Before reading another word, press play on The Pusher.
Yes. Steppenwolf. That one.
Volume at “this isn’t entertainment anymore.”
If the lyrics feel uncomfortably accurate for 2026 so far, you’re exactly where this was meant to land.
The OpenAI “Hail Mary”
“May I pass along my congratulations for your great interdimensional breakthrough. I am sure, in the miserable annals of the Earth, you will be duly enshrined.”
— Lord John Whorfin, The Adventures of Buckaroo Banzai Across the 8th Dimension
It is heartening to see that $20 billion in revenue and nearly 2 gigawatts of power, enough to run a small country or a very aggressive hairdryer, finally resulted in the same business model as a 2010 flashlight app. I can’t wait for my philosophical inquiry into the nature of consciousness to be interrupted by a 15 second unskippable ad for Tactical Grilling Aprons.
GPT-5.2:
“The meaning of life is… but first, have you joined the 500 million players in RAID: Shadow Legends? Use code OPENAI to get a free Legendary Champion, 50,000 Silver, and access to the Secret Cow Level if you act now, while we process your existential crisis. Found in the pursuit of purpose.”
This isn’t a failure of technology.
It’s capitalism completing the tech tree.
The Donation That Achieved Escape Velocity
And then there’s the donation.
Not an investment. Not a contract. Not a convertible note. A donation. A thing you give away because you no longer wish to own it.
Which makes the attempt to sue for roughly 3,600 times its original value feel less like litigation and more like discovering your old couch has secretly been accruing venture capital returns in the garage.
This is not a 3600x comeback story.
This is a gift that allegedly appreciated faster than Bitcoin, Nvidia, and several emerging economies combined.
The theory appears to be:
“I donated this freely, without expectation… except the expectation that, years later, it would mature into $138 billion.”
At this point, philanthropy isn’t generosity. It’s time delayed arbitrage.
Donate now. Wait patiently. Sue later. Let compound irony do the rest.
In that case, I’d like my money back from the Red Cross.
Not because of fraud.
Not because of misrepresentation.
But because we now need it.
For our ketamine addiction.
I mean, sorry, for depression.
I donated under the wildly outdated belief that a donation was a gift.
Apparently, the modern interpretation is that it’s a revocable emotional asset, redeemable later when things don’t go great and the stock market hurts your feelings.
“Dear Charity,
I gave you $50 in good faith.
Due to unforeseen psychological market volatility,
I now require it back with appreciation
for medical reasons.
Please advise.”
This is not philanthropy.
This is self care with a clawback clause.
The Beijing RoboCop
Nothing says “the future is here” like a humanoid robot barking at cyclists in Anhui while the government bans the very chips that could make it smart enough to recognize a bicycle. We are officially living in a world where the robot has a badge, a reflective vest, and the processing power of a singing birthday card, paired with the unshakable confidence of something that has never once been wrong because it has never once been allowed to doubt.
God help you if you’re holding a screwdriver.
Not menacingly. Just tightening your license plate so it stops rattling down the highway. But the robot has already logged: metal object detected. Human intent unclear. Escalation protocol engaged.
Context is unavailable. Nuance is deprecated.
Please remain still while the system consults a laminated decision tree that ends, as all good bureaucratic systems do, in Maximum Response Just to Be Safe.
It’s Blade Runner if the replicants were innocent, the cop was a kiosk, and every tragedy began with the words:
Error: Human behavior outside expected parameters.
The Raspberry Pi “Escape Pod”
The solution to global AI surveillance is apparently a $130 circuit board that looks like it was salvaged from a 1990s VCR. If you want to keep your data private, you just have to accept that your AI assistant now has the memory of a goldfish and the speed of a tectonic plate.
But that’s the trade.
You can have omniscience with ads, subpoenas, and surprise policy updates.
Or you can have a small, loyal idiot humming quietly on your desk that has never once tried to sell you anything.
Final Thought
This week didn’t reveal a fork in the road.
It revealed the menu.
Option A: Free intelligence, subsidized by your attention.
Option B: State intelligence, fully uniformed and deeply confused by context.
Option C: A shoebox computer that forgets your name every 20 minutes but has never asked for a credit card.
Choose wisely.
And finally, as leader of the free, I mean ad supported free world, since you didn’t pick me for your team in dodgeball and I didn’t get a gold star on my homework, I have no choice but to declare that all your base are belong to us.
After reading your reply, I realize now that I require a legal disclaimer and a moment of silence. (Or, as you advised last week, lawyers, guns, and money.)
I love your concluding thesis about the menu. We have a list of tradeoffs. All of them bad.
Also, for the record, if philosophical inquiry ever gets interrupted by a Tactical Grilling Apron ad, then I will gladly use the Raspberry Pi!
;)
Thanks for sharing your insights. They were wonderfully disturbing.
(I'll see you in the 8th dimension, assuming the Secret Cow Level doesn't get us first.)
This connects back to our exchange the other week, Mike, about “physical work as a safe harbor.” 😊
We landed in roughly the same place: it’s not safe — it’s buffered. The China traffic police robots make that buffer visible, as early execution under very specific deployment conditions.
What’s shifted for me since then is how similar this looks to what’s happening inside organizations. Displacement isn’t just about capability — it’s about reliability and cost per productive hour.
In that sense, both the “safe harbor” story and the “imminent takeover” story were incomplete. The real variable is timing.
I love how you tied this back to our earlier exchange about the Boston Dynamics robots + DeepMind integration.
I also think your "buffered, not safe" framework is back into play. The Chinese traffic police robots are the perfect real-world use case. They operate in an area that's relatively constrained, with repeatable conditions, and relatively low need for babysitting.
I’m very curious to see how these systems perform in sustained, real-world deployment. But who knows if we'll get to see any deployment data. Regardless, it feels like 2026 is shaping up to be a real-world inflection point for embodied AI. (The year of the robots?)
Thanks again for such thoughtful comments and for pulling these threads across time and domains. This kind of longitudinal thinking is precisely what I'm trying to cultivate here.
This is such a clear way of laying out what’s actually at stake right now, which I would otherwise have no clue about. Not just new tools, but who controls them, who profits, and what we quietly give up in exchange for convenience. I always appreciate how you connect everyday user choices to much larger structural shifts without turning it into panic or hype. It helps the moment feel understandable instead of overwhelming.
Comments like THIS are ones I like to save for when I'm having a lousy day.
THANK YOU.
I love how you mentioned structural shifts. I spend so much time trying to articulate what all these stories mean in the real-world, in ways that impact all of us.
Your comments always make me feel like someone's actually getting what I'm trying to do, which honestly keeps me writing when I'm wondering if anyone cares. (And, it makes all of the work worth the while.)
Thank you for being such a thoughtful reader and friend. This made my whole week. 💙
Yeah you absolutely sparked my interest. It was a great read! You brought a lot of things to my attention!
I hope this isn't too random but I have a question! Whats your take on decentralize ai vs. centralized ai?
I've been looking into this company called Bittsenor – its a decentralized AI platform (I can explain more, but I'm already feeling like I'm accidentally re-routing the convo too much). But I ask because I think centralized has always made me nervous as it grows - i think of surveillance, control, police robots, etc.
I'd love to know your take– from a curious individual who is only looking to learn :)
I want you to know your questions are always welcome here. I’m thankful there are like-minded nerds who want to dig into these nerdy computer science topics, lol.
The biggest difference between centralized vs. decentralized AI is who holds the power.
With centralized AI, a small number of companies train the biggest models. They also run the infrastructure and set the rules. That usually means the experience is smoother and the models are stronger. But it also means one gatekeeper can decide what’s allowed, how it’s priced, what gets logged, and how the system evolves. The problem with centralized AI is that surveillance, control, and mission creep can become very easy to scale. 👀 (Your concern is totally warranted.)
With decentralized AI, the goal is to spread that control out across a network so no single entity can unilaterally steer everything. That can reduce the risk that one company can flip a switch. But it doesn’t magically make things safer or more ethical. Playing devil’s advocate, decentralized systems can also make it harder to enforce standards, prevent abuse, or even know who’s accountable when something goes wrong.
So centralized AI optimizes for capability and convenience. Decentralized AI optimizes for reducing single-point control, but you trade off simplicity and clear accountability.
As for Bittensor specifically, I haven’t dug deep enough into their architecture to have a strong take yet. That said, the concept of decentralized AI networks is definitely worth watching. If you explore it, I’d love to hear what you find!
Thanks again, Natalie, for your fun questions. I hope you have a beautiful day. And if you have more questions, feel free to ask.
Always happy to be among fellow nerds who enjoy digging into this stuff!! 🙂
Your take around where power sits def helped me think about this, especially the tradeoff between capability/convenience and accountability. That devil’s-advocate point you raised is exactly why i'm curious about this!
I’ve been loosely following projects like Bittensor, which seem to be experimenting with “tokenizing intelligence” in a way that’s loosely analogous to how Bitcoin incentivizes security. But its a bit over my head
I also follow Michael Casey, who advocates for decentralized AI but openly acknowledges many of the same risks you mentioned. It makes me wonder whether the future isn’t purely centralized or decentralized, but some kind of coexistence where each keeps each other in check
Anyway, thank you for engaging my question so thoughtfully, this definitely gave me more to think about!
Thank you so much for this thoughtful note. You made my day!
And it’s always a joy bumping into a fellow nerd in the wild. We should really come with warning labels. Lol.
Also, totally with you on the coexistence possibility. My gut says the future is a hybrid where centralized systems win on convenience and capability, and decentralized systems help keep power from concentrating too tightly.
I just searched for Michael Casey and found a bunch of his interviews on YouTube, so I’m going to check some of those out. ;)
Thanks, Mike! Another good tech-savvy-but-not-jargon-heavy informative article. As someone who worked in local law enforcement and was Army military police a very long time ago, I can appreciate a robot taking over traffic duty! Could be tedious, thankless, and occasionally dangerous. But then that mission creep after that, right? What comes next may not be something I would appreciate.
Totally agree. Robots doing traffic duty is genuinely cool. If it reduces risk for humans, that’s a win.
I also share your concern about mission creep. Do we really want anyone building legions of these things without clear limits and oversight? 👀
Also, I didn’t know you worked local law enforcement and Army MP. Respect. I spent years doing technical support work at Hanscom AFB. (I was never enlisted). Picture a heavily bearded, four-eyed nerd trying to look squared away around disciplined pilots and officers, lol. Good times.
Too funny! But, hey, thanks for lending your knowledge, skills, and abilities to the Air Force! I was an MP back in the 80s when the Cold War still raged. I was stationed in West Germany (yes, it was still West back then) and Fort Hood Texas (not sure if it is called Fort Hood anymore.
It was genuinely my pleasure to support the USAF. Stressful at times. But a lot of fun too. (And yes, they fed me well, lol).
And truly, thank you for YOUR service. Being an MP in the 80s, with the Cold War still raging, especially while stationed in West Germany, must have been intense. (I'm sure you have stories.)
Also, I think Fort Hood may go by a different name now. But either way, what a pair of assignments!!!
Thanks for sharing, Daniel. Hope you have a beautiful week.
Thanks, Mike. Oh, I got stories alright, both good, bad, and ugly. lol. I loved Germany, though. Would love to visit there someday again. I hated being at Fort Hood after Germany! :-)
Lol. I love your AGI joke. It's devastatingly accurate. "Ads Generating Income" is the breakthrough OpenAI has been building toward all this time. 😂
Are you going to try running some local LLMs on a Raspberry Pi? I'm a huge fan of the ecosystem. (Full disclosure: I also use mine to play retro video games when I should be working on AI research. It's a surprisingly capable little retro-console, lol.)
That said, you definitely don't need a Raspberry Pi to use local LLMs. You can run private, open-source LLMs on almost any decent laptop these days. But there's just something cool about the Raspberry Pi.
Thanks again for writing and contributing to the work! It means so much to me.
I am hoping, because I already pay, we (the paying subscribers) will not be subject to ads any time soon.
As I read your very thorough research, the question I haven't yet looked into: What are the repercussions of me losing all my ChatGPT projects? There are a few very keyboard growing out my business.
How do I back up everything easily? Copy and paste to a couple of hard drives?
Thank you so much for writing and for your kind words.
I have great news for you. Based on what OpenAI has shared so far, ads are expected to apply only to "free" and "ChatGPT Go" accounts. Paid plans such as Plus, Pro, Business, Enterprise, and Edu are not supposed to include ads. So, as a paying subscriber, you should be safe for now. Thankfully.
On backing things up, I actually looked into this recently. You can export your ChatGPT data, including chats, from your account settings.
1. Click your profile icon and go to Settings
2. Select Data Controls and then Export Data
3. Confirm, and download the ZIP file from the email they send
However, I should mention that after researching this, many people complain that it doesn't contain ALL of the data. And some old chats, projects, or conversations might get excluded. So it seems rather quirky, unfortunately.
I also confess that I am a bit messy with my chatbots myself. I use ChatGPT, Claude, Gemini, and Perplexity, so things can end up scattered. What has helped me most is keeping local copies of my essential prompts and outputs, often as simple PDFs. It feels a little archaic. But it gives me peace of mind.
For your most critical projects, manually copying key prompts, instructions, and outputs into a document and backing that up to a couple of drives or cloud storage is still the most reliable method I know right now. It is tedious, but dependable.
Thanks again for reading and for raising such an important question!
Thank you Mike for your reply. I appreciate you and the time you invest in helping me / us! I will take your advise with respect to backing up, before I get too far down the rabbit hole.
Wishing you a blessed and lovely rest of your week.
Thank you for this Post, @MrComputerScience. Your AI/Raspberry Pi report brought forward some SLM thinking that I was toying with late last year. The article beneath is the result.
(I don’t know if it notifies you, but I gave you a shoutout at the end.)
I really appreciate you taking the time to write and share your article with me. And thank you for the shoutout at the end! That's so kind of you.
I'm excited to read "The Missing Middle" and see where your SLM thinking took you. It's always great when something sparks a deeper exploration like this. (I'm honored that I inspired you, even if in a small way.)
Thanks again for engaging with the work and for sharing your insights.
Mike, in amongst all the WOAH, you just helped me sort out what to get my Dad for his birthday soon 😂 We got him a Raspberry Pi several years ago and it is the most used little piece of tech ever! Now, we've going to up the game with the AI version and the control over data part will be a feature he will love so much.
Also, police bots are terrifying. Let's not do that 🤯
I cannot see a world where ads are a neutral addition to any LLM. Altman has already proven many times his words mean nothing, of course there is something they are selling business who choose to buy ChatGPT ads!
It's so funny that your father has a Raspberry Pi and loves it!
I love Raspberry Pi, too. I have a few of them. I mostly just use them for web browsing and retro video games, like Mario Kart and Zelda, lol. I keep telling people that I will make a birdfeeder camera with one. But, I haven't done it yet. The retro video games are excellent though. 10/10.
I also love how Raspberry Pi has lots of tutorials you can follow for various nerdy projects. If your dad likes to tinker, they're perfect.
The robot cop concept is fascinating engineering, but yeah. AI overlords patrolling humans is a dystopia I'd like to skip. 👀
As for OpenAI ads, I think the real shock is the speed. "Free tier monetization" was inevitable. But rolling it out this fast feels like they're racing toward something (or away from something). Either way, it's a signal worth watching.
Hope your dad has fun with the Raspberry Pi. Let me know if he builds anything cool with it!
So cool, Mike!! He uses it for electronic projects - random gadgets he creates for the house like testing if the peace lily has soil with the right moisture content or has the mail lady delivered anything in the mail box, he's also got a sensor in there! I get over tracking the inventions as they seem to make life more complicated 😂
Interesting shift. Freemium AI makes access broader, but policing the line between open use and responsible safeguarding is going to be the real challenge
I completely share your concern. The line between open access and responsible safeguarding is incredibly tricky. Honestly, I have no clue how to fix it. I'm waiting for someone way smarter than me to figure this out, lol.
In the meantime, I will continue advocating for free, low-cost, and open AI systems. They become even more valuable as we wrestle with harder questions about responsible use, accessibility issues, and the growing number of freemium tiers Big Tech will likely introduce in the coming years.
Insightful post as always. 😊 Ads were bound to happen, I’m almost surprised they took so long to roll them out, but I guess in the end it’s a good strategy, they first got us hooked quite well. Hehe
Thank you so much for reading and writing. I'm honored! 😊
I think you're 100% correct. Over the last few years, OpenAI's goal was likely to get as many folks using their platform as possible. Now that they have millions of "free" end-users, lol, they need to figure out how to make some money from them. ;)
The "get them hooked first" strategy is classic tech playbook. It definitely worked on me!
Thanks again for engaging. I really appreciate it.
great post.
ai goes ads…. uhg! and i bet the others will follow. for now while it’s just openai, im betting its not going to help their market share
Hey ToxSec!!!
I agree 100%.
I think if OpenAI succeeds with their "ad gambit", all of the other heavy hitters will have a "freemium" tier soon to follow. 😢
(I just hope uBlock Origin will be able to block all of the AI ads when they roll out soon enough, lol.)
;)
Cordially,
Mike D
absolutely. i worry about the more insidious threat, having chatbots direct users to preferred products and solutions.
“what’s the best backpack should i get for thru hiking?”
[responds with secretly paid for reference instead of the best one]
completely invisible to users!
Mike
Great write up as always.
Before reading another word, press play on The Pusher.
Yes. Steppenwolf. That one.
Volume at “this isn’t entertainment anymore.”
If the lyrics feel uncomfortably accurate for 2026 so far, you’re exactly where this was meant to land.
The OpenAI “Hail Mary”
“May I pass along my congratulations for your great interdimensional breakthrough. I am sure, in the miserable annals of the Earth, you will be duly enshrined.”
— Lord John Whorfin, The Adventures of Buckaroo Banzai Across the 8th Dimension
It is heartening to see that $20 billion in revenue and nearly 2 gigawatts of power, enough to run a small country or a very aggressive hairdryer, finally resulted in the same business model as a 2010 flashlight app. I can’t wait for my philosophical inquiry into the nature of consciousness to be interrupted by a 15 second unskippable ad for Tactical Grilling Aprons.
GPT-5.2:
“The meaning of life is… but first, have you joined the 500 million players in RAID: Shadow Legends? Use code OPENAI to get a free Legendary Champion, 50,000 Silver, and access to the Secret Cow Level if you act now, while we process your existential crisis. Found in the pursuit of purpose.”
This isn’t a failure of technology.
It’s capitalism completing the tech tree.
The Donation That Achieved Escape Velocity
And then there’s the donation.
Not an investment. Not a contract. Not a convertible note. A donation. A thing you give away because you no longer wish to own it.
Which makes the attempt to sue for roughly 3,600 times its original value feel less like litigation and more like discovering your old couch has secretly been accruing venture capital returns in the garage.
This is not a 3600x comeback story.
This is a gift that allegedly appreciated faster than Bitcoin, Nvidia, and several emerging economies combined.
The theory appears to be:
“I donated this freely, without expectation… except the expectation that, years later, it would mature into $138 billion.”
At this point, philanthropy isn’t generosity. It’s time delayed arbitrage.
Donate now. Wait patiently. Sue later. Let compound irony do the rest.
In that case, I’d like my money back from the Red Cross.
Not because of fraud.
Not because of misrepresentation.
But because we now need it.
For our ketamine addiction.
I mean, sorry, for depression.
I donated under the wildly outdated belief that a donation was a gift.
Apparently, the modern interpretation is that it’s a revocable emotional asset, redeemable later when things don’t go great and the stock market hurts your feelings.
“Dear Charity,
I gave you $50 in good faith.
Due to unforeseen psychological market volatility,
I now require it back with appreciation
for medical reasons.
Please advise.”
This is not philanthropy.
This is self care with a clawback clause.
The Beijing RoboCop
Nothing says “the future is here” like a humanoid robot barking at cyclists in Anhui while the government bans the very chips that could make it smart enough to recognize a bicycle. We are officially living in a world where the robot has a badge, a reflective vest, and the processing power of a singing birthday card, paired with the unshakable confidence of something that has never once been wrong because it has never once been allowed to doubt.
God help you if you’re holding a screwdriver.
Not menacingly. Just tightening your license plate so it stops rattling down the highway. But the robot has already logged: metal object detected. Human intent unclear. Escalation protocol engaged.
Context is unavailable. Nuance is deprecated.
Please remain still while the system consults a laminated decision tree that ends, as all good bureaucratic systems do, in Maximum Response Just to Be Safe.
It’s Blade Runner if the replicants were innocent, the cop was a kiosk, and every tragedy began with the words:
Error: Human behavior outside expected parameters.
The Raspberry Pi “Escape Pod”
The solution to global AI surveillance is apparently a $130 circuit board that looks like it was salvaged from a 1990s VCR. If you want to keep your data private, you just have to accept that your AI assistant now has the memory of a goldfish and the speed of a tectonic plate.
But that’s the trade.
You can have omniscience with ads, subpoenas, and surprise policy updates.
Or you can have a small, loyal idiot humming quietly on your desk that has never once tried to sell you anything.
Final Thought
This week didn’t reveal a fork in the road.
It revealed the menu.
Option A: Free intelligence, subsidized by your attention.
Option B: State intelligence, fully uniformed and deeply confused by context.
Option C: A shoebox computer that forgets your name every 20 minutes but has never asked for a credit card.
Choose wisely.
And finally, as leader of the free, I mean ad supported free world, since you didn’t pick me for your team in dodgeball and I didn’t get a gold star on my homework, I have no choice but to declare that all your base are belong to us.
Dear Mirror Malfunction,
After reading your reply, I realize now that I require a legal disclaimer and a moment of silence. (Or, as you advised last week, lawyers, guns, and money.)
I love your concluding thesis about the menu. We have a list of tradeoffs. All of them bad.
Also, for the record, if philosophical inquiry ever gets interrupted by a Tactical Grilling Apron ad, then I will gladly use the Raspberry Pi!
;)
Thanks for sharing your insights. They were wonderfully disturbing.
(I'll see you in the 8th dimension, assuming the Secret Cow Level doesn't get us first.)
Cordially,
Your nerdy friend,
(And still ad free. For now.)
Mike D
This connects back to our exchange the other week, Mike, about “physical work as a safe harbor.” 😊
We landed in roughly the same place: it’s not safe — it’s buffered. The China traffic police robots make that buffer visible, as early execution under very specific deployment conditions.
What’s shifted for me since then is how similar this looks to what’s happening inside organizations. Displacement isn’t just about capability — it’s about reliability and cost per productive hour.
In that sense, both the “safe harbor” story and the “imminent takeover” story were incomplete. The real variable is timing.
Hi Iwette! 😎
I love how you tied this back to our earlier exchange about the Boston Dynamics robots + DeepMind integration.
I also think your "buffered, not safe" framework is back into play. The Chinese traffic police robots are the perfect real-world use case. They operate in an area that's relatively constrained, with repeatable conditions, and relatively low need for babysitting.
I’m very curious to see how these systems perform in sustained, real-world deployment. But who knows if we'll get to see any deployment data. Regardless, it feels like 2026 is shaping up to be a real-world inflection point for embodied AI. (The year of the robots?)
Thanks again for such thoughtful comments and for pulling these threads across time and domains. This kind of longitudinal thinking is precisely what I'm trying to cultivate here.
Hope you have a beautiful week, Iwette.
Cordially,
Mike D
This is such a clear way of laying out what’s actually at stake right now, which I would otherwise have no clue about. Not just new tools, but who controls them, who profits, and what we quietly give up in exchange for convenience. I always appreciate how you connect everyday user choices to much larger structural shifts without turning it into panic or hype. It helps the moment feel understandable instead of overwhelming.
BRITT!! 🙏🔥
Comments like THIS are ones I like to save for when I'm having a lousy day.
THANK YOU.
I love how you mentioned structural shifts. I spend so much time trying to articulate what all these stories mean in the real-world, in ways that impact all of us.
Your comments always make me feel like someone's actually getting what I'm trying to do, which honestly keeps me writing when I'm wondering if anyone cares. (And, it makes all of the work worth the while.)
Thank you for being such a thoughtful reader and friend. This made my whole week. 💙
Cordially,
Mike D
Totally looking into a raspberry pi after this!
Hi Natalie!
Thank you so much for reading and commenting.
Honestly, I was UNSURE if I was going to include the Raspberry Pi article, lol.
(I felt like folks wouldn't care at all.)
But, it seems like it was everyone's favorite.
I'm so glad I included it!
Cordially,
Mike D
Yeah you absolutely sparked my interest. It was a great read! You brought a lot of things to my attention!
I hope this isn't too random but I have a question! Whats your take on decentralize ai vs. centralized ai?
I've been looking into this company called Bittsenor – its a decentralized AI platform (I can explain more, but I'm already feeling like I'm accidentally re-routing the convo too much). But I ask because I think centralized has always made me nervous as it grows - i think of surveillance, control, police robots, etc.
I'd love to know your take– from a curious individual who is only looking to learn :)
Hi Natalie!
Your kind words made my day. Thank you so much.
I want you to know your questions are always welcome here. I’m thankful there are like-minded nerds who want to dig into these nerdy computer science topics, lol.
The biggest difference between centralized vs. decentralized AI is who holds the power.
With centralized AI, a small number of companies train the biggest models. They also run the infrastructure and set the rules. That usually means the experience is smoother and the models are stronger. But it also means one gatekeeper can decide what’s allowed, how it’s priced, what gets logged, and how the system evolves. The problem with centralized AI is that surveillance, control, and mission creep can become very easy to scale. 👀 (Your concern is totally warranted.)
With decentralized AI, the goal is to spread that control out across a network so no single entity can unilaterally steer everything. That can reduce the risk that one company can flip a switch. But it doesn’t magically make things safer or more ethical. Playing devil’s advocate, decentralized systems can also make it harder to enforce standards, prevent abuse, or even know who’s accountable when something goes wrong.
So centralized AI optimizes for capability and convenience. Decentralized AI optimizes for reducing single-point control, but you trade off simplicity and clear accountability.
As for Bittensor specifically, I haven’t dug deep enough into their architecture to have a strong take yet. That said, the concept of decentralized AI networks is definitely worth watching. If you explore it, I’d love to hear what you find!
Thanks again, Natalie, for your fun questions. I hope you have a beautiful day. And if you have more questions, feel free to ask.
Cordially,
Mike D
Always happy to be among fellow nerds who enjoy digging into this stuff!! 🙂
Your take around where power sits def helped me think about this, especially the tradeoff between capability/convenience and accountability. That devil’s-advocate point you raised is exactly why i'm curious about this!
I’ve been loosely following projects like Bittensor, which seem to be experimenting with “tokenizing intelligence” in a way that’s loosely analogous to how Bitcoin incentivizes security. But its a bit over my head
I also follow Michael Casey, who advocates for decentralized AI but openly acknowledges many of the same risks you mentioned. It makes me wonder whether the future isn’t purely centralized or decentralized, but some kind of coexistence where each keeps each other in check
Anyway, thank you for engaging my question so thoughtfully, this definitely gave me more to think about!
Natalie,
Thank you so much for this thoughtful note. You made my day!
And it’s always a joy bumping into a fellow nerd in the wild. We should really come with warning labels. Lol.
Also, totally with you on the coexistence possibility. My gut says the future is a hybrid where centralized systems win on convenience and capability, and decentralized systems help keep power from concentrating too tightly.
I just searched for Michael Casey and found a bunch of his interviews on YouTube, so I’m going to check some of those out. ;)
Cordially,
Mike D
haha we **should** come with warning labels!!
Thanks, Mike! Another good tech-savvy-but-not-jargon-heavy informative article. As someone who worked in local law enforcement and was Army military police a very long time ago, I can appreciate a robot taking over traffic duty! Could be tedious, thankless, and occasionally dangerous. But then that mission creep after that, right? What comes next may not be something I would appreciate.
Daniel,
Totally agree. Robots doing traffic duty is genuinely cool. If it reduces risk for humans, that’s a win.
I also share your concern about mission creep. Do we really want anyone building legions of these things without clear limits and oversight? 👀
Also, I didn’t know you worked local law enforcement and Army MP. Respect. I spent years doing technical support work at Hanscom AFB. (I was never enlisted). Picture a heavily bearded, four-eyed nerd trying to look squared away around disciplined pilots and officers, lol. Good times.
Cordially,
Mike D
Too funny! But, hey, thanks for lending your knowledge, skills, and abilities to the Air Force! I was an MP back in the 80s when the Cold War still raged. I was stationed in West Germany (yes, it was still West back then) and Fort Hood Texas (not sure if it is called Fort Hood anymore.
PS: Fort Hood was renamed Fort Cavazos on May 9, 2023. It was renamed back to Fort Hood on June 11, 2025.
Dear Daniel,
It was genuinely my pleasure to support the USAF. Stressful at times. But a lot of fun too. (And yes, they fed me well, lol).
And truly, thank you for YOUR service. Being an MP in the 80s, with the Cold War still raging, especially while stationed in West Germany, must have been intense. (I'm sure you have stories.)
Also, I think Fort Hood may go by a different name now. But either way, what a pair of assignments!!!
Thanks for sharing, Daniel. Hope you have a beautiful week.
;)
Cordially,
Mike D
Thanks, Mike. Oh, I got stories alright, both good, bad, and ugly. lol. I loved Germany, though. Would love to visit there someday again. I hated being at Fort Hood after Germany! :-)
Daniel,
I can totally see that.
Germany sounds like a grand adventure. I imagine Fort Hood felt more like going to the principal’s office after Germany.
;)
Hope you enjoy a beautiful weekend!
Cordially,
Mike D
Very interested in the private AI opportunity so thanks for the heads up.
I love the joke going around that OpenAI has finally defined AGI. Ads Generating Income.
Hey Joe!
Thank you so much for replying.
Lol. I love your AGI joke. It's devastatingly accurate. "Ads Generating Income" is the breakthrough OpenAI has been building toward all this time. 😂
Are you going to try running some local LLMs on a Raspberry Pi? I'm a huge fan of the ecosystem. (Full disclosure: I also use mine to play retro video games when I should be working on AI research. It's a surprisingly capable little retro-console, lol.)
That said, you definitely don't need a Raspberry Pi to use local LLMs. You can run private, open-source LLMs on almost any decent laptop these days. But there's just something cool about the Raspberry Pi.
Thanks again for writing and contributing to the work! It means so much to me.
Hope you have a great day.
Cordially,
Mike D
I am hoping, because I already pay, we (the paying subscribers) will not be subject to ads any time soon.
As I read your very thorough research, the question I haven't yet looked into: What are the repercussions of me losing all my ChatGPT projects? There are a few very keyboard growing out my business.
How do I back up everything easily? Copy and paste to a couple of hard drives?
Hi Christine! 😎
Thank you so much for writing and for your kind words.
I have great news for you. Based on what OpenAI has shared so far, ads are expected to apply only to "free" and "ChatGPT Go" accounts. Paid plans such as Plus, Pro, Business, Enterprise, and Edu are not supposed to include ads. So, as a paying subscriber, you should be safe for now. Thankfully.
On backing things up, I actually looked into this recently. You can export your ChatGPT data, including chats, from your account settings.
1. Click your profile icon and go to Settings
2. Select Data Controls and then Export Data
3. Confirm, and download the ZIP file from the email they send
However, I should mention that after researching this, many people complain that it doesn't contain ALL of the data. And some old chats, projects, or conversations might get excluded. So it seems rather quirky, unfortunately.
I also confess that I am a bit messy with my chatbots myself. I use ChatGPT, Claude, Gemini, and Perplexity, so things can end up scattered. What has helped me most is keeping local copies of my essential prompts and outputs, often as simple PDFs. It feels a little archaic. But it gives me peace of mind.
For your most critical projects, manually copying key prompts, instructions, and outputs into a document and backing that up to a couple of drives or cloud storage is still the most reliable method I know right now. It is tedious, but dependable.
Thanks again for reading and for raising such an important question!
Hope you have a beautiful week.
Cordially,
Mike D
Thank you Mike for your reply. I appreciate you and the time you invest in helping me / us! I will take your advise with respect to backing up, before I get too far down the rabbit hole.
Wishing you a blessed and lovely rest of your week.
Christine
Christine,
Thank you so much for your kind words!
I love researching this nerdy AI stuff.
I'm thankful and honored that there are like-minded folks like you to brainstorm with around here.
:)
Thanks again, and I hope you have a gorgeous week!!
(Let me know if the ChatGPT backup breaks, lol.)
Cordially,
Mike D
Thank you for this Post, @MrComputerScience. Your AI/Raspberry Pi report brought forward some SLM thinking that I was toying with late last year. The article beneath is the result.
(I don’t know if it notifies you, but I gave you a shoutout at the end.)
https://reciprocalinquiry.substack.com/p/the-missing-middle
Hey Ruv! Thank you so much!
I really appreciate you taking the time to write and share your article with me. And thank you for the shoutout at the end! That's so kind of you.
I'm excited to read "The Missing Middle" and see where your SLM thinking took you. It's always great when something sparks a deeper exploration like this. (I'm honored that I inspired you, even if in a small way.)
Thanks again for engaging with the work and for sharing your insights.
Cordially,
Mike D
Mike, in amongst all the WOAH, you just helped me sort out what to get my Dad for his birthday soon 😂 We got him a Raspberry Pi several years ago and it is the most used little piece of tech ever! Now, we've going to up the game with the AI version and the control over data part will be a feature he will love so much.
Also, police bots are terrifying. Let's not do that 🤯
I cannot see a world where ads are a neutral addition to any LLM. Altman has already proven many times his words mean nothing, of course there is something they are selling business who choose to buy ChatGPT ads!
Hi Dallas! 😎
It's so funny that your father has a Raspberry Pi and loves it!
I love Raspberry Pi, too. I have a few of them. I mostly just use them for web browsing and retro video games, like Mario Kart and Zelda, lol. I keep telling people that I will make a birdfeeder camera with one. But, I haven't done it yet. The retro video games are excellent though. 10/10.
I also love how Raspberry Pi has lots of tutorials you can follow for various nerdy projects. If your dad likes to tinker, they're perfect.
The robot cop concept is fascinating engineering, but yeah. AI overlords patrolling humans is a dystopia I'd like to skip. 👀
As for OpenAI ads, I think the real shock is the speed. "Free tier monetization" was inevitable. But rolling it out this fast feels like they're racing toward something (or away from something). Either way, it's a signal worth watching.
Hope your dad has fun with the Raspberry Pi. Let me know if he builds anything cool with it!
Cordially,
Mike D
So cool, Mike!! He uses it for electronic projects - random gadgets he creates for the house like testing if the peace lily has soil with the right moisture content or has the mail lady delivered anything in the mail box, he's also got a sensor in there! I get over tracking the inventions as they seem to make life more complicated 😂
Hey Dallas!
I love that your dad is out there building little Raspberry Pi gadgets for everyday life. That is excellent.
And I totally get what you mean about it making things more complicated. One project turns into another fast.
I’m really glad I shared that Raspberry Pi story. I had no clue how popular these things are, lol.
Hope you have a beautiful weekend, Dallas!!
Cordially,
Mike D
Have a great weekend too, Mike!
Interesting shift. Freemium AI makes access broader, but policing the line between open use and responsible safeguarding is going to be the real challenge
Hey Pietro!
Thank you so much for responding.
I completely share your concern. The line between open access and responsible safeguarding is incredibly tricky. Honestly, I have no clue how to fix it. I'm waiting for someone way smarter than me to figure this out, lol.
In the meantime, I will continue advocating for free, low-cost, and open AI systems. They become even more valuable as we wrestle with harder questions about responsible use, accessibility issues, and the growing number of freemium tiers Big Tech will likely introduce in the coming years.
Thanks again for engaging with the work.
Cordially,
Mike D
Insightful post as always. 😊 Ads were bound to happen, I’m almost surprised they took so long to roll them out, but I guess in the end it’s a good strategy, they first got us hooked quite well. Hehe
Hey Cristina!
Thank you so much for reading and writing. I'm honored! 😊
I think you're 100% correct. Over the last few years, OpenAI's goal was likely to get as many folks using their platform as possible. Now that they have millions of "free" end-users, lol, they need to figure out how to make some money from them. ;)
The "get them hooked first" strategy is classic tech playbook. It definitely worked on me!
Thanks again for engaging. I really appreciate it.
Hope you have a beautiful week.
Cordially,
Mike D