Mitchell Hashimoto's ghost in the machine; Microsoft open sources Phi-4

Welcome to Runtime! Today: HashiCorp co-founder Mitchell Hashimoto is back with a new terminal emulator, Microsoft moves one of its Phi LLMs into the open, and the quote of the week.

(Was this email forwarded to you? Sign up here to get Runtime each week.)

Ship it

Terminal velocity: Most people never need to interact directly with their computer's operating system through the command-line interface, but software developers do, and they have opinions about which terminal emulator is best for the job. Over the holiday break HashiCorp co-founder Mitchell Hashimoto released Ghostty, a new open-source terminal emulator that works with MacOS and Linux.

"While there are many excellent terminal emulators available, they all force you to choose between speed, features, or native UIs. Ghostty provides all three," Hashimoto said in a post that provides more details on how it works. He said nothing about any potential business plans for the project, but Hashimoto's open-source projects turned into big business for HashiCorp.

Open season: Security information and event management (SIEM) software has become widely used, but automating how that software works with other tools can be a lot of work. On Tuesday Exabeam announced that its New-Scale Security Operations Platform is now compatible with the OpenAPI specification (OAS), which "defines a standard, language-agnostic interface to HTTP APIs which allows both humans and computers to discover and understand the capabilities of the service.".

Security teams tend to create "playbooks" for responding to certain types of threats, because responding as quickly as possible is a huge factor in limiting the damage. With support for OAS, Exabeam customers will be able to incorporate other security tools in their playbooks with far less configuration and implementation work required to link those tools together thanks to the common API spec, according to the company.

Phi-ghting chance: It's not clear how much demand there is for open AI models among enterprises, according to a report this week from The Information. But open models do play an important role helping researchers and others understand how these models work, and Microsoft added one of its Phi models to that list this week.

On Wednesday Microsoft released Phi-4, a "small language model" first introduced last month, on HuggingFace under the permissive MIT License. Anyone interested in using the model can download the weights used to build it, which allows them to customize the model for whatever purpose they like.

Twinning: Talk of the industrial metaverse feels very pre-ChatGPT at this point, but some of its original proponents are still working on that dream. This week at CES Nvidia unveiled "Mega, an Omniverse Blueprint for developing, testing and optimizing physical AI and robot fleets at scale in a digital twin before deployment into real-world facilities."

The idea is that Mega would allow factories to use Nvidia's hardware to test "robot brains" in virtual reality before letting them loose in production to move goods around a warehouse or execute tasks on an assembly line. “In the future, every factory will have a digital twin,” Nvidia CEO Jensen Huang said in his keynote speech, according to VentureBeat, and Runtime is taking the under on that bet.

This one goes to 11: Database companies are scrambling to improve their vector search capabilities in hopes of winning generative AI business, and Oracle is no exception. This week the company introduced Exadata 11M, the newest version of its database system, touting its improved performance for vector queries.

Specifically, Oracle said Exadata 11M delivers "up to 55% faster persistent vector index (IVF) searches" and "in-memory vector index queries (HNSW) are up to 43 percent faster." The Exadata product is a combination of hardware and software that customers can run in their own data centers, in Oracle's cloud, or in any of the Big Three clouds.

Stat of the week

One reason software developers are excited about the potential of AI coding assistants is that a much greater portion of their time than most people realize is spent on really, really boring tasks. According to a new survey conducted by Harness, "78% of developers spend at least 30% of their time on manual, repetitive tasks," and anything that helps them plow through that busywork is going to attract attention.

Quote of the week

"I think it's a huge misconception to come in and think that you're just going to plug in the agents and it is just magically going to figure it out." — Alex Reinke, co-founder and co-CEO of Celonis, dropping a truth bomb on the agentic AI frenzy.

The Runtime roundup

X plans to spend $1 billion on AI servers from HPE, according to Bloomberg, beating out Dell and Supermicro.

Microsoft Azure customers using its East US 2 region experienced some hiccups Friday after the company reported problems stemming from "a configuration change in a regional networking service."

Thanks for reading — see you Tuesday!