Quick Takes: February 2025
I’ve got some longer technical blogs in the works. For this week, it’s time again for some of my “Quick Takes”: articles I thought interesting and noteworthy, that you might have missed. Not that other blogs I’ve seen are not equally interesting!
Large-Scale IPv6
Thank you, I”ll duck the “do we need IPv6” debate and cut to the chase. I’ll note I’m not holding my breath for the “flag day” when everyone hustles to get IPv6 in place because most devices and apps expect it.
China is building new IPv6-only cities, or parts thereof. The following is well-written and not overly technical, and good reading for those planning a large IPv6 deployment.
And for those contemplating IPv6, you might want to check out the experts at HexaBuild for training and consulting services.
AI Misc
I’m having fun skimming the firehose of information about AI. And while I’d love to be more hands-on, I have to keep reminding myself I’m allegedly retired and limiting myself to about 8 hours/week of tech reading and blogging.
I came up with an AI question/idea, posted it somewhere (BlueSky probably), and shortly thereafter saw something similar briefly mentioned in a posting about DeepSeek.
Namely, to keep LLM size and training time more reasonable, why not create LLM’s dedicated to various technical or other areas of expertise, and have the “main LLM” be capable of recognizing when it has to delegate tasks to one of the “focused LLM’s” or “LLM domain experts”?
I’ve been keeping an eye out for further mentions of this sort of approach in conjunction with DeepSeek, but not seen anything. “Modular AI”, to coin a term?
I’m feeling a little cynical about the US AI firms’ huge LLM’s focus. The conspiracy theory version of that is probably something along the lines that it was all a ploy to justify huge capitalization, with code and run-time optimization to follow. And no, I don't subscribe to that theory. I do see commentary that people think the huge cost factor was to scare off more competition (putting it bluntly).
Some perhaps novel or interesting AI blogs/links: Tools for Service-Provider Agentic AI
It discusses what else is needed to provide agentic AI services.
Another good discussion: Techstrong.AI: Companies Looking to Get Started in AI May Be Overthinking It. In particular, there are different AI lifecycles, and you may not need all of them. As in, consume LLM’s but maybe limit yourself to RAG. It helps to make this explicit!
And the MIT Tech Review magazine recently printed a number of good high-level AI articles, albeit some behind a paywall.
And one curious one (paywall): These AI Minecraft Characters Did Weirdly Human Stuff All on Their Own
Something from their up-river Cambridge competition (last November): Watching the Generative AI Hype Bubble Deflate
And an AI learning resource: InstructLab bsky post
HPE, IOT/Edge, and Cellular
In a previous blog covering IOT and vendors, I concluded that HPE didn’t seem to have much footprint in that market. A recent GestaltIT blog clued me in, that their IOT focus is just different: more on leveraging their existing hardware to support “indoors IOT”, to coin a term. Which includes Retail and Edge support.
HPE Expands Wireless Networking Portfolio for the Edge: HPE announced a cellular bridge. This provides a cellular primary or backup path for Edge e.g. Retail sites, also temporary or new locations. This sounds like the cellular access device Meraki recently announced that I blogged about.
The article also mentions the CX 8325H switch ideal for kiosks and pop-up stores where small size matters. It has 18 ports and fits compact spaces, with lower power and cooling requirements. It notes that most such sites connect via private 5G.
领英推荐
Qualcomm IOT Connectivity
Qualcomm announced modules that can be used for IOT smart home, smart appliances, and other applications. Not consumer tech unless you’re into home-brew hardware. Interesting (to me) because it suggests that cost and form factor are shrinking, enabling smarter IOT and appliance uses.
QCC730M is dual-band micro-power Wi-Fi 4 for battery-powered IOT use, e.g. IP cameras, sensing, and smart locks.
QCC74xM is a programmable RISC-V based small module that does Wi-Fi 6, Buletooth, Thread, and Zigbee, with CAN abnd Ethernet interfaces. They suggest it is useful for products like IOT hubs or hotspots, and smart home applications.
Virtual 5G Cellular RAN
Virtual RAN is where the cellular Radio Access Network runs on general-purpose chips, rather than specific cellular hardware. The rationale would be to leverage the economics and rapid performance improvement of general-purpose chips. The article discusses the factors affecting that.
Selector AI CoPilot
Selector / John Capobianco have now made their Packet Copilot available, for the price of filling in a short form (and likely subsequent emails).
Key features cited:
Links:
Miscellany
Reminder: you may want to check back on my articles on LinkedIn to review any comments or comment threads. They can be a quick way to have a discussion, correct me, or share you perspectives on technology.
Hashtags: #PeterWelcher #CCIE1773 #QuickTakes #SelectorCopilot #HPEIOT #AIMisc? #IPv6City
FTC disclosure statement: https://www.dhirubhai.net/pulse/ftc-disclosure-statement-peter-welcher-y8wle/
Twitter: @pjwelcher
LinkedIn: Peter Welcher, https://www.dhirubhai.net/in/pjwelcher/
Mastodon: @[email protected]
? ?