From Windows Central:
- Microsoft is working on a major new AI feature for Windows 11, which it plans to unveil during an event on May 20.
- The feature, dubbed AI Explorer internally, will be able to document and organize everything you see on your PC, and turn everything you do into a searchable memory with natural language.
- It looks like AI Explorer will launch first on new Arm PCs, as its system requirements list Qualcomm’s new Snapdragon X Elite NPU as a baseline.
- This would omit Intel’s new AI PCs from taking advantage of this year’s blockbuster AI feature.
It’s an open secret that Microsoft is working on a significant AI update for Windows 11 this year, timed with new “AI PCs” that are set to begin being made available in June. In fact, Microsoft is holding a special event on May 20, where it intends to unveil its plans for AI in Windows 11, along with new Arm-powered Surface hardware.
We already know that the big new AI feature coming this year is something called AI Explorer, which I detailed earlier this month. AI Explorer turns everything you do on your computer into a searchable memory using natural language. It works across any app and enables the ability to search for previously opened conversations, documents, web pages, and images. It can also understand context and streamline tasks based on what’s currently on screen.
Now, we’re starting to learn more about the system requirements for AI Explorer. Spotted by @thebookisclosed on Twitter, the current system requirements for AI Explorer are set to an ARM64 CPU, 16GB RAM, 225GB storage, and a Qualcomm Snapdragon X Elite NPU. This suggests that AI Explorer will be exclusive to new Arm PCs that are shipping this summer.
If we had to take an educated guess, an ARM64 CPU is required because Qualcomm’s Snapdragon X platform outputs 45 TOPS for its NPU, whereas Intel’s NPU is around 10 TOPS and AMD is 16 TOPS, making them far weaker compared to Qualcomm’s. Indeed, Qualcomm’s NPU is so powerful it could be considered a ‘Gen 2’ NPU, as that is where Intel and AMD will go next with their next-gen CPUs, due in late 2024 and early 2025.
Turns out Windows 11 build 26100 (purported 24H2 RTM) contains the AI Explorer requirements 📃 baked into the OS💠 ARM64 CPU💠 16GiB of RAM💠 225GiB system drive (total, not free space)💠 Snapdragon X Elite NPU (HWID QCOM0D0A)I guess that’s one way to drive ARM64 adoption
Link to the rest at Windows Central and thanks to F. for the tip.
Looks like a direct shot at Intel to PG.
One ‘use’ for AI would be to update software – isolating the parts it can’t do, to be handed to humans.
There is a program I want to use, and the only difference is that it is written for a 32-bit machine, and now the machines I want to run on are 64-bit machines.
Updating sounds incredibly tedious for humans, and I’ve been waiting for years. Maybe a very careful AI could do most of the job.
You are correct.
Programing is prime LLM territory. One use is for a model to analyze code and optimize it and/or comment it. One early report saw 70% productivity improvements.
And it has been available at Github since before ChatGPT went public via Github Copilot.
(Github being yet another microsoft business.)
LLM use has since been ratcheting up to where some folks (mistakenly) think there is no future in codesmithing careers.
https://www.windowscentral.com/software-apps/weve-shifted-the-responsibility-of-extracting-relevant-context-for-software-engineering-tasks-from-developers-to-the-ai-agents-microsofts-ai-based-framework-turns-developers-to-overnight-mere-supervisors
“Microsoft recently published a research paper that painted a clearer picture, highlighting the future of coding and developers as artificial intelligence becomes more widespread. The paper provides an in-depth analysis of AutoDev — an AI-powered framework designed to ‘assist’ developers with software development, ultimately redefining coding and automation.
“The research paper further details instances where the framework was tested and performed well by providing repositories to tackle technical software engineering work. It’s worth noting that the technology also ships with AI-powered capabilities to validate its outcomes. AutoDev supports file editing, retrieval, build processes, execution, testing, and git operations.”
—-
Some techies are spooked by the tool’s power and think it will replace them (much like hyperventilating writers fretting over “AI” fiction) forgetting that code generators need explicit instructions of exactly what the code is supposed to do. Like telling it to convert 32-bit to 64-bit or transcode x86 to ARM code, graphics routines from one GPU family to another. Codesmithing exists as a means to an end and LLM tools do not operate in a vacuum. At its most basic, somebody needs to feed it a mission, be it a prompt, flowchart, or existing code.
There will still be a need for humans deciding what the code is to do. But the code quality should improve and have less bugs. And be quicker and cheaper to generate. One industry in desperate need of this boost is video games. (A whole story right there.) One of many ways LLMs will be used in gaming.
“
Bloomberg has a good writeup on Github Copilot’s current capabilities:
https://www.msn.com/en-us/money/other/microsoft-s-ai-copilot-is-starting-to-automate-the-coding-industry/ar-BB1lLIYz?cvid=debae6b5e4664595e22197330d4979ff&ei=66
“Three years later, and now infused with the latest version of OpenAI’s GPT-4 technology, GitHub’s Copilot can do a lot more, including answering engineers’ questions and converting code from one programming language to another. As a result, the assistant is responsible for an increasingly significant percentage of the software being written and is even being used to program corporations’ critical systems.
“Along the way, Copilot is gradually revolutionizing the working lives of software engineers—the first professional cohort to use generative AI en masse. Microsoft says Copilot has attracted 1.3 million customers so far, including 50,000 businesses ranging from small startups to corporations like Goldman Sachs, Ford and Ernst & Young. Engineers say Copilot saves them hundreds of hours a month by handling tedious and repetitive tasks, affording them time to focus on knottier challenges.”
GitHub emphasizes that the tool is an assistant, not a substitute for human programmers, and has put the onus on customers to use it wisely. Robust guidelines are required to prevent lazy programmers from simply accepting what Copilot suggests, said GitHub Chief Executive Officer Thomas Dohmke. He expressed confidence that engineers would keep one another honest.
“The social dynamic of the team will make sure that those that are cheating by accepting code too fast and that don’t actually go through the process defined by the team, that code will not make it into production,” he said in an interview.”
More at the source.
Current generation AI PCs are a tad weak to run full Copilot and equivalent models locally since that requires a 40 Trillion Operations per second (TOPS) Neural Processing Unit (NPU) in addition to the old standard CPU and GPU. The recently introduced BROADCOM mobile chip can do that.
However, in classical IBM-style account control, Intel has preannounced their next generation chipset (Codenamed Lunar Lake) will do 100TOPS. It is slated for late ’24-early ’25.
At the same time, optimized-function SLMs are showing up that run in under 10TOPS so the current AI PCs aren’t useless but much like the late 80’s, the time to jump into the new era is when you have an immediate need because “new and improved” will be arriving regularly every six months or so. Prices will drop and software will improve on a similar cadence, rendering older tools laughably limited in a couple of years.
If you can conceive a use for this tech that isn’t available yet, wait a bit. It’ll get here soon enough.