DebtDeflation

In my experience there's really only three true prompt engineering techniques:

- In Context Learning (providing examples, AKA one shot or few shot vs zero shot)

- Chain of Thought (telling it to think step by step)

- Structured output (telling it to produce output in a specified format like JSON)

Maybe you could add what this article calls Role Prompting to that. And RAG is its own thing where you're basically just having the model summarize the context you provide. But really everything else just boils down to tell it what you want to do in clear plain language.

Maybe that's why the current administration is pushing so hard to bring back low end manufacturing.

If you're a SWE, Accountant, Marketer, HR person, etc. put out of work by AI, now you can screw together iPhones for just over minimum wage. And if we run out of those jobs, there's always picking vegetables now that all the migrants are getting deported.

It would not surprise me one bit if the Tech CEOs see things this way.

It's happening in every industry. CEOs moving back office jobs to India and telling Wall St they replaced the jobs with "AI" to get a stock price boost. I'm convinced this dynamic is a major cause of the "white collar recession" we're experiencing now. Perhaps the intent is to eventually replace the Indians with AI (Artificial Intelligence), but right now it's very much AI (Anonymous Indians) doing the work.

That's why I said AI is not a fad. It's everywhere and is providing real utility. It's just not going to let CEOs replace the majority of their workforce like they think it will.

> Some of the smartest people I know share a bone-deep belief that AI is a fad — the next iteration of NFT mania

It's not that it's a fad. It's that the hype has gotten way ahead of the capability. CEOs laying off double digit percentages of their workforce because they believe that in 6 months AI will actually be able to do all those jobs and they want to get the message out to Wall St to juice the stock price today.

We don't talk about our mutant superpowers outside of the group chat.

> I do buy into the narrative that selection bias for lucky risk-takers explains a lot of bad leaders

It explains a lot more than bad leaders, it explains almost everything in society. The winners write history and you can't really win unless you take big risks.

Wouldn't the inverse be true also? Why would non-believers bet yes? If they lose (which they see as the near certain case) they're out the money, if they win the money doesn't really offset eternal damnation. Seems like betting no is the only smart play regardless of personal beliefs.

The first 42 paragraphs read like something that could have been written by any libertarian free-market thinker at any point during the last 80 years. The final 3 paragraphs read like something written in the last few months to appease the current administration.

I'm a massive quantum computing skeptic. However, I do believe that if there is going to be an actual real and practical QC use case with verifiable quantum advantage in our lifetimes, it's NOT going to be factoring with Shor's or doing search with Grover's, it's going to be molecular interaction simulation as described here. Even so, I give it less than 50% chance of coming to fruition.

It's absurd.

>"The product will be capable of being fully aware of a user’s surroundings and life, will be unobtrusive, able to rest in one’s pocket or on one’s desk, and would be a third core device a person would put on their desk after a MacBook Pro and an iPhone."

Why exactly would I want a "third" device for my desk or a second one for my pocket?

The only additional device I'd consider at this point would be an eyeglass sized VR/AR.

So basically an Alexa/Echo except with always-on microphone and camera communicating with OpenAI's servers?

Or, near as I can tell, any customers or any revenue. And apparently 55 employees.

The question is who is setting these OKRs/Metrics for management and why?

It seems to me to be coming from the CEO echo chamber (the rumored group chats we keep hearing about). The only way to keep the stock price increasing in these low growth high interest rate times is to cut costs every quarter. The single largest cost is employee salaries. So we have to shed a larger and larger percentage of the workforce and the only way to do that is to replace them with AI. It doesn't matter whether the AI is capable enough to actually replace the workers, it has to replace them because the stock price demands it.

We all know this will eventually end in tears.

Whatever happened to feature extraction/selection/engineering and then training a model on your data for a specific purpose? Don't get me wrong, LLMs are incredible at what they do, but prompting one with a job description + a number of CVs and asking it to select the best candidate is not it.

>The sale of this data should be treated as PII and "transfers of ownership" should not be implied.

Completely agree. The usual social media esque "it's a free service so you are the product" arguments don't apply here, people paid 23andme for genetic testing.

The good news is:

>"Regeneron pledged to comply with 23andMe’s privacy policy, which allows customers to have their personal information deleted upon request."

There needs to me a massive campaign to convince every single customer to file a data deletion request.

Hadoop was fundamentally a batch processing system for large data files that was never intended for the sort of online reporting and analytics workloads for which the DW concept addressed. No amount of Pig and Hive and HBase and subsequent tools layered on top of it could ever change that basic fact.

> automated decision systems

So if a bank has an automated loan approval system that consists of a series of IF-THEN statements, and one of those statements amounts to IF (applicant.race != "White"), loan.reject; this ban would forbid a state from taking action?

I've worked in the field of Contact Center AI for almost 10 years now. Rough order of magnitude you can expect to deflect around 30% of inbound contacts with an AI chatbot. You can also expect to reduce agent Average Handle Time by 15-20% using AI enabled semantic search and workflow automation. This is ballpark, if your contacts are simpler you can do a little better, if they're complex you'll do a little worse.

The problem is that vendors are telling companies they can eliminate 80-90% (maybe even 100% if they can keep a straight face) of their customer service agent jobs with AI, and that is nonsense.

>It’s ok to assume that you absorb 100% of what you eat

That's not really true. If you've ever done the keto diet, you know that your body expels unburned ketones through your breath, sweat, and urine. Protein can be used to repair structures rather than burned or stored for energy.

There's also something called the "thermic effect of feeding". Your body requires more energy to process protein (20-30% of calories consumed) than it does carbs (5-10%) than it does fats (0-3%).

That started awhile ago. Google "the great flattening".

It unravels when unemployment spikes and all of these automatic 401k contributions stop and eventually reverse as the unemployed start draining their retirement savings to survive.

This is how it was in the US too growing up in the 1970s and 1980s, everyone had a clothes line in their yard. But by the late 1980s it seemed everyone started getting clothes dryers.

>your clothes would not last that long

Washing on cold or warm, gentle cycle, and then either tumble drying on low or hang drying will greatly extend the life of your clothes. Washing on hot with a more vigorous cycle and then drying on hot not only risks shrinkage in the short term but will cause your clothes to wear out and fall apart much faster.

On a similar note, we had the big push towards "BPA-free" plastic a few years ago. Manufacturers just replaced the BPA with related bisphenols like BPF which is probably just as bad as BPA and BPS which is probably worse than BPA.

The study looks at 11 occupations in Denmark in 2023-24.

Maybe instead look at the US in 2025. EU labor regulations make it much harder to fire employees. And 2023 was mainly a hype year for GenAI. Actual Enterprise adoption (not free vendor pilots) started taking off in the latter half of 2024.

That said, a lot of CEOs seem to have taken the "lay off all the employees first, then figure out how to have AI (or low cost offshore labor) do the work second" approach.

I was on a plane during the Northeast Blackout of 2003. Landed, got in my car, and attempted to leave the parking garage but with no power the automated EZ-Pass payment wasn't working nor were the credit card machines. Most people, myself included, had neither sufficient cash nor a checkbook on hand. Huge logjam of cars. The workers ended up getting the old school handheld metal credit card machines that created an imprint of the credit card on carbon paper from some long forgotten storeroom and using them.

That's an issue once the tow truck gets where it's going, but the GPS tracker will record/broadcast the path there.

The current gold standard for vehicle theft protection is:

IGLA system to block the CAN bus, LIN bus, and ODBII port. It also protects against key fob cloning/relay attacks.

+

A hidden physical kill switch that cuts off the fuel pump relay (the company 41.22 makes a drop in that doesn't require wire splicing).

+

A hidden GPS tracker with an onboard backup battery in the event the car battery is disconnected.

None of this stops someone with a flatbed from simply towing your vehicle away, but at least the GPS tracker will give you a window to locate them.

This after they cut 15,000 employees (over 10% of the workforce at the time) last August? How many times in a row can you cut 10%, 20%, of the workforce without impacting ability to deliver? Either these people were doing nothing of value (in which case leadership is incompetent) or they were doing something of value and the company performance will be impacted by their departure (in which case leadership is incompetent), so which is it?