A significant portion of OpenAI’s business is focused on assisting enterprise customers in developing AI products, despite the company’s consumer-facing products, such as ChatGPT and DALL-E, receiving the majority of attention. They are now receiving new tools for those customers.

Corporate clients that power their AI tools with OpenAI’s application programming interface (API) will receive improved security features, the company announced in a blog post, including the option to use single sign-on and multi-factor authentication by default. In order to lessen the chance of any data leaks onto the public internet, OpenAI has also implemented 256-bit AES encryption during data transfers.

Additionally, OpenAI has introduced a new Projects feature that makes it easier for businesses to manage who has access to various AI tools. Companies should find it easier to stick to their budgets with the new cost-saving features, according to OpenAI. One such feature is the ability to use a Batch API to reduce spending by up to 50%.

Although the OpenAI announcement this week isn’t as exciting as a new GPT-4 version or text-to-video generation capabilities, it’s still significant. With OpenAI’s toolset, businesses all over the world are developing a wide range of AI tools for both internal and external use. If certain essential security and cost-savings improvements aren’t made, those businesses might look elsewhere or, worse yet, decide against pursuing AI projects altogether.

Security improvements may be especially important to companies and employees, as well as the eventual customers using their AI tools. If AI can deliver stronger security features, both company and user data is safer.

OpenAI stated that its new features not only address security and cost-savings, but also some of the requests made by its customers. Ingesting 10,000 files into AI tools is now possible for businesses, compared to just 20 files earlier. Additionally, according to the company, OpenAI’s platform should be less expensive to run and easier to use thanks to new file management features and the ability to control usage on the go.

Now accessible are all of OpenAI’s new API features. The company intends to continue enhancing its platform with cost-saving and security features in the future.

Within the field of artificial intelligence, “small language models” have gained significant traction lately due to their ability to operate locally on a device rather than requiring cloud-based data center-grade computers. On Wednesday, Apple unveiled OpenELM, a collection of minuscule AI language models that are available as open source and small enough to run on a smartphone. For now, they’re primarily proof-of-concept research models, but they might serve as the foundation for Apple’s on-device AI products in the future.

Apple’s new AI models, collectively named OpenELM for “Open-source Efficient Language Models,” are currently available on the Hugging Face under an Apple Sample Code License. Since there are some restrictions in the license, it may not fit the commonly accepted definition of “open source,” but the source code for OpenELM is available.

A similar goal is pursued by Microsoft’s Phi-3 models, which we discussed on Tuesday. These models are small, locally executable AI models that can comprehend and process language to a reasonable degree. Although Apple’s OpenELM models range in size from 270 million to 3 billion parameters across eight different models, Phi-3-mini has 3.8 billion parameters.

By contrast, OpenAI’s GPT-3 from 2020 shipped with 175 billion parameters, and Meta’s largest model to date, the Llama 3 family, has 70 billion parameters (a 400 billion version is on the way). Although parameter count is a useful indicator of the complexity and capability of AI models, recent work has concentrated on making smaller AI language models just as capable as larger ones were a few years ago.

Eight OpenELM models are available in two flavors: four that are “pretrained,” or essentially a next-token version of the model in its raw form, and four that are “instructional-tuned,” or optimized for instruction following, which is more suitable for creating chatbots and AI assistants:

The maximum context window in OpenELM is 2048 tokens. The models were trained using datasets that are publicly available, including RefinedWeb, a subset of RedPajama, a version of PILE that has had duplications removed, and a subset of Dolma v1.6, which contains, according to Apple, roughly 1.8 trillion tokens of data. AI language models process data using tokens, which are broken representations of the data.

According to Apple, part of its OpenELM approach is a “layer-wise scaling strategy” that distributes parameters among layers more effectively, supposedly saving computational resources and enhancing the model’s performance even with fewer tokens used for training. This approach has allowed OpenELM to achieve 2.36 percent accuracy gain over Allen AI’s OLMo 1B (another small language model) with half as many pre-training tokens needed, according to Apple’s published white paper.

In addition, Apple made the code for CoreNet, the library it used to train OpenELM, publicly available. Notably, this code includes reproducible training recipes that make it possible to duplicate the weights, or neural network files—something that has not been seen in a major tech company before. Transparency, according to Apple, is a major objective for the organization: “The reproducibility and transparency of large language models are crucial for advancing open research, ensuring the trustworthiness of results, and enabling investigations into data and model biases, as well as potential risks.”

By releasing the source code, model weights, and training materials, Apple says it aims to “empower and enrich the open research community.” However, it also cautions that since the models were trained on publicly sourced datasets, “there exists the possibility of these models producing outputs that are biased, or objectionable in response to user prompts.”

Though the company may hire Google or OpenAI to handle more complex, off-device AI processing to give Siri a much-needed boost, Apple has not yet integrated this new wave of AI language model capabilities into its consumer devices. It is anticipated that the upcoming iOS 18 update—which is expected to be revealed in June at WWDC—will include new AI features that use on-device processing to ensure user privacy.

The company announced this week that users of Dingtalk, the workplace communication platform from Alibaba Group, can now turn to AI agents from outside providers for assistance with a variety of tasks.

Over 200 AI-powered agents with a focus on enterprise-facing features, industry-specific services, and productivity tools are available in DingTalk’s newly launched marketplace.

The platform also improved DingTalk AI Assistant, its in-house created AI agent, so it can now take in data from more sources, such as photos and videos.

“We think AI agents have the potential to be the mainstay of applications in the future. Ye Jun, President of DingTalk, stated, “Our goal is for DingTalk’s AI Agent Store to become a preeminent center for the development and interchange of AI agents.”

AI agents, a type of software, are being used by businesses all over the world to increase productivity.

In a survey conducted by Accenture last year, the overwhelming majority of C-suite executives (96%) said they thought AI agent ecosystems would offer their companies a big opportunity over the next three years.

DingTalk is keeping up, with over 700 million users as of last year.

In April 2023, the platform made its first use of generative AI technology when it collaborated with Alibaba Cloud’s large language model Qwen to introduce DingTalk AI Assistant.

In less than a year, Dingtalk’s AI capabilities have been used by over 2.2 million corporations, including about 1.7 million monthly active enterprises.

Artificial Intelligence

With the ability to create and share AI agents on the platform, the most recent development of DingTalk positions it as a formidable ally for Software-as-a-Service (SaaS) companies as well as individual developers.

Similar to conventional chatbots, these computer programs react to natural language commands, but they offer far more features. They are capable of carrying out both inside and outside of the DingTalk platform, from planning trips to producing insights from business analyses.

Ye stated, “We anticipate the rise of a thriving commercial marketplace and a flourishing ecosystem centered around AI agents.”

The more than 200 agents on DingTalk’s marketplace have cross-application integration and industry-specific knowledge.

AI agents created by third parties are required to apply for approval before they can be listed on DingTalk in order to guarantee a high standard of service.

Advantage of Multimodality

DingTalk has improved its AI Assistant even more by making it multimodal, or able to process data in multiple formats.

Up to 500 pages of text can be processed at once by Dingtalk AI Assistant, and users can request summaries to expedite work and learning.

Dingtalk AI agent is also capable of understanding images and extracting data from photos, pictures, videos, and other media thanks to Qwen-VL, Alibaba Cloud’s large vision language model.

DingTalk AI Assistant’s comprehension of visual cues enables it to produce subtitles, interpret images, transcribe videos, and even look up more information in response to a graphic prompt.

For example, someone who happened to take a photo of one of the temples dotted around the shore of Hangzhou’s West Lake could upload it. A quick synopsis of the site’s past would be provided to the user.