Too Much of a Bad Thing is a new thing in AI. ARD #117
Today’s theme: Too much of a bad thing can be a thing. We’ve all heard the phrase “too much of a good thing” — well, too much of a bad thing can be a thing too, even when good was the original intent. Three separate AI events today make the point: AI ‘vibe coding’ and its emerging quality-control problem in open source, leading economists piling in on AI’s coming impact on jobs, and the exploding complexity of choices inside every frontier AI model. Three events for the AI Tech Wave — each with my Take first, then my Overall Take.
(1) AI Vibe Coding Has a Quality-Control Issue — Especially on the Open-Source Side
MP TAKE: Vibe coding — the term coined by Andrej Karpathy, a coder’s coder, one of the original OpenAI co-founders and an architect of Tesla’s self-driving AI, now at Anthropic — is a net good thing. It means almost all of us, not just advanced programmers, can now build software with our voice and tools like ChatGPT, Gemini and Claude Code. But here’s the part almost nobody thinks about: over 80% of the world’s software runs on open source — the generosity of strangers, thousands of small, passionate, very technical teams maintaining little bits of code that make everything work. And AI is now flooding those projects with machine-generated code submissions — mostly well-intentioned — that still have to be evaluated, monitored, measured and fixed by human maintainers.
The problem is a lopsided balance: a flood, versus the resources to manage said flood. Closed-source companies can fight fire with fire — they have compute budgets and AI agents to review AI-generated code. Open-source maintainers mostly don’t, because that compute is expensive. So the next time some gizmo or app of yours doesn’t quite do what it’s supposed to, it may not be the company’s fault at all — it may be some bit of open-source plumbing that got snarled up by vibe-coded updates that weren’t flagged and triple-checked. Given how much of the world runs on this software, it’s a key new issue to have on your radar.
Sources, in narrative order: FT — Who cleans up after the vibe-coding party?. For longtime readers, in narrative order: ‘Anthropic’s moment in the Vibe Coding sun’ in AI-RTZ #960; ‘Developers loving AI Vibe Coding’ in AI-RTZ #661; and ‘How Nvidia and Apple can be the global US open-source AI champions vs China’ in AI-RTZ #1089.
(2) Leading Economists and Technologists Pile In on AI Job Concerns
MP TAKE: Almost 200 highly visible economists and technologists — a bunch of them Nobel laureates — signed on to a warning that AI could cause job losses faster than the lagging benefits arrive, with Anthropic’s co-founder Jack Clark leading the charge. It’s a never-ending issue with technology, and I’ve been watching it for over thirty years, on every tech wave. They cite the Industrial Revolution, electricity, the railroads — all massive change, good and bad. Within about ten years of the car, horses disappeared from our roads (which got a lot cleaner, and then we had traffic). The pattern is old; what’s new is the pace — and now hundreds of economists are running their models on it.
Here’s my issue with the models: you can absolutely forecast the X number of jobs lost when AI replaces people. What the models cannot measure is the Z number of things that weren’t possible before — the new work that suddenly needs a hundred times more people, arriving faster than the same economists envision. It’s not a failure of vision: those things haven’t been invented yet. And I have never seen a tech wave move at this pace with this many gargantuan innovations, where the opportunities on the surface qualitatively outweigh the relative costs. Meanwhile the piling-on adds to already-historic highs of public concern — regulators will pay attention, and people can be politically enraged to stand in front of data-center sites with signs. Remember: when cars were invented, people seriously demanded they be capped at the speed of horses so as not to scare them. That was real — it’s in books. For Anthropic, the clarion cry burnishes the safety brand. But it exaggerates the glass half empty, and barely looks at the glass half full. The arguments should flow freely on all sides — this one just keeps playing the same tune.
Sources, in narrative order: NY Times — Nearly 200 economists and tech leaders warn of AI threats. For longtime readers, in narrative order: ‘Jensen vs Dario debate on AI jobs impact’ in AI-RTZ #789; and ‘Anthropic founder/CEO Dario Amodei’s latest AI essay’ in AI-RTZ #980.
(3) Too Many Choices in Every Frontier Model — Massive Cognitive Load on Early Users
MP TAKE: The good news: we have a lot of great new models to try — Anthropic’s Fable 5 and OpenAI’s GPT-5.6 at the top, and the ‘best of the rest’ from Meta, SpaceXAI and others that I covered last week. The bad news: there’s now an explosion of choice within the models. Each company ships three or four sizes — large, medium, small — and each size comes in different intensities: light, medium, fast. Do the combinatorial math and you’re staring at 20 or 30 different choices per model before you’ve even decided what task to run or how many agents to spin up. Advanced users end up watching Pareto curves of input and output token costs — bills that can run tens of thousands of dollars a day for developer-types, and hundreds for regular users, depending on the choice.
Why is this happening? Because the labs are balancing scarce, super-expensive AI data-center compute against the maximum revenue they can earn — a la carte, tiered pricing — which matters urgently to at least two of them, OpenAI and Anthropic, ahead of their mega-AI IPOs this year and next. The result: advanced developers post 90-minute YouTube videos just explaining which GPT-5.6 variant to use for which task — and after watching, you might have a couple of ideas. And then of course you feel forced to watch similar hour plus videos on how to best select and use all the variants of Anthropic’s Claude Fable 5. Then Meta’s Muse Spark 1.1. Then Grok 4.5. Then… well, you get the idea. You end up spending as much time evaluating these things as you save using the AI in the first place. WIth your head exploding with choices and the ongoing math of ‘how much compute I’ve burned’, and ‘was it worth it’? Especially when the boss or the client being billed comes asking that question. That’s a huge cognitive load on precisely the most avid customers and fans — and they’re complaining. I’ve seen versions of this in every tech wave, but never to this degree. Yes, it’ll get better and simpler. For now, it’s friction landing on the people who love these products most.
Sources, in narrative order: Axios — Finding your goldilocks in OpenAI’s GPT-5.6 model. Theo (t3.gg) — ‘This is absolute chaos’ — saving money with OpenAI GPT-5.6. For longtime readers, in narrative order: ‘Anthropic recasts Mythos as Fable 5’ in AI-RTZ #1114; and Friday’s ‘The Best of the Rest AI Models’ in ARD #116.
MP OVERALL TAKE
Each of today’s events is really about the same thing: the tyranny of choice. We’re familiar with it choosing what to watch on Netflix over the years. Takes as much time to figure out what to watch as watch the thing. AI is more complicated. The a la carte costs can blow up your budget in minutes. When so much comes in on the back of a good thing — AI — you suddenly have to cope with it: in selecting models, in weighing the risks versus benefits on jobs and the economy, and in the sheer volume of code produced by models, humans, and combinations thereof overwhelming the open-source systems that make our world run.
So the discipline is to watch every swing of the pendulum in the key areas of AI — especially as they gather momentum to swing too far on either side. Those are the places where over-reactions do the most damage, to the pendulums’ swings or to whether they exist at all. Over time this stuff gets solved. But in the near term, the swings are getting super wide, and they need to be watched more intently in this tech wave than in most prior waves I’ve seen. There are no easy one-point solutions to any of these — there are lots of partial ones — so all of the stakeholders need to stay vigilant. That’s the challenge, and the opportunity, of too much of a good thing turning into too much of a bad thing.
Gadget AI — How to Quickly Understand the Exploding Choices in USB-C Cables
MP Take: This one is geeky but it’s right under our nose With countless USB-C cables of all types and capabilities bunched up in our drawers. Without USB-C, we would not have the amazing connected technology world we have — a very prosaic item we never think about. The port was accidentally invented at Intel by an Indian-American engineer trying to fix a printer-cable problem for his wife; over decades it grew into an open standard, developed and productized by thousands of companies around the world — invented by no one company, which is also why there’s no industry solution to its big everyday annoyance: USB-C cables come in all manner of capabilities — the power they carry, the data they move, the wattage they charge at — and none of them has any display. There’s no physical way for a human to tell which cable does what, especially staring into a drawer full of them, grabbing one to charge your iPhone or some gizmo.
So the fix, for now, comes from software: The Verge covered a free Mac app that reads out the truth about whatever mystery cable you plug in. That’s the solution being cited — geeks can crack it — but the complexity is only going up, especially in the AI era as these cables carry huge amounts of data under improving standards. You often can’t tell a cable is wrong until you plug it in and something doesn’t work. We need more software and hardware solutions around this unsung hero — which leads straight to today’s questions.
Sources, in narrative order: The Verge — This free Mac app reveals the truth about your mystery USB-C cables. Google Gemini — The complexity of USB-C cables. The Atlantic — The broken promise of USB-C.
Questions
Q1 — What’s Michael’s favorite must-carry USB-C gadget?
Starting on a positive note: a power bank with a display. Mine happens to be from Anker, but others make them too. Plug in all manner of USB-C cords and the display tells you what’s being charged, at what power, at what data capacity. It has a convenient lanyard that’s also a USB-C cable, and a bulky enough battery to charge a whole bag of gizmos — and since airlines limit you to one of these things anyway, this is my gizmo of choice for cables, charging and power in one.
Q2 — What’s Michael’s most fervent wish with USB-C cables?
That Microsoft with Windows, and Apple with macOS, iPadOS and iOS, build this into their operating systems. Whenever any USB-C cable is plugged into a phone, tablet, laptop or computer, it should automatically pop up all the information: this cable can do this much power, this much wattage, these data speeds — all the capabilities, on both sides of the cable. Built-in. Out of the box. No extra gadgets needed, and no hunting for the kind of free software The Verge covered. That would be genuinely helpful — and it’s all knowable from the cable’s own electronics.
Wrapping up
Today’s AI-RTZ #1146 — OpenAI draws Apple ‘First Blood’ — is on Apple coming after OpenAI with a lawsuit over AI devices and talent. OpenAI has hired over four hundred people to work under Jony Ive on all manner of AI devices — potentially even a competing AI smartphone — and Apple cites at least two ex-employees, one of them nearly twenty-five years at Apple, alleged to have taken files, parts and information they shouldn’t have, and to have inculcated a culture of taking more Apple technology than they should. It’s a real issue for OpenAI ahead of its expected IPO in the next twelve-ish months. Lots of pros, cons and issues — recommended as today’s reading.
Tomorrow — ARD 118 on AI-RTZ 1147.
Thanks for joining us today, AI Curious Folk. Stay tuned.
— MP
Full Source Reading —
For the broader context, see the canonical sources for ARD 117 — in today’s narrative order:
Event 1 — AI Vibe Coding’s Open-Source Quality Problem
-
AI-RTZ #960 — Anthropic’s moment in the ‘Vibe Coding’ Sun · AI-RTZ #661 — Developers loving AI ‘Vibe Coding’ · AI-RTZ #1089 — How Nvidia and Apple can be the global US open-source AI champions vs China
Event 2 — Economists Pile In on AI Jobs
-
NY Times — Nearly 200 economists and tech leaders warn of AI threats
-
AI-RTZ #789 — Jensen vs Dario debate on AI jobs impact · AI-RTZ #980 — Anthropic founder/CEO Dario Amodei’s latest AI essay
Event 3 — The Tyranny of Choice in Frontier Models
-
Axios — Finding your goldilocks in OpenAI’s GPT-5.6 model · Theo (t3.gg) — ‘This is absolute chaos’ — saving money with GPT-5.6
-
AI-RTZ #1114 — Anthropic recasts Mythos as Fable 5 · ARD #116 — The ‘Best of the Rest’ AI Models
Gadget AI — USB-C Cable Complexity
-
The Verge — This free Mac app reveals the truth about your mystery USB-C cables · Google Gemini — The complexity of USB-C cables · The Atlantic — The broken promise of USB-C
Clips from today
Segment 1 — The Tyranny of Choice in AI Models
Frontier models now ship in 20-30 variants each — sizes, intensities, price tiers — before you even pick a task. Advanced users watch Pareto curves of token costs; bills run tens of thousands a day at the high end.
MP Take: It’s a tyranny of choice — like Netflix, but worse: wrong choices lose you real money, not just time. The labs are balancing scarce, expensive compute against maximum revenue ahead of their mega-IPOs. You end up spending as much time evaluating models as you save using them.
Segment 2 — AI Job Losses: The Unseen Opportunities
Nearly 200 economists and tech leaders — many of them Nobel laureates — warn AI could destroy jobs faster than the benefits arrive.
MP Take: The models can forecast the losses, but they can’t measure the things that weren’t possible before AI — the opportunities haven’t been invented yet. Same movie every wave: when cars arrived, people wanted them capped at the speed of horses.
Clip 1 — AI’s Unseen Opportunities
Economists’ models can forecast AI’s job losses — but not the new things that were impossible before AI, which arrive faster and bigger than anyone envisions.
MP Take: The models don’t measure the opportunities because those haven’t been invented yet. I’ve never seen a wave with this many gargantuan innovations where the surface opportunities outweigh the relative costs.
Clip 2 — USB-C Cable Info Wish
Every USB-C cable carries different power and data capabilities — and nothing on the cable tells you which one you’re holding.
MP Take: My fervent wish: Windows, macOS, iPadOS and iOS should pop up a cable’s power, wattage and data speeds the moment you plug it in. Built-in. Out of the box. No extra gadgets needed.
Clip 3 — USB C: The Unsung Hero of Tech
Without USB-C we wouldn’t have the connected world we have — a standard born at Intel from one engineer’s printer-cable problem, then built out by thousands of companies.
MP Take: We use it every day and take it for granted. The complexity is going up, and you can’t tell what a cable does until it doesn’t work. We need better software and hardware solutions around this prosaic little hero.
Clip 4 — USB-C Power Bank with Display
MP’s one must-carry USB-C gadget: a power bank with a display that shows exactly what’s being charged, at what power and data capacity.
MP Take: This one’s from Anker, but others make them too. The lanyard doubles as a USB-C cable, the battery charges a bag of gizmos — and airlines limit you to one anyway. My gizmo of choice.
About AI Ramblings Daily (ARD), and AI-RTZ
Both are daily. Both are free. Both are about AI. But they’re different mediums carrying different messages.
AI-RTZ is the morning text — a deeper written take on one idea, published by at least 5 AM EST. Today: post #1146.
AI Ramblings Daily is the afternoon video + podcast — my ad hoc takes and perspective on the day’s AI issues & news flow, around 20 minutes, with short 1-2 minute clips for quick topic views. Today: episode #117.
Subscribe to either or both on michaelparekh.substack.com. They run as separate Sections you can opt into or out of.
Links used in today’s show (already embedded inline above; listed here for reference)
Take 1 — AI Vibe Coding’s Open-Source Quality Problem:
Take 2 — Economists Pile In on AI Jobs:
-
NY Times — Nearly 200 economists and tech leaders warn of AI threats
-
AI-RTZ #980 — Anthropic founder/CEO Dario Amodei’s latest AI essay
Take 3 — The Tyranny of Choice in Frontier Models:
Gadget AI — USB-C Cable Complexity:
Companion text:
(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here.)
Subscribe to AI: Reset to Zero for daily AI Ramblings + Sunday Bigger Picture posts.
