 OpenAI unveiled GPT-4, its latest and most advanced large language model that accepts both text and images. GPT-4 is now available to the public through OpenAI's ChatGPT Plus paid subscription. The latest GPT has also already been powering Microsoft's Bing search engine AI, the company revealed. More: - While GPT-4 accepts both text and images as input, its output remains text only.
- For example, a user could upload a photo or image, and GPT-4 will respond with a text description or caption for what it "sees."
- OpenAI has also launched a waitlist for a GPT-4 API for developers.
- Early adopters of GPT-4 include Duolingo, Morgan Standley, Khan Academy, and Stripe.
GPT-4 improvements: - The new GPT is more powerful than GPT-3.5, which has powered ChatGPT.
- GPT-4 supports up to 25,000 words of text, an increase from 1,000.
- The model is said to be "safer and more aligned" than prior models. In internal testing, it was 40% more likely to generate factual information compared to GPT 3.5.
- OpenAI says GPT-4 scored high on various tests and benchmarks. It performed in the 93rd percentile of an SAT reading exam and the 89th percentile of the SAT Math exam (see image above).
Quotable: - OpenAI CEO Sam Altman said GPT-4 is more creative than its predecessors, displays less bias, and hallucinates a lot less.
- It passed a simulated bar exam and scored a 5 on some AP exams, he wrote.
- However, GPT-4 is "still flawed, still limited, and it still seems more impressive on first use than it does after you spend more time with it," according to Altman.
- GPT-4 is also limited when it comes to current events, as most of its training data dates back to before September 2021.
THE NEW YORK TIMES 
| |
 Google said it's bringing generative AI technology to its Workspace products, starting with AI-enabled writing features in Gmail and Google Docs. The latter features will begin rolling out to a limited number of U.S. testers this month. For example, testers will be able to prompt Gmail's AI to automatically compose an email or use AI in Docs to polish summaries of meetings. More: - Google Workspace, its productivity suite, claims nearly 3 billion users, including individual and company users.
- Google says it will roll out generative AI tools to most or all of its Workspace products, which include Gmail, Chat, Meet, Calendar, Drive, Docs, Sheets, Slides, Forms, and Sites.
- In Sheets, for example, AI will automatically create formulas. Slide's AI will be able to auto-generate images, audio, and more for presentations.
- Google says it will launch the AI writing tools in Gmail, Docs, and other apps to the public sometime later this year.
New AI developer tools: - The AI tools are built on Google's large language model known as PaLM (Pathways Language Model).
- PaLM also forms the basis for other newly announced Google Cloud services that cover new tools and APIs for developers.
- These include PaLM API and the related MakerSuite, an environment for developers creating apps to test out generative AI projects.
- Google said it's also adding generative AI to Vertex AI, its platform for businesses to build and deploy AI and ML applications at scale using its models.
- Meanwhile, Google Cloud's Generative AI App Builder enables developers and business users to create AI apps such as chatbots "in minutes or hours."
- The app builder and Vertex AI update are also rolling out to a limited number of testers, including Totoya and Mayo Clinic, starting today.
THE VERGE
| |
 A message from TOPTAL Don’t waste weeks or months on the typical hiring process. In just 2-5 days, Toptal can provide you with the carefully vetted experts you need for your mission-critical projects. Because of our rigorous screening process, only the top 3% of applicants are accepted into the Toptal network. With over 10,000 experts around the globe, our talent pool is made up of exceptional developers, designers, project managers, and more. Since 2010, we’ve worked with over 20,000 clients, including industry leaders like Motorola, Zendesk, and Kraft-Heinz. Whether you’re looking for just one expert or an entire team, Toptal has the world-class specialists you need, all available on demand. Get Started Today
| |
Microsoft has reportedly laid off the ethics and society team within its AI unit as part of companywide layoffs announced in January. At the time, Microsoft said it would lay off 10,000 employees, or about 4.5% of its total workforce, through March 31. More: - The company's roughly seven-person ethics and society team was announced as part of those layoffs this week.
- The team helped to ensure that Microsoft products and services are designed in ways that accurately reflect its responsible AI principles before launching those tools to the public.
- According to TechCrunch, team members were recently exploring the risks posed by OpenAI technology, such as ChatGPT, as that tech is further incorporated into Microsoft's product suite.
Response: - In a statement, Microsoft said it's still committed to "developing and designing AI products and experiences safely and responsibly."
- The company said its investment in AI has evolved, resulting in adjustments to team structures.
- It has added more employees directly within its product teams who are focused on incorporating AI principles.
AI Principles: - Microsoft said it's also expanded the scope of its separate Office of Responsible AI, which develops its AI principles and rules.
- That Office "provides cross-company support for things like reviewing sensitive use cases and advocating for policies that protect customers," according to the company.
- However, one former ethics and society team member said their colleagues did not always know how to apply the principles coming out of the Office.
- "Our job was to show them and to create rules in areas where there were none," that person said.
TECHCRUNCH
| |
Adept AI, a startup working on AI models that can perform tasks from text commands, has brought in $350M as part of a Series B funding round, giving it a post-money valuation of at least $1B, Forbes reports. San Francisco-based Adept emerged from stealth nearly a year ago. The company has since shown off a working demo of an early AI digital assistant that could respond to user requests by taking action, freeing humans to focus on other work. More: - General Catalyst and Spark Capital led the $350M in financing. Adept last announced a $65M Series A round in April 2022.
- Since then, the company has announced its first large AI model for performing actions from natural language requests.
- Called ACT-1, that model can "take complex user requests across a range of software tools and execute them, even coordinating across multiple tools when needed," according to Adept.
- ACT-1 started out being able to answer questions and can now do more complex tasks, such as "importing LinkedIn URLs into recruiting software," Forbes reports.
- Adept's long-term goal is to develop an "AI teammate" that's trained to use every available software tool and API.
- It would take in a variety of text commands and perform related tasks to automate workflows for knowledge workers.
Who's who: - Adept was founded by CEO David Luan, a former Google Research director and former VP and director of engineering at OpenAI.
- Adept's co-founders include Ashish Vaswani (chief scientist) and Niki Parmar (CTO), former Google Brain researchers who co-authored the deep learning model known as the Transformer.
FORBES
| |
New job postings in AI: - AI/ML Senior Engineer
- Staff Writer, Artificial Intelligence & Algorithms Reporter
- The Washington Post | San Francisco, California
- Artificial Intelligence (AI) Engineering Subject Matter Expert
- Executive Recruiting | Arlington, Virginia
- Senior AI/ML Engineer
- Apple | Cupertino, California
- Artificial Intelligence Leader, Life Sciences — SFL Scientific
- Deloitte | Multiple locations
- Head of the Centre for Artificial Intelligence
- AstraZeneca | Cambridge, U.K., Gaithersburg, Maryland, or Gothenburg, Sweden
- Postdoctoral Data Science, Computer Vision
- Johnson & Johnson | Cambridge, Massachusetts
| |
QUICK HITS - End the madness of manual expense reporting with Divvy. Take a demo of Divvy and get a college basketball jersey of your choice.*
- Morgan Stanley plans to introduce a chatbot based on OpenAI's GPT-4. The AI tool has been undergoing testing and will roll out to all the bank's financial advisers in the upcoming months.
- Optiwe added AI and ChatGPT features to its messaging platform to help sales team members communicate effectively with prospective clients.
- Lowe's is testing Knightscope's K5 autonomous robots at certain U.S. stores. The egg-shaped robots are patrolling the retailer's parking lots in an effort to increase security.
- Ready to grow your business through influencer partnerships? Get the visual guide to the essential stages.*
*This is sponsored content.
| |
UPCOMING EVENTS - March 14 - Inside Startups Coffee Break (Register Here)
- March 15 - Monthly Meditation guided by Nicholas Whitaker (Register Here)
- March 20 - Inside Interview with Rollworks - Clearing the Path to ABM Success: Addressing the Issues Holding You Back (Register Here)
- March 27 - Insider Interview with Zscaler - Zero Trust Defense Strategies to Take Your Security to the Next Level (Watch On Demand)
- April 18 - Human Resources Summit'23 (Register Here)
| |
 |  | Beth is a writer and editor at Inside.com covering artificial intelligence and daily technology news. She has written the Inside AI newsletter since 2019. You can reach her at Beth.Duckett@inside.com. |   | | Editor | Aaron Crutchfield is based in the high desert of California. Over the last two decades, he has spent time writing and editing at various local newspapers and defense contractors in California. When he's not working, he can often be found looking at the latest memes with his kids or working on his 1962 and 1972 Fords. | |
Why delay success? Hire world-class talent in just 2-5 days with Toptal. | |
Comments
Post a Comment