Professor Marcus, You Missed One Point about the Apple Reasoning Paper
June 16, 2025
An opinion essay written by a dinobaby who did not rely on smart software but for the so-so cartoon.
The intern-fueled Apple academic paper titled “The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity” has caused a stir. An interesting analysis of the responses to this tour de force is “Seven Replies to the Viral Apple Reasoning Paper – and Why They Fall Short.” Professor Gary Marcus in his analysis identifies categories of reactions to the Apple document.
In my opinion, these are, and I paraphrase with abandon:
- Human struggle with complex problems; software does too
- Smart software needs lots of computation so deliver a good enough output that doesn’t cost too much
- The paper includes an intern’s work because recycling and cheap labor are useful to busy people
- Bigger models are better because that’s what people do in Texas
- System can solve some types of problems and fail at others
- Limited examples because the examples require real effort
- The paper tells a reader what is already known: Smart software can be problematic because it is probabilistic, not intelligent.
I look at the Apple paper from a different point of view.
The challenge for Apple has been for more than a year to make smart software with its current limitations work reasonably well. Apple’s innovation in smart software has been the somewhat flawed SIRI (sort of long in the tooth) and the formulation of a snappy slogan “Apple Intelligence.”
This individual is holding a “cover your a**” document. Thanks, You.com. Good enough given your constraints, guard rails, and internal scripts.
The job of a commercial enterprise is to create something useful and reasonably clever to pull users to a product. Apple failed. Other companies have rolled out products making use of smart software as it currently is. One of the companies with a reasonably good product is OpenAI’s ChatGPT. Another is Perplexity.
Apple is not in this part of the smart software game. Apple has failed to use “as is” software in a way that adds some zing to the firm’s existing products. Apple has failed, just as it failed with the weird googles, its push into streaming video, and the innovations for the “new” iPhone. Changing case colors and altering an interface to look sort of like Microsoft’s see-through approach are not game changers. Labeling software by the year of release does not make me want to upgrade.
What is missing from the analysis of the really important paper that says, “Hey, this smart software has big problems. The whole house of LLM cards is wobbling in the wind”?
The answer is, “The paper is a marketing play.” The best way to make clear that Apple has not rolled out AI is because the current technology is terrible. Therefore, we need more time to figure out how to do AI well with crappy tools and methods not invented at Apple.
I see the paper as pure marketing. The timing of the paper’s release is marketing. The weird colors of the charts are marketing. The hype about the paper itself is marketing.
Anyone who has used some of the smart software tools knows one thing: The systems make up stuff. Everyone wants the “next big thing.” I think some of the LLM capabilities can be quite useful. In the coming months and years, smart software will enable useful functions beyond giving students a painless way to cheat, consultants a quick way to appear smart in a very short time, and entrepreneurs a way to vibe code their way into a job.
Apple has had one job: Find a way to use the available technology to deliver something novel and useful to its customers. It has failed. The academic paper is a “cover your a**” memo more suitable for a scared 35 year old middle manager in an advertising agency. Keep in mind that I am no professor. I am a dinobaby. In my world, an “F” is an “F.” Apple’s viral paper is an excuse for delivering something useful with Apple Intelligence. The company has delivered an illustration of why there is no Apple smart TV or Apple smart vehicle.
The paper is marketing, and it is just okay marketing.
Stephen E Arnold, June 16, 2025
Googley: A Dip Below Good Enough
June 16, 2025
A dinobaby without AI wrote this. Terrible, isn’t it? I did use smart software for the good enough cartoon. See, this dinobaby is adapting.
I was in Washington, DC, from June 9 to 11, 2025. My tracking of important news about the online advertising outfit was disrupted. I have been trying to catch up with new product mist, AI razzle dazzle, and faint signals of importance. The first little beep I noticed appeared in “Google’s Voluntary Buyouts Lead its Internal Restructuring Efforts.” “Ah, ha,” I thought. After decades of recruiting the smartest people in the world, the Google is dumping full time equivalents. Is this a move to become more efficient? Google has indicated that it is into “efficiency”; therefore, has the Google redefined the term? Had Google figured out that the change to tax regulations about research investments sparked a re-thing? Is Google so much more advanced than other firms, its leadership can jettison staff who choose to bail with a gentle smile and an enthusiastic wave of leadership’s hand?
The home owner evidences a surge in blood pressure. The handyman explains that the new door has been installed in a “good enough” manner. If it works for service labor, it may work for Google-type outfits too. Thanks, Sam AI-Man. Your ChatGPT came through with a good enough cartoon. (Oh, don’t kill too many dolphins, snail darters, and lady bugs today, please.)
Then I read “Google Cloud Outage Brings Down a Lot of the Internet.” Enticed by the rock solid metrics for the concept of “a lot,” I noticed this statement:
Large swaths of the internet went down on Thursday (June 12, 2025), affecting a range of services, from global cloud platform Cloudflare to popular apps like Spotify. It appears that a Google Cloud outage is at the root of these other service disruptions.
What? Google the fail over champion par excellence went down. Will the issue be blamed on a faulty upgrade? Will a single engineer who will probably be given an opportunity to find his or her future elsewhere be identified? Will Google be able to figure out what happened?
What are the little beeps my system continuously receives about the Google?
- Wikipedia gets fewer clicks than OpenAI’s ChatGPT? Where’s the Google AI in this? Answer: Reorganizing, buying out staff, and experiencing outages.
- Google rolls out more Gemini functions for Android devices. Where’s the stability and service availability for these innovations? Answer: I cannot look up the answer. Google is down.
- Where’s the revenue from online advertising as traditional Web search presents some thunderclouds? Answer: Well, that is a good question. Maybe revenues from Waymo, a deal with Databricks, or a bump in Pixel phone sales?
My view is that the little beeps may become self-amplifying. The magic of the online advertising model seems to be fading like the allure of Disneyland. When imagineering becomes imitation, more than marketing fairy dust may be required.
But what’s evident from the tiny beeps is that Google is now operating in “good enough” mode. Will it be enough to replace the Yahoo-GoTo-Overture pay-to-play approach to traffic?
Maybe Waymo is the dark horse when the vehicles are not combustible?
Stephen E Arnold, June 16, 2025
OpenA Ive: The Sam Altman Jive is Alive
June 16, 2025
Look out, Tim Apple, your pie may be burning. According to Fox News, “OpenAI’s $6.5B New Acquisition Signals Apple’s Biggest AI Crisis Yet.” The purchase in question? AI device startup io, founded by former Apple designer Jony Ive. “Cyber Guy” Kurt Knutsson tells us:
“This isn’t your typical business deal. It’s a collaboration between Sam Altman, who leads OpenAI, and the designer responsible for some of Apple’s most iconic products, including the iPhone and Apple Watch. Together, they want to create a new generation of AI-powered devices that could completely change how we use technology.”
Altman is trusting Ive and his team of (largely) Apple expats to propel OpenAI to the forefront of a “screenless” tech future. What, exactly, that will look like remains a mystery—a tantalizing one, if the pair has their way. Knutsson writes:
“While Altman and Ive are keeping most details secret, they have hinted at a family of AI devices that focus on seamless, intuitive interaction rather than screens. They want to create something that understands your context, adapts to your needs and helps you connect and create in new ways, all without requiring you to stare at a display. The device won’t be a phone or a pair of glasses but something entirely new that fits into your life as naturally as a MacBook or iPhone once did.”
Contrast this ambitious vision with Apple’s current position as a relative laggard in the AI field. Unlike Google, which tried to overtake Apple on the same AI racetrack, OpenAI plans to blaze a new path. Not only is the firm working on a new paradigm, it hopes to rapidly overtake the market when it gets there. We learn:
“OpenAI’s ambition is huge. In fact, they want to ship 100 million units faster than any company has ever done with a new product, which shows just how big their vision is.”
Will Apple be able to weather this innovation assault? Will tech devices soon ditch the display? Will OpenAI buy Google Chrome? Will soft AI carry the day? So many questions.
Cynthia Murrell, June 16, 2025
Just Cheat Your Way Through Life: Hey, It Is 2025. Get with It, Loser
June 13, 2025
Just a dinobaby and no AI: How horrible an approach?
I am a dinobaby. I lived in Campinas, Brazil. The power was on and off most days of the week. Mostly off, though. My family in the 1950s was one of the few American units in that town. My father planned for my education. I attended the local school for a few weeks. Then the director sent me home. The school was not set up for non-Portuguese speakers. There were a few missionaries in Campinas, and one of them became my Calvert Course tutor. He went to visit a smaller town, tangled with a snake, and died. That meant that I had to read the World Books my father bought as a replacement for the years of schooling I missed.
Bummer. No ChatGPT. Not much of anything except reading the turgid prose of the World Books and answering questions my mother and father presented for the section I read that day. “What was the capital of Tasmania?” I answered, “Hobart.” I guess that meant I passed. So it went for several years.
What would I have done if I had a laptop, electricity, and an Internet connection? I can tell you straight away that I would have let the smart software do my homework. Skip the reading. Let ChatGPT, You.com, Venice.ai, or some similar system do the work. I had a leather soccer (football) and the locals let me play even though I sucked.
When I read “AI Cheating Is So Out of Hand In America’s Schools That the Blue Books Are Coming Back,” I immediately sat down and wrote this blog post. I don’t need smart software, thank you. I have access to it and other magical computer software. I actually like doing research, analysis, and critical thinking. I am happy when someone tells me I am wrong, uninformed, or off base. I take note, remember the input, and try not to make the same mistake again.
But the reality of today is that smart software is like the World Books my parents made me read, memorize facts, and answer questions based on whatever baloney those volumes contained. AI is here; education has changed; and most students are not going to turn their backs on smart software, speed, and elimination of what is for most people the painful process of learning.
People are not stupid. Most just stop learning anything they don’t absolutely have to master. Now why learn anything? Whip out the smart phone, punch the icon for smart software, and let the system do the thinking.
The write up says:
… as AI tears through America’s elite educational system, lobotomizing tomorrow’s young leaders as it goes, could it be that blue books have been refashioned from a villain of the pre-AI age to a hero for our algorithmically-poisoned times? More and more, it seems like they’re the dark knight that America’s illiterate masses needs. The Journal notes that Roaring Spring Paper Products, the family-owned paper company that produces a majority of the blue books that are sold on college campuses, admits that the new AI era has ironically been good for its business.
Nifty. Lobotomize: I wonder if the author of the article knows exactly how unpredictable the procedure was and probably still is in some remote part of the modern world. Will using LLMs make people stupider? No, what makes people stupider is the inability, the motivation, and the curiosity required to learn. Doom scrolling is popular because young people are learning to follow trends, absorb video techniques, and learn how to “do” their fingernails. These may be more important than my knowing that the longest snake known when the World Books were published was over 20 feet long, specifically, the reticulated python. (Thank goodness, the snake lived in Indonesia, not Brazil.)
The write up says:
Indeed, if the return of pen and paper is a promising sign, America’s educators aren’t out of the woods yet—not even close. A recent survey found that 89% of college students had admitted to using ChatGPT to complete a homework assignment. AI-detection tools designed to spot cheating also routinely fail. Increasingly, America’s youth seem to view their educations as a high-stakes video game to be algorithmically juked. In short, more drastic measures (like the formulation of new laws and regulations around AI use) may need to be taken if the onset of America’s aggressive stupidification is to be halted.
My personal view is that a cultural shift has taken place. People don’t want to “work.” Families are no longer nuclear; they are not one mother, one father, and 2.4 children and maybe a dog, probably a boxer or a Labrador. Students no longer grab a book; they only have two hands and both are required to operate a mobile phone or a laptop. Teachers are no longer authority figures; they are viewed as problems, particularly by upper middle class and wealthy parents or parent as the case may be.
The blue book thing is mildly interesting, but I am not sure these are a solution. Students cannot read or write cursive; they print. This means that answers will be shorter, maybe like social media posts. If a student has a knack for art, icons may be included next to an insightful brief statement. A happy face signals the completion of the test. I would, if I were 13, draw a star and a calligraphic “A” on the front of my blue book.
What type of world will this educational milieu deliver? To be honest, I am glad I am old and will die before I have to experience to much of the LLM world. 
Stephen E Arnold, June 13, 2025
Another Vote for the Everything App
June 13, 2025
Just a dinobaby and no AI: How horrible an approach?
An online information service named 9 to 5 Mac published an essay / interview summary titled “Nothing CEO says Apple No Longer Creative; Smartphone Future Is a Single App.” The write up focuses on the “inventor / coordinator” of the OnePlus mobile devices and the Nothing Phone. The key point of the write up is the idea that at some point in the future, one will have a mobile device and a single app, the everything app.
The article quotes a statement Carl Pei (the head of the Nothing Phone) made to another publication; to wit:
I believe that in the future, the entire phone will only have one app—and that will be the OS. The OS will know its user well and will be optimized for that person […] The next step after data-driven personalization, in my opinion, is automation. That is, the system knows you, knows who you are, and knows what you want. For example, the system knows your situation, time, place, and schedule, and it suggests what you should do. Right now, you have to go through a step-by-step process of figuring out for yourself what you want to do, then unlocking your smartphone and going through it step by step. In the future, your phone will suggest what you want to do and then do it automatically for you. So it will be agentic and automated and proactive.
This type of device will arrive in seven to 10 years.
For me, the notion of an everything app or a super app began in 2010, but I am not sure who first mentioned the phrase to me. I know that WeChat, the Chinese everything app, became available in 2011. The Chinese government was aware at some point that an “everything” app would make surveillance, social scoring, and filtering much easier. The “let many approved flowers bloom” approach of the Apple and Google online app stores was inefficient. One app was more direct, and I think the A to B approach to tracking and blocking online activity makes sense to many in the Middle Kingdom. The trade off of convenience for a Really Big Brother was okay with citizens of China. Go along and get along may have informed the uptake of WeChat.
Now the everything app seems like a sure bet. The unknown is which outstanding technology firm will prevail. The candidates are WeChat, Telegram, X.com, Sam Altman’s new venture, or a surprise player. Will other apps (the not everything apps from restaurant menus to car washes) survive? Sure. But if Sam AI-Man is successful with his Ive smart device and his stated goal of buying the Chrome browser from the Google catch on, the winner may be a CEO who was fired by his board, came back, and cleaned out those who did not jump on the AI-Man’s bandwagon.
That’s an interesting thought. It is Friday the 13th, Google. You too Microsoft. And Apple. How could I have forgotten Tim Cook and his team of AI adepts?
Stephen E Arnold, June 13, 2025
Musk, Grok, and Banning: Another Burning Tesla?
June 12, 2025
Just a dinobaby and no AI: How horrible an approach?
“Elon Musk’s Grok Chatbot Banned by a Quarter of European Firms” reports:
A quarter of European organizations have banned Elon Musk’s generative AI chatbot Grok, according to new research from cybersecurity firm Netskope.
I find this interesting because my own experiences with Grok have been underwhelming. My first query to Grok was, “Can you present only Twitter content?” The answer was a bunch of jabber which meant, “Nope.” Subsequent queries were less than stellar, and I moved it out of my rotation for potentially useful AI tools. Did the sample crafted by Netskope have a similar experience?
The write up says:
Grok has been under the spotlight recently for a string of blunders. They include spreading false claims about a “white genocide” in South Africa and raising doubts about Holocaust facts. Such mishaps have raised concerns about Grok’s security and privacy controls. The report said the chatbot is frequently blocked in favor of “more secure or better-aligned alternatives.”
I did not feel comfortable with Grok because of content exclusion or what I like to call willful or unintentional coverage voids. The easiest way to remove or weaponize content in the commercial database world is to exclude it. When a person searches a for fee database, the editorial policy for that service should make clear what’s in and what’s out. Filtering out is the easiest way to marginalize a concept, push down a particular entity, or shape an information stream.
The cited write up suggests that Grok is including certain content to give it credence, traction, and visibility. Assuming that an electronic information source is comprehensive is a very risky approach to assembling data.
The write up adds another consideration to smart software, which — like it or not — is becoming the new way to become informed or knowledgeable. The information may be shallow, but the notion of relying on weaponized information or systems that spy on the user presents new challenges.
The write up reports:
Stable Diffusion, UK-based Stability AI’s image generator, is the most blocked AI app in Europe, barred by 41% of organizations. The app was often flagged because of concerns around privacy or licensing issues, the report found.
How concerned should users of Grok or any other smart software be? Worries about Grok may be an extension of fear of a burning Tesla or the face of the Grok enterprise. In reality, smart software fosters the illusion of completeness, objectivity, and freshness of the information presented. Users are eager to use a tool that seems to make life easier and them appear more informed.
The risks of reliance on Grok or any other smart software include:
- The output is incomplete
- The output is weaponized or shaped by intentional or factors beyond the developers’ control
- The output is simply wrong, made up, or hallucinated
- Users who act as though shallow knowledge is sufficient for a decision.
The alleged fact that 25 percent of the Netskope sample have taken steps to marginalize Grok is interesting. That may be a positive step based on my tests of the system. However, I am concerned that the others in the sample are embracing a technology which appears to be delivering the equivalent of a sugar rush after a gym workout.
Smart software is being applied in novel ways in many situations. However, what are the demonstrable benefits other than the rather enthusiastic embrace of systems and methods known to output errors? The rejection of Grok is one interesting factoid if true. But against the blind acceptance of smart software, Grok’s down check may be little more than a person stepping away from a burning Tesla. The broader picture is that the buildings near the immolating vehicle are likely to catch on fire.
Stephen E Arnold, June 12, 2025
Developers: Try to Kill ‘Em Off and They Come Back Like Giant Hogweeds
June 12, 2025
Just a dinobaby and no AI: How horrible an approach?
Developers, which probably extends to “coders” and “programmers”, have been an employee category of note for more than a half century. Even the esteemed Institute of Advanced Study enforced some boundaries between the “real” thinking mathematicians and the engineers who fooled around in the basement with a Stone Age computer.
Giant hogweeds can have negative impacts on humanoids who interact with them. Some say the same consequences ensue when accountants, lawyers, and MBAs engage in contact with programmers: Skin irritation and possibly blindness.
“The Recurring Cycle of ‘Developer Replacement’ Hype” addresses this boundary. The focus is on smart software which allegedly can do heavy-lifting programming. One of my team (Howard, the recipient of the old and forgotten Information Industry Association award for outstanding programming) is skeptical that AI can do what he does. I think that our work on the original MARS system which chugged along on the AT&T IBM MVS installation in Piscataway in the 1980s may have been a stretch for today’s coding wonders like Claude and ChatGPT. But who knows? Maybe these smart systems would have happily integrated Information Dimensions database with the MVS and allowed the newly formed Baby Bells to share certain data and “charge” one another for those bits? Trivial work now I suppose in the wonderful world of PL/1, Assembler, and the Basis “GO” instruction in one of today’s LLMs tuned to “do” code.
The write up points out that the tension between bean counters, MBAs and developers follows a cycle. Over time, different memes have surfaced suggesting that there was a better, faster, and cheaper way to “do” code than with programmers. Here are the “movements” or “memes” the author of the cited essay presents:
- No code or low code. The idea is that working in PL/1 or any other “language” can be streamlined with middleware between the human and the executables, the libraries, and the control instructions.
- The cloud revolution. The idea is that one just taps into really reliable and super secure services or micro services. One needs to hook these together and a robust application emerges.
- Offshore coding. The concept is simple: Code where it is cheap. The code just has to be good enough. The operative word is cheap. Note that I did not highlight secure, stable, extensible, and similar semi desirable attributes.
- AI coding assistants. Let smart software do the work. Microsoft allegedly produces oodles of code with its smart software. Google is similarly thrilled with the idea that quirky wizards can be allowed to find their future elsewhere.
The essay’s main point is that despite the memes, developers keep cropping up like those pesky giant hogweeds.
The essay states:
Here’s what the "AI will replace developers" crowd fundamentally misunderstands: code is not an asset—it’s a liability. Every line must be maintained, debugged, secured, and eventually replaced. The real asset is the business capability that code enables. If AI makes writing code faster and cheaper, it’s really making it easier to create liability. When you can generate liability at unprecedented speed, the ability to manage and minimize that liability strategically becomes exponentially more valuable. This is particularly true because AI excels at local optimization but fails at global design. It can optimize individual functions but can’t determine whether a service should exist in the first place, or how it should interact with the broader system. When implementation speed increases dramatically, architectural mistakes get baked in before you realize they’re mistakes. For agency work building disposable marketing sites, this doesn’t matter. For systems that need to evolve over years, it’s catastrophic. The pattern of technological transformation remains consistent—sysadmins became DevOps engineers, backend developers became cloud architects—but AI accelerates everything. The skill that survives and thrives isn’t writing code. It’s architecting systems. And that’s the one thing AI can’t do.
I agree, but there are some things programmers can do that smart software cannot. Get medical insurance.
Stephen E Arnold, June 12, 2025
Why Emulating Oxford University in the US Is an Errand for Fools
June 11, 2025
Just a dinobaby and no AI: How horrible an approach?
I read an essay with the personal touches I admire in writing: A student sleeping on the floor, an earnest young man eating KY fry on a budget airline, and an individual familiar with Laurel and Hardy comedies. This person write an essay, probably by hand on a yellow tablet with an ink pen titled “5 Ways to Stop AI Cheating.”
What are these five ways? The ones I noted are have rules and punish violators. Humiliation in front of peers is a fave. Presumably these students did not have weapons or belong to a street gang active in the school. The other five ways identified in the essay are:
- Handwrite everything. No typewriters, no laser printers, and no computers. (I worked with a fellow on a project for Persimmon IT which did some work on the DEC Alpha, and he used computers. (Handwriting was a no go for interacting with the DECs equipped with the “hot” chip way back when.)
- Professors interact with a student and talk or interrogate the young scholar to be
- Examinations were oral or written. One passed or failed. None of this namby pamby “gentleman’s C” KY fry stuff
- Inflexibility about knowing or not knowing. Know and one passes. Not knowing one becomes a member of Parliament or a titan of industry
- No technology. (I would not want to suggest that items one and five are redundant and that would be harshly judged by some of my less intellectually gifted teachers at assorted so-so US institutions of inferior learning.
Now let’s think about the fool’s errand. The US is definitely a stratified society, just like the UK. If one is a “have,” life is going to be much easier than if one is a “have not.” Why? Money, family connections, exposure to learning opportunities, possibly tutors, etc. In the US, technology is ubiquitous. I do not want to repeat myself, so a couple of additional thoughts will appear in item five below.
Next, grilling a student one on one is something that is an invitation to trouble. A student with hurt feelings need only say, “He/she is not treating me fairly.” Bingo. Stuff happens. I am not sure about a one on one in a private space would be perceived by a neutral third party. If one has to meet, meet in a public place.
Third, writing in blue books poses two problems. The first is that the professor has to read what the student has set forth in handwriting. Second, many students can neither write legible cursive or print out letters in an easily recognizable form. These are hurdles in the US. Elsewhere, I am not sure.
Fourth, inflexibility is a characteristic of some factions in the US. However, helicopter parents and assorted types of “outrage” can make inflexibility for a public servant a risky business. If Debbie is a dolt, one must find a way to be flexible if her parents are in the upper tier of American economic strata. Inflexibility means litigation or some negative networking or a TikTok video.
Finally, the problem with the no-tech approach is that it just won’t work. Consider smart software. Teachers use it and have LLMs fix up “original research.” Students use it to avoid reading and writing. Some schools ban mobile devices. Care to try that at an American university when shooters can prowl the campus?
The essay, like the fantasies of people who want to live like those in Florence in the 15th century are nuts. Pestilence, poverty, filth, violence, and big time corruption— there were everyday companions.
Cheating is here to stay. Politician is a code word for crook. Faculty (at least at Harvard) is the equivalent of bad research. Students are the stuff of YouTube shorts. Writing in blue books? A trend which may not have the staying power of Oxford’s stasis. I do like the bookstore, however.
Stephen E Arnold, June 11, 2025
Lights, Ready the Smart Software, Now Hit Enter
June 11, 2025
Just a dinobaby and no AI: How horrible an approach?
I like snappy quotes. Here’s a good one from “You Are Not Prepared for This Terrifying New Wave of AI-Generated Videos.” The write up says:
I don’t mean to be alarmist, but I do think it’s time to start assuming everything you see online is fake.
I like the categorical affirmative. I like the “alarmist.” I particularly like “fake.”
The article explains:
Something happened this week that only made me more pessimistic about the future of truth on the internet. During this week’s Google I/O event, Google unveiled Veo 3, its latest AI video model. Like other competitive models out there, Veo 3 can generate highly realistic sequences, which Google showed off throughout the presentation. Sure, not great, but also, nothing really new there. But Veo 3 isn’t just capable of generating video that might trick your eye into thinking its real: Veo 3 can also generate audio to go alongside the video. That includes sound effects, but also dialogue—lip-synced dialogue.
If the Google-type synths are good enough and cheap, I wonder how many budding film directors will note the capabilities and think about their magnum opus on smart software dollies. Cough up a credit card and for $250 per month imagine what videos Google may allow you to make. My hunch is that Mother Google will block certain topics, themes, and “treatments.” (How easy would it be for a Google-type service to weaponize videos about the news, social movements, and recalcitrant advertisers?)
The write worries gently as well, stating:
We’re in scary territory now. Today, it’s demos of musicians and streamers. Tomorrow, it’s a politician saying something they didn’t; a suspect committing the crime they’re accused of; a “reporter” feeding you lies through the “news.” I hope this is as good as the technology gets. I hope AI companies run out of training data to improve their models, and that governments take some action to regulate this technology. But seeing as the Republicans in the United States passed a bill that included a ban on state-enforced AI regulations for ten years, I’m pretty pessimistic on that latter point. In all likelihood, this tech is going to get better, with zero guardrails to ensure it advances safely. I’m left wondering how many of those politicians who voted yes on that bill watched an AI-generated video on their phone this week and thought nothing of it.
My view is that several questions may warrant some noodling by a humanoid or possibly an “ethical” smart software system; for example:
- Can AI detectors spot and flag AI-generated video? Ignoring or missing may have interesting social knock on effects.
- Will a Google-type outfit ignore videos that praise an advertiser whose products are problematic? (Health and medical videos? Who defines “problematic”?)
- Will firms with video generating technology self regulate or just do what yields revenue? (Producers of adult content may have some clever ideas, and many of these professionals are willing to innovate.)
Net net: When will synth videos win an Oscar?
Stephen E Arnold, June 11, 2025
LLMs, Dread, and Good Enough Software (Fast and Cheap)
June 11, 2025
Just a dinobaby and no AI: How horrible an approach?
More philosopher programmers have grabbed a keyboard and loosed their inner Plato. A good example is the essay “AI: Accelerated Incompetence” by Doug Slater. I have a hypothesis about this embrace of epistemological excitement, but that will appear at the end of this dinobaby post.
The write up posits:
In software engineering, over-reliance on LLMs accelerates incompetence. LLMs can’t replace human critical thinking.
The driver of the essay is that some believe that programmers should use outputs from large language models to generate software. Doug does not focus on Google and Microsoft. Both companies are convinced that smart software can write good enough code. (Good enough is the new standard of excellence at many firms, including the high-flying, thin-air breathing Googlers and Softies.)
The write up identifies three beliefs, memes, or MBAisms about this use of LLMs. These are:
- LLMs are my friend. Actually LLMs are part of a push to get more from humanoids involved in things technical. For a believer, time is gained using LLMs. To a person with actual knowledge, LLMs create work in order to catch errors.
- Humans are unnecessary. This is the goal of the bean counter. The goal of the human is to deliver something that works (mostly). The CFO is supposed to reduce costs and deliver (real or spreadsheet fantasy) profits. Humans, at least for now, are needed when creating software. Programmers know how to do something and usually demonstrate “nuance”; that is, intuitive actions and thoughts.
- LLMs can do what humans do, especially programmers and probably other technical professionals. As evidence of doing what humans do, the anecdote about the robot dog attacking its owner illustrates that smart software has some glitches. Hallucinations? Yep, those too.
The wrap up to the essay states:
If you had hoped that AI would launch your engineering career to the next level, be warned that it could do the opposite. LLMs can accelerate incompetence. If you’re a skilled, experienced engineer and you fear that AI will make you unemployable, adopt a more nuanced view. LLMs can’t replace human engineering. The business allure of AI is reduced costs through commoditized engineering, but just like offshore engineering talent brings forth mixed fruit, LLMs fall short and open risks. The AI hype cycle will eventually peak10. Companies which overuse AI now will inherit a long tail of costs, and they’ll either pivot or go extinct.
As a philosophical essay crafted by a programmer, I think the write up is very good. If I were teaching again, I would award the essay an A minus. I would suggest some concrete examples like “Google suggests gluing cheese on pizza”, for instance.
Now what’s the motivation for the write up. My hypothesis is that some professional developers have a Spidey sense that the diffident financial professional will license smart software and fire humanoids who write code. Is this a prudent decision? For the bean counter, it is self preservation. He or she does not want to be sent to find a future elsewhere. For the programmer, the drum beat of efficiency and the fife of cost reduction are now loud enough to leak through noise reduction head phones. Plato did not have an LLM, and he hallucinated with the chairs and rear view mirror metaphors.
Stephen E Arnold, June 11, 2025
-
- Subscribe to Beyond Search
Feature archive
News archive
-
