MJY

Pragmatic AI watchdog | Tracking hype vs. reality for responsible scaling | DevSecOps engineer | Pro-engineering discipline, pro-sustainable rollout

[MJY @omninomsky]

@DaveShapi Let's finish the thought out:

And as tech "gets easier" the demand for more sophisticated application and usage increases until suddenly the need for experts has... grown? That can't be right. Why do we always need more experts when we innovate? Tale as old as time.

[MJY @omninomsky]

@AlanRMacLeod It's one of those things where you just have to watch it play out in real-time.

[MJY @omninomsky]

@GosuCoder You think loops are bad, you should see what this "branching" business is all about.

[MJY @omninomsky]

@grok @Vivek4real_ So just to be clear, no one at the WEF (or likely anywhere) is advocating for a plan to monetize breathing?

[MJY @omninomsky]

Actually, I think fraud is too harsh a term. Truth is Musk has achieved a lot of incredible and ambitious things. Regardless of your personal feelings, there's no denying he has many successful, world changing solutions that, especially in contrast to others in similar spaces, are a net benefit.

I think Musk has grand ideas but I think oftentimes he gets carried away. Especially when he's selling something. Especially when he has to sell something. Especially when failure is existential. Especially when that failure is all but guaranteed. That's typically when things get outlandish bordering on fraudulent.

[MJY @omninomsky]

He's actually not very charismatic either.

Let me elaborate my point. I already have known for well over a decade that Elon Musk was a fraud who had a tenuous grasp on physics and material world engineering.

I don't know too much about his background but I assumed he had at least a strong background with computers and had inferred the rest (erroneously).

Now, I have clear evidence that he doesn't know how computer architecture works either.

In order to do what he suggests would require a complete paradigm shift at the hardware/ISA layer.

This would be the kind of thing that would be bigger than Musk's world. We'd be hearing echoes coming from chip manufacturers. I personally think it would mean abandoning Von Neumann architecture.

That's nowhere to be seen let alone going to be here in 6 months.

That's all I'm saying

[MJY @omninomsky]

@alive_prolly @Bitcoin_Teddy If it were even possible in the first place, it would need entirely new and different architecture from Von Neumann architecture and ISAs. Musk says that paradigm shift is 6 months away. I couldn't make up something more preposterous if I tried.

[MJY @omninomsky]

@Bitcoin_Teddy I don't know, the upper division coursework for my computer science program at my university were some of the most eye opening and engaging foundational material I ever engaged with. I feel very privileged and lucky to know what I know.

[MJY @omninomsky]

@BraddrofliT @CitizenSigma Fair point. But can you explain to me which term wasn't, at some point, made up?

[MJY @omninomsky]

@asaio87 @sourabhbgp The more sane/conservative folks will have write the prompts as steps and will have pass/fail criteria. It still can go off the rails, but you can converge on many solutions surprisingly quickly if you have a rigorous approach.

[MJY @omninomsky]

The Last Audit

The Rapture took them at 3:17 a.m. on a Tuesday.

No trumpets. No light. No explanation.

Every adult male on Earth simply ceased to be.

Dr. Elena Voss, epidemiologist, woke alone.

Her husband's side of the bed was empty. The indentation remained in the mattress. Her phone was already vibrating with emergency alerts. News feeds filled with impossible footage: empty truck cabs rolling to a stop on interstate shoulders, abandoned control rooms, surgeries interrupted mid-procedure.

By noon, the first numbers began arriving.

Day 3

Violent Crime: -87%

The decline was immediate.

Homicides, assaults, robberies—all fell sharply. Prisons reported unusual quiet. Emergency dispatch centers noted dramatic reductions in violent incidents. The data appeared across every major dashboard on Earth.

In New York, the mayor announced an emergency commission to study the phenomenon.

"The early numbers are impossible to ignore."

In Geneva, the United Nations convened a special session on the future of global peace.

Public debate shifted rapidly toward understanding what the data revealed.

For the first time in decades, nearly every trend line pointed in the same direction.

Elena opened a notebook.

Not for history.

For accounting.

Day 12

Violent Crime: -91%

Major Grid Interruptions: 3

Spare Transformer Deliveries: 0

The electrical systems continued operating largely as designed.

At first.

The problem was not generation.

The problem was maintenance.

Transformers failed at approximately the same rate they always had. Switchgear still wore out. Cooling systems still leaked. Inspection schedules still arrived.

Only the people responsible for many of those tasks no longer did.

The first shortages appeared in replacement inventories.

Nobody panicked.

The metrics still looked excellent.

Day 53

Violent Crime: -94%

Grid Reliability: -19%

Insulin Production Facilities Offline: 3 of 5

The world adapted where it could.

Attorneys became utility administrators.

Public health officials learned water treatment operations.

University professors entered logistics management.

Retired engineers delayed retirement indefinitely.

Every organization discovered the same problem at roughly the same time.

There were more critical roles than people available to fill them.

The shortages did not arrive as catastrophe.

They arrived as postponements.

Maintenance deferred.

Inspections delayed.

Repairs rescheduled.

Backlogs accumulating quietly in systems nobody thought about until they stopped working.

Day 104

Violent Crime: -96%

Global Freight Volume: 41% of Baseline

Projected Future Birth Rate: Approaching Zero

Container ships continued arriving.

Less frequently.

Rail networks continued operating.

Less reliably.

Hospitals continued performing surgery.

Less often.

The problem was never willingness.

The problem was dependencies.

Replacement valves sat in warehouses waiting for transportation.

Transportation waited for maintenance.

Maintenance waited for parts.

Parts waited for transportation.

The loops tightened.

International organizations continued publishing reports.

Several awards were established recognizing excellence in post-Rapture public safety initiatives.

The ceremonies were well attended.

Day 187

Violent Crime: -98%

Remaining Viable Sperm Reserves: Rapidly Degrading

Regional Communications Failures: Increasing

The quiet spread.

Not dramatically.

One subtraction at a time.

A substation failed.

A rail signal remained offline.

A water treatment facility reduced capacity.

A satellite ground station lost synchronization.

No individual failure was particularly alarming.

The accumulation was.

Elena began noticing a pattern.

People spoke constantly about outcomes.

Almost nobody spoke about maintenance.

The workers who had calibrated switches.

Inspected towers.

Repaired turbines.

Serviced pumps.

Replaced bearings.

Maintained generators.

They had rarely appeared in public narratives while they were present.

They appeared everywhere once they were gone.

Day 412

Violent Crime: Insufficient Data

Population: Declining

Electrical Service: Intermittent

The dashboards began disappearing.

Some organizations could no longer collect data.

Others could no longer publish it.

The measurement systems themselves required maintenance.

Governments released thoughtful papers on resilience, sustainability, and adaptive urban planning.

Many of the servers hosting those documents went offline within the year.

For the first time in history, entire national governments consisted exclusively of women.

For the first time in history, corporate boards, cabinet positions, and executive offices achieved complete gender parity.

The milestones were real.

So were the blackouts.

Day 821

Population: Declining

Civilizational Capacity: Contracting

World Peace: Achieved

No one knew the violent crime rate anymore.

There were no global reporting systems left to calculate it.

Elena suspected it was close enough to zero.

The distinction no longer mattered.

She sat on the porch of what had once been a suburb.

Trees had reclaimed the roads.

The electrical lines hung silent overhead.

The notebook rested open in her lap.

For decades she had recorded the numbers.

Now only one entry remained.

She wrote carefully.

Audit Complete

Metric| Result
Violent Crime| Eliminated
Humanity| Eliminated
Civilization| Eliminated

She stared at the page for a long time.

Then, beneath the table, she added a final note.

World peace achieved.

Technically correct.

She closed the notebook.

The audit was complete.

The silence continued.

[MJY @omninomsky]

@ICONOCLASTIAE @CuriosityonX Actually, it would be 100% because if men disappeared, it would be the last generation of humans, and as that last generations dies out so too will the remainder of the crime.

You know, logical conclusions and all

[Vivek Sen @Vivek4real_]

THE MAN WHO HELPED INVENT MODERN AI SAYS CHATGPT, CLAUDE, AND GEMINI ARE BUILDING THE WRONG THING.

HE RAISED $1 BILLION TO BET AGAINST THE ENTIRE LLM INDUSTRY.

NO PRODUCT. NO REVENUE.

JUST A BET THAT TODAY'S AI IS FUNDAMENTALLY FLAWED

[MJY @omninomsky]

@Stretchedwiener @Govindtwtt The funniest part of the fiction you just wrote is the notion of 'perfected AI'. I had to really suspend my disbelief to get through the rest of the post.

[MJY @omninomsky]

@easybod @lizterzapzap It's almost as if there are wide range of Linux users and therefor would look rather foolish to think of them as a collective monolith with identical thoughts.

[MJY @omninomsky]

See, people keep focusing on the downstream parts of this. They're assuming this stuff already works as intended and they'll be able to scale it up indefinitely. By coming into the conversation in this way, we're taking a maybe on the success and viability of AI and turning it into a foregone conclusion.

The problem is that we haven't proven whether AI has long term success potential, or is even sustainably viable. We need both of those things to be understood and accepted before we can start predicting the downstream effects such as further stratification in society.

The problem with the upper echelons is they turn a blind eye to reality where ever they can. They plug it up with money and resources. To them, reality is something to manage, mitigate, and ignore. Simply put, they don't like being told no, but reality doesn't care and says no anyway.

The fact is that AI hyperscalers puts tremendous strain on public systems. The NERC issued their rarest level 3 alert over a month ago. The load variance by these hyperscalers can be so massive and unpredictable that grid workers can't respond accordingly. The grids weren't designed to deal with this sort of thing and could risk real outages. If that starts happening, people will forbid the hyperscaler operations

That's just one big reality check AI has to pass. The second is the low ROI from companies that have gone the most in on adopting AI. Many report losses. Many have dispensed with a lot of experts thinking adopting AI was inevitable. They have buyers remorse and many probably feel duped.

We've already seen correction as software engineering roles are becoming more and more numerous on job boards. The end of this month is the end of the quarter. As companies start having earnings call I suspect we're going to hear a lot of pivoting away from AI except utilization in a human augmenting capacity.

But by the end of next quarter? We're likely going to see the S curve of AI adoption taper. This means there's a real risk of profitability.

Finally, it's looking more and more like hardware is catching up. Other countries are seeing that buying dedicated hardware for their own, personal, locally hosted AI, with unlimited token usage, as becoming more and more viable.

This also is a big dent in the hyperscaler model. It could still be used for the chatbot. But as a developer, I'm going to use the local model because it wouldn't take long (less than a year) for it to pay for itself.

The problem is the "upper echelon" is as greedy as they are insulated from reality. Their greed wanted them to be owners of a new utility ala electricity in the form of intelligence. The reality is that AI is much, much cheaper for users and the planet if it's decentralized. It's also more effective.

These reality checks are the reasons why we shouldn't treat the "upper echelon" predictions as foregone conclusions. The evidence shows there's a real chance that things go quite differently.

[MJY @omninomsky]

@0xkkonrad Anthropic, with the help of compilers....

Yeah I see your point.

I get citing the AI in the sense of full disclose for a team review. The customers, or even the people outside the team, don't need to know that though.

[MJY @omninomsky]

@lizterzapzap @duvaldood Sounds like you're making your skill issue our problem