Michael Leonard

A spot-on description from Urs on the core challenge and essence of Firefly Clock Synchronization. paper: https://lnkd.in/gQsScQhu Pooria Namyar Yuliang Li Weitao Wang KK Yap Junzhi Gong Chen Chen Peixuan (Shane) Gao Devdeep Ray Gautam Kumar Yidan Ma Ramesh Govindan

59

I like to think I’m good at making decisions. I make a lot of decisions. But yesterday I had an uncomfortable experience that made me look deeper into how I prioritize. This piece investogates focus, execution, opportunity cost, and the uncomfortable difference between pausing an idea and killing it.

20

2 Comments

AMD Ryzen 9 9950X and DDR5 NVMe mean nothing if your power dies during the trading session. Redundant power, failover routing, and RAID-protected storage are infrastructure quality. CPU specs are marketing. Fund managers and EA operators know this. They judge VPS on uptime SLA and disaster recovery, not processor model. Silent failures cost more than obvious downtime. A power blip that triggers failover lag. Partial disk corruption nobody notices until the EA crashes mid-trade. Those kill trust faster than any outage. CPU cores get printed on the sales page. Redundancy lives in the data center design. In the backup systems. In the operational discipline that catches problems before traders do. Trust is built through consistent operational excellence, not published specifications. The traders running serious capital already know the difference. The question is. does your provider.

2

2 Comments

Something this blog nailed is that leaders blame models & hardware for slow AI, when the network is the real culprit. Bounce data across the public internet & latency turns into a guessing game. Proximity is the fix hiding in plain sight. Trace the lag: https://eqix.it/4y49A94

6

AT&T's chief technology officer didn't dodge the question this week. He confirmed it. Fewer employees, five years from now. Not a rumor. Not a leak. A direction, stated plainly. Jeremy Legg also did something most executives avoid: he pushed back on the number everyone was repeating. Reports pegged the cuts at 85,000 workers. Legg called that wrong, telling WIRED it was "a direction for headcount, not an agreed endpoint." AT&T has already gone from roughly 139,000 employees to about 131,000 this year. The company is chasing $4 billion in annual savings by 2028, built on automation, AI, and a shift from copper to fiber that needs far fewer truck rolls, fewer manual repairs, fewer people climbing poles. Here's what got buried in the headline: AT&T still expects to hire. Not fewer jobs. Different jobs. People to build and govern AI agents. Technicians who maintain fiber, not copper. The company isn't shrinking into nothing , it's re-shaping around what the new infrastructure actually requires. We are familiar with these situations: Leadership announces "efficiency gains," everyone hears "layoffs," and the real story what kind of organization is being built on the other side never gets told. Legg told it anyway, publicly, with an actual number instead of a rumor. So the question for leaders watching this isn't how many roles AI will eliminate. It's what roles will only exist because of AI and whether we're already building the people who can fill them. Decisions I'd take from AT&T's move if I were running a technology org right now: 1-Name the target, not the myth. A vague fear of automation is worse than a specific, even uncomfortable, number. 2-Fund the new roles before you cut the old ones. Efficiency without reinvestment is just shrinkage with better PR. 3-Treat infrastructure change as workforce strategy. Copper to fiber wasn't just a network decision. It was a five-year people plan disguised as an engineering one. Automation doesn't ask permission to change what an organization looks like. Leaders decide whether it changes for the better. Where is your organization telling the truth about the shift already underway and where is it still hoping nobody asks? #Leadership #ArtificialIntelligence #EnterpriseAI #TechnologyLeadership #DigitalTransformation #Telecommunications #AIInfrastructure #FutureOfWork. https://lnkd.in/eKJP7MjQ

5

2 Comments

Very interesting, and I think I agree with what he's saying to a point. However, I would argue two counterpoints: 1. There IS a meaningful / perceivable difference between analog and digital audio. A preference for analog may be rooted in nostalgia for some people ("records sound like how I remember music sounding!") but even people who have no nostalgia for LP's can tell a difference, and many of them immediately prefer the way "lower quality" analog recordings sound. It's similar to why The Retro Roadshow only uses "inferior" CRT displays when we exhibit classic video games: the handcrafted art for those games was painstakingly designed to take advantage of the limitations of CRT technology, and they literally don't look right on modern "high resolution" displays without significant effort to digitally replicate the "lower quality" effect of a CRT display. 2. A future where the vast majority of software is "prompted" rather than coded is ALSO a future in which the vast majority of software developers have no idea how software actually works. As a result, they will be limited to producing only the code that their AI tools already know how to produce. It will result in a huge increase in the quantity of apps, and a huge decrease in innovation, experimentation, and (I suspect) subjective quality. AI can only repeat and remix what it has seen done before, it cannot create truly new work. For that, you need human ingenuity, which will be handcuffed when a whole generation of "developers" don't understand how anything actually works. AI is obviously an astonishing new tool that opens up countless avenues towards increased productivity, but I desperately hope that it doesn't replace the essence of human creativity. AI has to support and empower US - otherwise, what's the point? Thanks to Sarai Dunsany for sending me this post! #AI #SaturdayDeepThoughts

5

8 Comments

Re-posting for whoever may need to see this. Steal this idea -- users moving from VMware to KVM/Proxmox/whatever, esp using NFS shared storage, can use a narrow strategy completely skips VDDK/ etc. by converting a vmdk to qcow2 by metadata/header manupulation, vmdks of arbitrary disk size (think TBs) to a qcow2 in guaranteed constant time O(1). Migration might require some additional virt-v2v-inplace and other steps that your AI-coded migration scripts/app can easily figure out and incoporate. The outcome: migrate terabyte sized VMware VMs in seconds (maybe minutes), not hours and days.

1

AWS's inevitable destiny: becoming the next Lumen - latest from Corey Quinn in which he posits that AWS is doomed to become a background lower-margin infra provider: https://lnkd.in/gFMuPcW2

5

2 Comments