Retired tech exec, playing music in an idyllic beach town.
Vero Beach, FL
Joined December 2025
- Tweets3.2K
- Following169
- Followers288
- Likes365
Not a hot take at all. Anyone who has supervised a team of superstars knows that it's really hard work.
LLMs are a reflection of how we think, for better or worse.
Changing the value of a few intermediate "neurons" can produce wide swings in behavior.
The more I study LLM behavior, the more I learn about human behavior.
My answer to Fermi's Paradox?
If there are many trillions of galaxies each with many billions of stars, planets, etc. -- why haven't we been contacted by alien life?
I think it's because we haven't advanced nearly enough yet.
Basically, it's an "us" problem. Held at arms length, we're a sloppy, emotional and violent species.
I've found living in Florida generally awesome.
It compares well with the SF Bay Area, the Pacific Northwest, the Boston burbs, Boulder and a few other cool places I've lived.
I'm in a beach resort town where every day feels like, well, living in a beach resort. It doesn't suck.
It's a beach day today. This morning everyone was out in the great warm weather hiking, biking, rowing, fishing or whatever they do outdoors.
Classic cars and motorcycles everywhere. Lots of expensive toys as well. Young families out and about. Everyone does their thing.
When people ask "what do you do?" they mean what do you do to keep yourself busy, and not necessarily what line of work.
There are golfers, boaters, pickleballers, tennis players, cruise enthusiasts, charity leaders, travelers and so on. And, of course, plenty of folks still holding down a job.
I run a great local band. People love good live music here. It's something I've always wanted to do.
By default, most everybody is happy, polite and engaging. It's expected that you'll make small talk with just about everyone you meet, including service people.
There are at least 40x more churches in town than liquor stores or pawn shops. Lots of people carry. It's about a safe a place as you could imagine.
They call my town "Mayberry By The Sea", and there's a certain truth in it. Everyone loves being here and wants to keep it the special place that it is.
The town and county aren't anti-growth, but they do keep a close eye on things so Vero Beach doesn't end up like West Palm Beach.
In all fairness, the summer humidity can break you after a while, though. If you have a place to escape to farther north, that's ideal. The snowbird thing works.
Just my two cents for all you retirement-adjacent folks who might be looking around for your piece of paradise.
Lutz nails it here.
Many conservative voters refuse to participate in the polling charade and lump obviously biased results in the media under "more fake news".
What matters more than someone's opinion is who is engaged enough to vote, and who isn't.
And the only way to determine engagement is via focus groups. You literally pay people to sit in a room and answer questions honestly.
At one point in my career I was running dozens of focus groups a year, then later would write big checks to have professionals do this on our behalf.
The other huge flaw in political polling is that there's FAR more money behind swaying the results than determining reality.
🚨🇺🇸 Pollster Frank Luntz is NOT buying the blue wave...
He says Trump voters have a tendency NOT to participate in polling, and that many of those polled may not actually end up voting.
He also cast doubt on whether pollsters are even using the right measurements or asking the right questions.
Underestimating GOP turnout by even a small turnout could mean the blue wave is more like a blue puddle...
We'll know who's right in 13 days!!!
Writer: Michael
I would suggest that all investors stay a safe distance away from Anthropic.
They're not going to be a winner. It might be a good time to sell the company to IBM or similar.
Training costs wiil continue skyrocket and market prices for inference will continue to drop.
Just like any gold rush, the cost of production continues to skyrocket while market prices fall.
By comparison, Elon is building the full stack: chips, power, data centers in space and rockets to get them there.
Now that's a moat.
Yesterday he shared that he'd put his best team on building gas turbine blades, which is the current bottleneck to AI scaling. And that Terafab is moving ahead quickly.
Open source is moving fast as well. Deepseek is wild, they've released a "service provider" stack: most everything you need to stand up AI as a service.
And there will be hundreds if not thousands of people doing this very inexpensively.
If I were an investor, I'd consider shorting the Anthropic IPO.
🚨 ANTHROPIC's NUMBERS ARE CONCERNING -
Revenue ~$16,300,000,000
Net Loss ~$80,700,000,000
IPO Valuations ~$2,000,000,000,000
@grok How long can it run before declaring bankruptcy if it keeps on buring like this?
I'm still processing everything Elon said during the first 20 minutes of the SpaceX earnings call.
It was surreal.
It should be named themost mind-blowing earnings call in history (after his rocket car demo) and ended with him throwing out a $500B - $1 trillion ARR.
Apple is running at $466B, Amazon at $717B.
Just watching him talk and think is fascinating.
xitter.gq/i/broadcasts/1PKqrNbLZ…
And with DeepSeek's release, they've created a new category of "independent AI service provider" -- smaller players who can provide a higher level of customer service at a more reasonable cost than the big guys.
Wait six months, they'll be everywhere.
After watching Dems get over-polled for decades, I asked myself "why is this"?
It can't be completely explained away by "pay to say" pollsters. Respected names get it wrong regularly as well.
Frank Lutz nailed it: there are conservatives who won't share their opinions with pollsters. I am one of them, and I know there are many others.
Some will try to apply a correction bias of ~3.4%, supposedly based on past polling failures, which is a weak attempt to recognize the phenomenon.
You can't measure what refuses to be measured. there is no way to know how many of us there are, or if we will turn out to vote.
The only poll that matters is on election day.
💥NEW: NBC's Steve Kornacki *DUMPS COLD WATER* on The Today Show viewers' hopes of Dems winning Senate💥
Kornacki noted Democrat leads in key Senate races are "tenuous" — and that polling frequently underestimates Trump voters.
He asserted Republicans now confidently view polls where Democrats have slight leads as likely GOP victories.
Kornacki gave specific examples, such as Ohio and Iowa.
Remember the infamous Ann Selzer Iowa 2024 poll?
Kornacki particularly highlighted the Kansas Senate race in 2020. He said the Democrat was polling well — but Republican Roger Marshall ultimately smoked her in a landslide.
Marshall is now supposedly polling extremely tightly against his Democrat opponent once again.
Kornacki also pointed out that even blue states like Michigan and Minnesota appear remarkably close in polling.
He reminded viewers that Abdul El-Sayed barely won his primary after polling projected a huge victory.
Due to the tightness of these Senate races, underestimating Republican turnout by even a small margin could make all the difference.
Frank Luntz also recently warned Trump voters "don't respond" to pollster phone calls and asserted he was "skeptical" that polls were accurate.
Nobody is buying the polls anymore.
Democrats could be in for a rude awakening in 2 weeks.
One of my favorite Muskisms is the ability to predict the future being the highest form of intelligence.
I've given that a lot of thought, and it's painfully true.
I've seen many otherwise bright people get blindsided by easily predictable events. "What happens next?" can be really hard for some.
AI (now evolving into true super intelligence) won't trigger mass unemployment, but will force many to be supervisors rather than individual contributors.
Every bot team needs a human involved. There's a limit to how much bot work a human can inspect. More bots, more human supervisors/inspectors.
No supervisors paid to pay attention, no useful bot work. Just like with human workers.
The total amount of useful work performed by the workforce (a rough proxy for GDP) will increase by one order of magnitude, or more.
Plenty of jobs for those who can do it, or are willing to learn. Bot herders will be paid much more than non-bot herders for obvious reasons.
The predictions around having to work less and make more are likely the most accurate ones.
🤖 Made with AI
Their "networks" will likely be rebranded Starlink services.
The only company putting enough satellites in space is SpaceX for the foreseeable future.
Mobile companies have two core competencies: running a network, and billing for it.
Now they're just going to be retail outlets for Starlink services.
BREAKING: Four European telecom giants are joining forces to challenge Elon Musk’s Starlink.
Deutsche Telekom, Orange, Telefónica and Vodafone have signed a letter of intent to explore a joint bid for spectrum to deliver satellite connectivity directly to phones.
Four telecom giants. Still no Falcon 9. Still no Starship. SpaceX has no competition. Period.
This is great.
Very few mid-level dweebs are going to risk federal prosecution by knowingly committing a crime.
The obfuscation is fun to watch.
Previously, the CA Secretary of State "enforced" the rules. Hahaha. Ooops, now it's the DOJ.
It's simple: if any of the 20k suspect ballots get counted the DOJ will prosecute them with a federal crime.
Checkmate.
Looks like the organized protest industry is falling on hard times.
They can't seem to get it up, no matter how hard they try.
The dark money is gone along with the paid protestors. It just ain't the same without millions in funding.
Remember when everyone was protesting data centers? That money is gone too, so far fewer protestors.
Same with Palestine or whatever cause-of-the-day used to be protested.
When the money leaves, so do the protestors.
You'll see the Roadster flying car tech on the moon and Mars before long.
The existing pressurized gas system has 6x the range on the moon, maybe 30-40 sec of light as there's 1/6th the gravity. Fans, of course, are useless.
That could get you out of a few tight spots.
The main drawback of the tank is its weight fully loaded: 250 kg or 530 lbs here on Earth. But on the moon, a tank with six times the capacity wouldn't be hard.
Now we're talking 3-4 minutes of flight. Yes, you'd need to provide your own gas to compress, presumably by heating lunar rocks.
Very useful for exploring on the Moon.
Mars has 38% of the gravity of Earth, so a bigger/heavier tank could be supported as well.
I could easily see a lunar or martian vehicle built on Roadster tech that would drive normally but be able to jump obstacles and traverse canyons.
As Elon said, "where we're going, there are no roads".
I got an up-close look at the engineering INSIDE the new Tesla Roadster at yesterday’s unveil, and it’s pretty crazy. After watching this thing drive upside down, launch into the air, and do flips, I got to see the actual hardware that makes it all possible.
1/ The Electric drive unit is the heart of the Roadster’s electric powertrain, delivering power directly to the wheels.
2/ The battery pack has rows of cylindrical battery cells, with cooling systems and electrical connections running throughout. This is what powers the car before the rocket tech comes into play.
3/ The rocket thrusters has multiple cold-gas nozzles. Tesla says the four main rear thrusters blast compressed air faster than the speed of sound, producing more than 80,000 newtons of thrust.
4/ The SpaceX-inspired pressure tanks has two massive carbon-fiber-wrapped pressure vessels with a combined capacity of 600 liters, holding approximately 250 kg of compressed air. And the wild thing is the Roadster can recharge its own air supply, no special rocket fuel is required.
5/ An active rear wing has a variable-pitch wing that automatically adjusts downforce depending on what the car needs. It allows it to get more grip when cornering, less drag when going fast.
6/ The underbody engineering is the hardware underneath that brings everything together, including the systems that help the Roadster stick to the ground and launch into the air. And then there’s the fan suction system, which Tesla says can generate up to 2 g’s of suction force, enough to hold the car upside down.
Tesla and SpaceX have built an electric sports car with a battery, electric motors, rocket thrusters, high-pressure air tanks, active aerodynamics, and a suction system powerful enough to let it drive upside down. And yesterday, I watched it actually happen with my own eyes…
I’ve seen some incredible engineering from Tesla over the years, but getting this close to the hardware and then seeing what it can do was on a whole another level.
This is what happens when you bring together some of the best automotive and rocket engineers in the world and let them push the limits of what’s possible.
Great job Tesla! 👏
One of the reason many people irrationally dislike Elon is that he's so damn successful at so many things.
He's redefined existing industries and created entirely new ones.
And having the time of his life doing it.
Flying cars, rockets to Mars, robots, space AI ... he's the guy who is making the future real. He makes the impossible look easy.
Untold legions of young builders have now been inspired by his ethos.
History will be kind to him.
Elon Musk walked onto the Tesla Roadster stage so casually that people thought the unveiling would be boring.
Little did they know he was about to deliver one of the most INSANE unveils ever!
Then he said:
“You’re gonna see some crazy stuff tonight. We need an audience to tell if it’s not SI/AI.” 😂
ELON IS A F*CKING GENIUS!
Pay attention here, history is repeating itself.
Back in the day, everyone demanded their own server(s). VMware came along and industrialized virtualization.
Before long, everyone was running a private cloud.
The same opportunity exists for model sharing. One honking server (or, more typically, a cluster of servers) being shared across a trusted group of power users.
Right now, that market space is wide open. Expect lots of activity here before long.
Actual Computer turns your NVIDIA DGX Station into a department-scale AI factory.
Read the full case study:
actual.inc/company/blog/case…
It's 2026 and the race for desktop AI is on. Apple is a player.
You can now run a frontier-class open model on your desktop with a ~$25K rig - a pair of maxed out Mac Studios.
No subscription required.
That may sound exotic to some, but it's well within the range of many.
Last year at this time, nothing like the model or its platform existed.
When supply starts to catch up to demand (perhaps in 2027), you'll see prices drop as they have before.
Open models like this will put a big dent in the future IPO aspirations of the proprietary model vendors.
America is SO back
Zuck just announced Muse Spark Open is dropping later today, built with Mac Studios in mind
This is bigger than you think
The Meta Muse model was trained specifically for agent tasks. Now we can run it locally, plugged into agents like Hermes
This is the future of AI
Local, open weights models powering local open source harnesses running on our own hardware
Completely personalized, custom intelligence with complete privacy
5 years from now very person on earth will have this. Their own custom digital worker getting tasks done for them 24/7
And the best part is, it's all American made baby🇺🇸
If you haven't gotten into local models yet. NOW is the time
This is how Elon ended up buying Twitter.
Consider that Elon designed his EV from the ground up, knowing where the tech was going.
All the other auto manufacturers simply bolted EV tech onto their existing designs. It sucked by comparison.
No one is skeptical of Elon's EVs.
His Amazon clone would likely be branded as part of X, his "everything" brand. Fully automated, fully robotic, designed around first principles.
Historically, others wouldn't compete with Amazon because of the capital required to create a competitive offering.
The market is more than willing to give Elon money to do anything he wants to do.
Including building an Amazon competitor if he wants to.
Routinely sweeping the executive conference room for listening devices has been a standard practice for decades, as well as other aspects of operational security.
The difference here is the spy got caught quickly, which rarely happens when one of these turns up.
I think the ability of AI to help detect and prosecute crimes is vastly underrated by the general population.
holy crap GPT 6 was used to weed out a spy stealing secrets at OpenAI’s offices.
guess jacob coxon was right - foreign spies have been planted at major labs to steal blueprints of models, maybe even sabotage future releases:
- a hidden recording device was discovered at openai’s HQ
- GPT 6 was used to deduce who most likely planted it, FBI was called in.
- that employees now been terminated and under investigation.
AI models have effectively become akin to nuclear war codes or the keys to the royal mint but are significantly more vulnerable.
would not be surprised if the government gets involved now to protect future model development.
Big tech shifts always open up entirely new markets.
If I was in my 20s, I'd grab a couple of my fellow geeks, invest in a beefy local AI rig, and target small-to-medium sized businesses, particularly segments that are not well-served by traditional SaaS solutions.
Generating and qualifying leads shouldn't be hard. It's easier to get sizable deals done in this segment as you're usually working with the owner.
We'd put their company on a drip-feed of incremental new capabilities, layering on top of whatever they're using today -- and giving the organization time to adapt to new workflows.
Like the SaaS business before it, the real money is in maintenance and customization work.
Like the heroin business before it, first hit is free :)
Local AI is good enough for a startup. A 10-person startup built its own 128GB-VRAM AI server for $18,000.
One Threadripper box sitting locally with
🎮 4× Radeon AI PRO R9700 32GB
💾 128GB total VRAM
🧠 Qwen3.8 + DeepSeek V4 Flash
👥 16 concurrent AI sessions
And they're actually pushing it.
Qwen3.8-Flash-Next with 16 simultaneous sessions:
📚 48K context/user
⚡ 538.5 tps aggregate decode
At 128K context per user
⚡ 365 tps aggregate decode
The GPUs are even undervolted to under 210W each.
Link in ALT
ALT https://www.reddit.com/r/LocalLLaMA/comments/1x2u742/building_a_4x_r9700_setup_for_a_10_person_startup/