Sabitlenmiş Tweet
Joshua 1:6 Be strong and courageous, because you will lead these people to inherit the land I swore to their ancestors to give them.
English
Keith Townsend
85K posts
Keith Townsend
@CTOAdvisor
CTO Turned Advisor | Helping Vendors Resonate and IT Leaders Execute. Engage with my virtual twin https://t.co/7fh1X8hbEJ. Independent Advisor.
Chicago, IL Katılım Haziran 2009
498 Takip Edilen24.7K Takipçiler
This is a day to remember. My theatre geek granddaughter is having a bonfire to celebrate high school graduation. My daughter put on Hamilton, and the 20 theatre geeks went wild.
I wouldn't even be able to describe this to a 25-year-old Keith.
English
I’ve been experimenting with local AI on my Nvidia Spark all day. I’m comparing local 30B to 90B models to o3 for both job completion and loop control. I’m testing code repository loops. o3 doesn’t need to loop and finishes in 7 seconds consistently. The local models all take 60 minutes on the Spark and never successfully solves the coding challenge.
This is with general purpose models. I’m testing Qwen Coder 2.5 over the next few hours to see if it closes the gap on both latency and capability.
English
Your UI doesn't matter and agents consume your API. Where is the moat for your SaaS product now?
Good @a16z post that looks at this move to "headless" and what the new moats and sticky points are. tl;dr; The data layer is still the source of value.
a16z.news/p/is-software-…
English
Gemma 4 31B running in vLLM doing some hybrid AI loop control experimentation. This is where vLLM earns its due. I tried running this experiment with Ollama and it just wasn't up to snuff.
English
@IsForAt I purchased two generations of Surface Books. May rank as my all time favorite device for writing.
English
Who is a Surface fan?
I am asking this to just check in on the hardware business I built, by finding fans and getting the product team to engage and build the best PC ever built for those fans.
I am worried the Surface business is dead. it was important when there were lots of fans who knew it was the only tablet/laptop combo for work.
they have done nothing in the last 6 years, and have ruined the brand by munging it with concepts no one understands like "copilot plus PCs".
please tell me there are fans out there still...
English
If you aren't paying, you're the product. My business is no exception.
It's been almost a year since I left Futurum to restart my independent journey. My premiere product is The Buyer Room. A practioner-led discussion between IT Vendors and their target buyer persona.
What does that have to do with @thectoadvisor content? I've built substantial frameworks over the past year that are free to download. How exactly does that make money if they are free?
TLDR;
I built trust with you, the IT Buyer, via my content. That trust is used to sell vendors buyer room sessions. Vendors gain insights they couldn't otherwise gain via focus groups or analyst surveys. I get additional fuel for my frameworks.
The flywheel. Read more and see a sample of a buyer room artifact.
linkedin.com/pulse/why-ente…
English
Who is at Google I/O next week.
My first time attending. What should I expect?
English
@CTOAdvisor @QuinnyPig oh sweet summer child
"Billing for EC2 Mac instances is per second with a 24-hour minimum allocation period to comply with the Apple macOS Software License Agreement"
English
These things are over $9100 a month, making it incredibly hard for me to counter accusations that the cloud is overpriced. Yes, there's value in EBS/IAM/the rest of the ecosystem, but not that much value, frens...
CloudSensei@TheCloudSensei
Announcing general availability of Amazon EC2 M3 Ultra Mac instances dlvr.it/TSXx2s
English
@JoshLuedeman @NVIDIAAI @MrsCto Ha! She is very protective of my learning budget. If I don't spend $ xx,xxx on training/learning every year, she gets very concerned that my skills are getting stale.
English
GIF
English
I started my experimentation with Gemma 4, now I'm using Nemotron. I've been putting this @NVIDIAAI Spark to work the past couple of days!
Keith Townsend@CTOAdvisor
Everyone demos agents. The more useful question is who controls the loop. I wrote a requirements document for an experiment I have not run yet. The unusual part is not writing the requirements document first. The unusual part is publishing it before running the experiment. Why publish it now? Because the design decisions are the content. I have a DGX Spark in the lab. The interesting question is not whether it can run a model. Local inference is becoming practical enough that “can it run?” is the less interesting question. The harder question is whether a local model can participate in a real agentic system when execution, validation, and judgment are separated deliberately. My working hypothesis: local models may be excellent workers but unreliable governors. A local model can summarize, extract, classify, and propose a decision. But should it own the loop? Can it tell when evidence is weak? Can it detect when it is summarizing instead of deciding? Can it know when to stop? Can it know when local judgment is not enough? The architecture under test is not all-local or all-cloud. It is local execution, deterministic validation, and selective escalation to a stronger reasoning model when the workflow requires judgment. In plain English: local workers, deterministic control, and outsourced judgment only when needed. That is the Layer 2C question in my 4+1 AI Infrastructure Model. Where does judgment live? Full post: “Who Controls the Loop? A Requirements Document for Local-First Agentic AI” thectoadvisor.com/blog/2026/05/1…
English
@JoshLuedeman @NVIDIAAI Ha... I have a new problem. There's enough value that @MrsCTO recommended that I buy a 2nd one!
English
@CTOAdvisor @NVIDIAAI I keep going back and forth. Honestly I was hoping you would say a lot so I could have an excuse to pass on it 😂
English
@JoshLuedeman @NVIDIAAI It's not noticeable from an ambient perspective. It does get "warm" to the touch but not hot. It's actually how I tell if it's running because there's no power light!!!
English
@CTOAdvisor @NVIDIAAI How much heat does that thing generate?
English
@QuinnyPig @awscloud Now I am curious. Splunk is a money pit.
English
@CTOAdvisor @awscloud It's somehow less than Azure, Splunk, etc. And it tiers down to something more reasonable at volume.
English
“Be sure to instrument your calls with OTEL” says @awscloud, who charges 50¢ per GB of ingested CloudWatch Logs.
mark@r_marked
"Why can't you just go back to the old model of charging based on messages?"
English
I've been building this test all day. It has morphed into a coding-based test because my original test case was too subjective to test deterministically...
This is fun but at some point I have to put it down!
Keith Townsend@CTOAdvisor
Everyone demos agents. The more useful question is who controls the loop. I wrote a requirements document for an experiment I have not run yet. The unusual part is not writing the requirements document first. The unusual part is publishing it before running the experiment. Why publish it now? Because the design decisions are the content. I have a DGX Spark in the lab. The interesting question is not whether it can run a model. Local inference is becoming practical enough that “can it run?” is the less interesting question. The harder question is whether a local model can participate in a real agentic system when execution, validation, and judgment are separated deliberately. My working hypothesis: local models may be excellent workers but unreliable governors. A local model can summarize, extract, classify, and propose a decision. But should it own the loop? Can it tell when evidence is weak? Can it detect when it is summarizing instead of deciding? Can it know when to stop? Can it know when local judgment is not enough? The architecture under test is not all-local or all-cloud. It is local execution, deterministic validation, and selective escalation to a stronger reasoning model when the workflow requires judgment. In plain English: local workers, deterministic control, and outsourced judgment only when needed. That is the Layer 2C question in my 4+1 AI Infrastructure Model. Where does judgment live? Full post: “Who Controls the Loop? A Requirements Document for Local-First Agentic AI” thectoadvisor.com/blog/2026/05/1…
English