile, Salesforce announced last week is plans to integrate OpenAI’s generative AI technology across its customer relationship management platform, including Slack, the group-chat app Salesforce acquired in 2021. The new set of AI tools, dubbed Einstein GPT, is currently in the pilot phase.
Putting Generative AI To Work
A handful of enterprise customers are already working with GPT-4. For example, PwC wants to use the technology to speed up the work of its 4,000 lawyers. It has signed a 12-month contract with a start-up called Harvey, which uses software powered by GPT-4,. PwC said the technology would help them work more quickly on tasks such as analyzing contracts and carrying out due diligence. The Big Four firm said it also planned to find ways to use the service in its tax practice. Carol Stubbings, PwC’s global tax and legal services leader, told The Financial Times the technology “marks a huge shift in the way that tax and legal services will be delivered and consumed across the industry”. The firm said the technology would speed up decision-making by producing answers to questions, which would then be reviewed and added to by staff. Because it was capable of parsing vast quantities of text and writing convincing answers to questions, it could also be used to summarize key clauses from batches of contracts, PwC said, and eventually to produce initial due diligence reports based on instructions from lawyers.
Morgan Stanley Wealth Management is using GPT-4 to build a system that will instantly retrieve information from company documents and other records and serve it up to financial advisers in conversational prose. The American multinational financial services corporation hosts a content library comprising hundreds of thousands of pages covering investment strategies, market research, commentary, and analyst insights; a massive repository stored across numerous internal sites, primarily in PDF format, which necessitates advisors to sift through extensive information to address specific queries. This process can be both time-consuming and unwieldy, so Morgan Stanley is using GPT-4 to power an internal-facing chatbot that performs a comprehensive search of wealth management content and attempts to unlock the cumulative knowledge of Morgan Stanley Wealth Management, Jeff McMillan, Head of Analytics, Data & Innovation, whose team is leading the initiative, told AI Magazine. “You essentially have the knowledge of the most knowledgeable person in Wealth Management — instantly”, said McMillan. “Think of it as having our Chief Investment Strategist, Chief Global Economist, Global Equities Strategist, and every other analyst around the globe on call for every advisor, every day. We believe that is a transformative capability for our company.”
Stripe, the San Francisco-based global payments platform, currently has 14 GPT-4 application prototypes in the works, Emily Sands, Stripe’s head of information, told Reuters. The first offering will be a way for Stripe’s software developers to type out a question and receive summarized answers instead of having to search through developer documentation, Sands said.Another test in the works allows Stripe’s customers to make queries about their business analytics using natural language instead of needing to write database queries, she added.
Tastewise, an Israeli AI-powered market intelligence platform, is using generative AI to introduce what it calls lightning-fast conversational and contextual insights to help brands such as Nestle, Mars and PepsiCo to find opportunities in the food and beverage marketplaces.
Small to midsized firms are also starting to use generative AI for office work. A recent survey, by workforce management experts at WorkYard, found that some SMEs are using ChatGPT 3.5, a precursor to GPT-4, to perform a variety of tasks such as social media management, email outreach, content creation, customer service and data analysis.
Companies need to start preparing for the impact generative AI will have on the work force, AI and future of work expert Arun Sandararajan, the Harold Price Professor of Entrepreneurship and Professor of Technology, Operations and Statistics at New York University’s Stern School of Business, said in an interview with The Innovator. (See The Innovator’s Interview Of The Week for the full interview)
During pilot programs testing the new tools, Microsoft told the Wall Street Journal that issues that came up with corporate users included concerns around data control and security, and ensuring results produced by the AI apps did not violate the company’s ethical guidelines. To guard against these and other risks, Microsoft said the new tools are trained on data from customers’ own systems, helping to ground results in reality. At the same time, the apps never directly talk to customers or directly taking action. They only provide content that is edited and reviewed by a human.
Yet despite assurances from the enterprise market’s biggest players, most corporate are taking a wait-and-see approach to ChatGPT-like generative AI technology. Only 12% of nearly 500 information-technology decision makers recently surveyed by market research firm Enterprise Technology Research said they plan to use OpenAI technology, or allocate further resources after initially assessing it. As many as 44% of respondents said they were aware of OpenAI technology but had no plans to check it out.
That is not surprising given recent news stories about ChatGPT going off the rails.
In a blog post OpenAI said it spent six months making GPT-4 “safer.” It claims GPT-4 is 82% less likely to respond to requests for disallowed content and 40% more likely to produce factual responses from GPT 3.5 on internal evaluations. However, It still has a major problem with “hallucination,” or making stuff up, and isn’t factually reliable, the company said.
Industry observers say this could not only expose companies to legal issues it could have significant negative consequences for society, further blurring the lines between truth and fiction, leading to more polarization as well as more sophisticated forms of phishing and manipulation of voters.
In a recent interview with The Innovator Stuart Russell, a Professor of Computer Science at the University of California at Berkeley, holder of the Smith-Zadeh Chair in Engineering, and Director of the Center for Human-Compatible AI and the Kavli Center for Ethics, Science, and the Public, called for the creation of an equivalent of a government agency like the U.S. Food and Drug administration to oversee algorithms.
NYU Professor Gary Marcus and a member of the Canadian parliament have argued that it’s time for government to consider frameworks that allow for AI research under a set of rules that provide ethical standards and safety, while pausing the widespread public dissemination of potentially risky new AI technologies—with severe penalties for misuse – until there is assurance of the safety of new technologies that the world doesn’t yet fully understand.
The release of GPT-4 is only increasing concerns as OpenAI is refusing to release any information about how the technology was developed.
“It puts all of us in an extremely poor position to predict what GPT-4 consequences will be for society if we have no idea of what is in the training set and no way of anticipating which problems it will work on and which it will not,” Marcus said this week in a blog posting. “One more giant step for hype, but not necessarily a giant step for science, AGI [artificial general intelligence] or humanity.”
IN OTHER NEWS THIS WEEK
Spain’s IESE Business School Uses Quantum, Computing With PC Software In Classrooms
Multiverse Computing, a startup specializing in quantum and quantum-inspired software, said its software was leveraged by IESE Business School in its Master in Management) program to demonstrate how quantum algorithms can solve problems more efficiently than classical algorithms. Singularity, the company’s quantum computing software as a service (qSaaS) platform, solves challenges for large enterprises across all industries by allowing users to leverage quantum computing directly through Microsoft Excel.
IESE used the software in a recent class at its Madrid campus with over 100 students who were able to demonstrate that quantum solutions can do a better job solving specific problems than previously proposed classical solutions. The classroom problem focused on the global financial crisis of 2008, which forced financial institutions to increase the rigor of its banking practices. As the availability of credit became limited, banks tightened their lending systems and needed to anticipate risky loans more accurately. Data science teams were created to assess whether clients could repay their loans within the stipulated time. In class, the IESE students formulated the challenge as a machine learning problem and proposed a classification algorithm that considered the default status of the client based on characteristics such as a client’s checking balance and work status as well as the loan amount and duration, among other factors. This use case is an inherently complex problem due to the multiple factors at play. Using the proposed classical machine learning algorithm, the students were able to correctly forecast 74% of all loan defaults. Using Multiverse Computing’s Singularity Excel add-in, the IESE students were able to solve the same problem using a quantum machine learning algorithm on a D-Wave quantum machine with 54 active qubits. During the in-class demo, students achieved an accuracy of 78%. Using larger datasets and additional Multiverse Computing’s algorithms, accuracy can reach about 90% in some cases, according to the startup.
State-sponsored hackers from China have developed techniques that evade common cybersecurity tools and enable them to burrow into government and business networks and spy on victims for years without detection, according to Google researchers.Over the past year, analysts at Google’s Mandiant division have discovered hacks of systems that aren’t typically the targets of cyber espionage. Instead of infiltrating systems behind the corporate firewall, they are compromising devices on the edge of the network—sometimes firewalls themselves—and targeting software built by companies such as VMware. or Citrix Systems. These products run on computers that don’t typically include antivirus or endpoint detection software.
FOOD AND AGRICULTURE
Bayer And Microsoft Team On Tech Tools For The Food And Agriculture Industry
Bayer and Microsoft have formalized the partnership they announced just over a year ago to create a cloud-based set of data tools and data science solutions for the food and agriculture industry.The announcement comes in the same week that cloud competitor AWS launched a partnership with data infrastructure startup Leaf.
Should I Stay Or Should I Go?: VW Heads For North America
As predicted Volkswagen said it will build its first North American battery plant in Canada. The German carmaker on March 13 said its battery arm PowerCo would build its first plant outside Europe in St Thomas, Ontario. VW had told EU officials that it was putting on hold a planned battery plant in eastern Europe while it waited for the bloc to respond to Washington’s $369 billion package of subsidies in the Inflation Reduction Act.
Silicon Valley’s Bank Collapse Underscores Role Of U.S. Money In Global Tech Financing
The failure of Silicon Valley Bank reverberated through startups and venture-capital firms from China to Singapore and India during a roller coaster few days that shook confidence in Asia over reliance on U.S. tech financing.
BNP Parisbas Partners With StartUp To Move Into B2B BNPL Services
BNP Paribas is moving into the B2B buy now, pay later space through a partnership with fintech Hokodo. Built specifically for large multinationals, the BNPL app is integrated with existing checkout systems and provides instant buyer approval through a real-time credit decision process.
The full service includes proprietary credit decisioning, transaction financing, credit and fraud insurance, collections through an eMandate and dunning, as well as different financing options available to better fit merchants’ needs.
To read more of The Innovator’s News In Context articles click here.