News about Apple and Amazon

[John F Sowa]

Two items from the newsletter by Anand Sanwal show current trends.  

1. Apple is canceling their plans to use LLMs to upgrade Siri.  Instead, they're trying to find another company that can produce the next version of Siri.  Using LLMs is harder than they thought it would be.

2.  Amazon is firing more workers as robots take over, and they need more highly skilled workers.

This raises many questions:  Where do you get the highly trained workers?  How do you help the workers who lose their jobs?   Where do you get the money to train people, retrain people, help the unemployed, match people to jobs?  

John

__________________________________________

Not: Apple’s homegrown LLM strategy

 Apple might ditch its own AI models for Siri — and that's a pretty big deal for a company that loves building everything in-house. 

The iPhone maker is chatting with Anthropic and OpenAI about powering future Siri versions instead of using Apple's own foundation models. After testing both options, Apple execs think Anthropic's tech works better than what they've built internally. 

This comes after Apple had to push back its AI-enhanced Siri from early 2025 to spring 2026. While other big tech players like Amazon, Microsoft, and Meta are ramping up massive AI infrastructure spending — with combined capex hitting unprecedented levels — Apple's struggles highlight how even trillion-dollar giants can't always build the best AI in-house.

This comes after Apple had to push back its AI-enhanced Siri from early 2025 to spring 2026. While other big tech players like Amazon, Microsoft, and Meta are ramping up massive AI infrastructure spending — with combined capex hitting unprecedented levels — Apple's struggles highlight how even trillion-dollar giants can't always build the best AI in-house.

Not: Human factory workers 

 Amazon's warehouse workforce is shrinking as robots take over.  The company now operates over 1M robots globally, with facilities averaging just 670 human employees — the lowest in 16 years. 

CEO Andy Jassy confirmed the trend: fewer people will handle jobs that robots can do. While some workers get retrained for higher-paying technical roles, the math is stark — total employment is declining even as operations expand. 

This signals a broader transformation beyond Amazon. Despite early predictions of fully automated "dark warehouses," nearly 80% of facilities still rely on manual processes. The industry is instead embracing hybrid models where technology augments human capabilities, creating upskilled roles managing robotic systems alongside traditional workers.

[Dan Brickley]

Apple is canceling their plans to use *their own* LLMs to upgrade Siri

The article doesn’t support “Apple is canceling their plans to use LLMs to upgrade Siri”

I have no idea what is actually happening, of course 

Best,

Dan

--
All contributions to this forum are covered by an open-source license.
For information about the wiki, the license, and how to subscribe or
unsubscribe to the forum, see http://ontologforum.org/info
---
You received this message because you are subscribed to the Google Groups "ontolog-forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ontolog-foru...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/ontolog-forum/01f1ef5dedc941cb89d388456a135081%401ac59d87c3a744b5b8cce382b827b3e3.

[John F Sowa]

Dan,

Yes, the emphasis on the phrase "their own" is implied, since LLMs have made a major advance in processing natural languages, and everybody (or nearly everybody) is using them in one way or another for NLP.  But as many projects and publications have shown, the method of generating answers by LLMs is a kind of abduction or educated guessing.  And guessing can never be 100% accurate, especially for complex subject matter.

There are several ways of improving the quality of the guesses:

1. Avoid introducing any extraneous information.   Translating notations, natural or artificial, is fairly safe because the output has a very close relationship to the input , and it's unlikely that extraneous data will be introduced.  This is the first and most successful application of LLMs that Google developed.

But these translations are not perfect, since many specialized applications require technical terminology that is not the most probable in everyday language.  Examples include international treaties, legal contracts, proprietary or trade marked terminology by certain businesses, and new discoveries and inventions in science and engineering.  Special methods and/or human editors and translators are required.

2. Limit the source data to a narrow selection of texts.  An example would include Q/A methods about a collection of documents produced by a single company or individual.   But this is not perfect because the texts produced by a company or an individual may develop new terms and ways of using them.  Other errors may be caused by combining information from two or more documents on unrelated topics.

3. Scan the source data to eliminate unreliable, irrelevant, or erroneous texts.  This is impossible or impractical for sources as large as the Internet.   Some companies have hired cheap workers in countries such as Kenya to eliminated texts about pornography and violence.  But those methods can only eliminate a tiny fraction of the evil or false information on the WWW.

There is much more to say about these issues.  But the above examples show the still unsolved challenges that Apple and others have discovered.   The fact that companies that develop Q/A systems use cheap labor from Kenya to weed out just a tiny fraction of false or bad data shows that LLMs, by themselves, cannot do that task.

John

 

---

From: "Dan Brickley' via ontolog-forum" <ontolo...@googlegroups.com>

[Philip Jackson]

John,

Regarding your note and questions: "Amazon is firing more workers as robots take over, and they need more highly skilled workers. ... Where do you get the highly trained workers? How do you help the workers who lose their jobs? Where do you get the money to train people, retrain people, help the unemployed, match people to jobs?"

The topic of "technological unemployment" is discussed in my book "Toward Human-Level Artificial Intelligence", published by Dover in 2019. Almost all of the following somewhat lengthy discussion is also given in the third edition of my book "Introduction to Artificial Intelligence", also published by Dover in 2019:

In 1930, Keynes defined 'technological unemployment' as unemployment caused by technology eliminating jobs faster than it creates new jobs. He warned it would be a significant problem for future generations.

In 1983, Leontief, Duchin, and Nilsson each wrote papers about the potential for automation and AI to cause long-lasting unemployment. Leontief (1983a,b) reasoned the use of computers to replace human mental functions in producing goods and services would increasingly reduce the need for human labor. Nilsson (1983) predicted AI would significantly reduce the total need for labor, particularly for white-collar and service sector jobs. Duchin (1983) discussed methods for widely distributing incomes without paychecks. In the past two decades, several authors have warned about this potential problem and suggested possible solutions.

In the past two decades, several authors have warned about this potential problem and suggested possible solutions. They include Albus (2011), Brain (2013), Brynjolfsson and McAfee (2011), Ford (2009), Reich (2009 et seq.), Rifkin (1995 et seq.), and others. So, several economists (Brynjolfsson, Duchin, Leontief, McAfee, Reich, Rifkin) and computer technologists (Albus, Brain, Ford, Nilsson) have discussed this problem and developed similar viewpoints. To be concise in referring to these authors, they will here be called Leontief-Duchin-Nilsson (LDN) theorists, focusing only on their arguments regarding technological unemployment, automation, and AI—they may disagree about other topics. It would be incorrect to call them Keynesian economists, since this term refers to Keynes' theories more broadly. Nor is it accurate to call them Luddites or neo-Luddites, because they do not advocate halting technological progress.

Economists in general disagree about whether technological unemployment can have widespread and long-lasting effects on workers and the economy. Many economists have considered it is not a significant problem, arguing that workers displaced by technology will eventually find jobs elsewhere, and the long-term effect on an economy will be positive. However, Leontief, who was awarded the Nobel Memorial Prize in Economic Sciences in 1973, wrote that it is not valid to assume that someone who loses a job due to technological progress will always be able to find another job, even after retraining. Brynjolfsson and McAfee (2011) wrote that no economic law says technological progress will automatically benefit most of the people. A large majority of the people in a nation can have reduced wealth as a result of technological progress, even if overall wealth increases.

Since it is beyond the scope of this book to resolve disputes about theories of economics, the views of LDN theorists can at most be presented tentatively. Those writing in the past two decades appear to roughly agree at least implicitly, on the following points for the problem of technological unemployment:

1. In the next several decades of the 21st century, automation and AI could lead to technological unemployment affecting millions of jobs at all income levels, in diverse occupations, and in both developed and developing nations. This could happen with current and near-term technologies, i.e. without human-level AI. It has already occurred for manufacturing, agriculture, and many service sector jobs.

2. It will not be feasible for the world economy to create new jobs for the millions of displaced workers, offering equivalent incomes producing new products and services.

3. Widespread technological unemployment could negatively impact the worldwide economy, because the market depends on mass consumption, which is funded by income from mass employment. LDN theorists vary in discussing and describing the degree of impact.

4. The problem is solvable by developing ways for governments and the economy to provide alternative incomes to people who are technologically unemployed. LDN theorists have proposed methods for funding and distributing alternative incomes.

5. The problem can and should be solved while preserving freedom of enterprise and a free market economy.

6. The problem cannot be solved by halting or rolling back technological progress, because the world's population depends on technology for economic survival and prosperity.

7. Solutions to the problem could lead to greater prosperity, worldwide. LDN theorists vary in describing potential benefits: Nilsson (1984) envisioned automation and AI could provide the productive capacity to enable a transition from poverty to a "prosperous world society". Ford (2009) suggested extension of alternative incomes to people in poverty could create market demand causing a 'virtuous cycle' of global economic growth.

Based on the arguments of LDN theorists, the possibility that AI could help eliminate global poverty may be considered a 'potential best case event' for the economic risks and benefits of AI.

To summarize: If technological unemployment is a major economic problem, then global prosperity could require developing an economic system that provides alternative incomes to people who are technologically unemployed. The challenge could be to develop ways of funding a universal basic income while preserving freedom of enterprise and economic stability, and controlling monetary inflation.

Wray (1998) discussed how the federal government could be the 'employer of last resort' in the US economy, using 'modern monetary theory' (MMT). Work on repairing and upgrading the nation's infrastructure could offset or postpone technological unemployment for perhaps millions of workers for the next few decades. Much work could be funded in the service sector, providing care for the environment, people, and communities. MMT funding could also provide income for extended periods of time to people who are technologically unemployed. Such funding would minimize taxation and avoid over-taxation for all income levels, and enable capitalism and free markets to continue supporting economic growth. It could reduce the need for state and local taxes. (N.b: MMT and LDN theory originated separately.)

Phil

---

From: ontolo...@googlegroups.com <ontolo...@googlegroups.com> on behalf of John F Sowa <so...@bestweb.net>
Sent: Friday, July 4, 2025 12:05 AM
To: ontolog-forum <ontolo...@googlegroups.com>; CG <c...@lists.iccs-conference.org>
Subject: [ontolog-forum] News about Apple and Amazon

[John F Sowa]

Phil,

I strongly agree with those four points in your note:

4. The problem is solvable by developing ways for governments and the economy to provide alternative incomes to people who are technologically unemployed. LDN theorists have proposed methods for funding and distributing alternative incomes.

5. The problem can and should be solved while preserving freedom of enterprise and a free market economy.

6. The problem cannot be solved by halting or rolling back technological progress, because the world's population depends on technology for economic survival and prosperity.

7. Solutions to the problem could lead to greater prosperity, worldwide...

My primary concern is about electing leaders at the top who understand the issues and are willing and able to implement them.  I have serious doubts about the current leader, but that is not a topic for Ontolog Forum.

John

 

---

From: "Philip Jackson" <philipcj...@hotmail.com>