Yes. Those "subscriptions" are more like an auction, the way they are setup right now.

Yes. It's really sneaky. And try cancelling a subscription. I simply could not find the cancel button on an old subscription I no longer needed. I browsed many amazon.com pages. I could edit the subscription, skip, change frequency, etc. Cancel button nowhere to be found. Had to use AI to finally find it.

One really should be able to set a price limit. Re-ordering at 2x the original price is just deceptive. I would much rather be told it's not available, and shop elsewhere. That happened with wet cat food during the pandemic. Even Wal-mart was sold out. No price you could pay to get the right flavor of fancy feast.

Is what this service actually is, and should be called. Not subscription.

Here's the text reformatted for Slashdot's plain-text style — no tags, just spacing and ASCII conventions:

---

At least, that has been my experience.

Prompt I did today:

Is AI code generation excellent at maintaining software, keeping backwards compatibility, handling edge cases, and minimizing regressions?

--- Claude response ---

Short answer: not really -- at least not yet, and not without significant human oversight. Here's an honest breakdown:

Where AI code generation falls short on maintenance

Backwards compatibility is one of the harder problems. AI models have no persistent understanding of your codebase's history, contractual API promises, or downstream consumers. They'll refactor things that look "cleaner" without realizing they're breaking a public interface. They don't inherently know what's internal vs. external surface area unless you tell them.

Edge cases get handled inconsistently. AI is decent at the common cases it's seen many times in training data, but novel edge cases -- especially ones specific to your domain, your data shapes, or your infrastructure quirks -- are frequently missed or handled with superficially plausible but wrong logic. It also tends to handle the happy path well and paper over error paths.

Regressions are a real problem. Without a strong test suite that the AI is actually running and respecting, it will introduce regressions confidently. It has no memory of why something was written the way it was, so it will "fix" code that was deliberately defensive or working around a known upstream bug. Comments help, but only if they're there and specific.

Long-range consistency across a large codebase is poor. AI works well in a local context window but loses coherence across hundreds of files. Naming conventions drift, architectural patterns get inconsistently applied, and subtle invariants get violated.

Where it's actually useful in maintenance contexts

- Boilerplate changes (renaming, restructuring) with good test coverage to catch breakage
- Generating tests themselves (though coverage != correctness)
- Explaining legacy code before a human makes changes
- Drafting migration scripts or deprecation stubs with human review
- Catching obvious bugs during review (not as the sole reviewer)

The meta-issue

The core problem is that AI has no skin in the game. It doesn't run the software, doesn't get paged at 3am, and doesn't accumulate the institutional knowledge of why things are the way they are. The best results come when a human who does have that context is directing.

--- Gemini response ---

Can AI Actually Maintain Your Codebase Long-Term? (Spoiler: Not Really)

The short answer is no. LLMs are excellent for rapid prototyping, translating logic between languages, and generating boilerplate -- but they fall apart on the nuanced demands of long-term software maintenance. Here's the breakdown:

Maintaining Software: Poor to Fair

Even with massive context windows, an AI doesn't "understand" a repository of millions of lines with legacy dependencies and tribal knowledge. It treats code as a statistical text completion problem, not a systemic architecture. Worse, repeated use without strict human oversight introduces inconsistent design patterns, copy-pasted logic, and micro-architectural drift -- *increasing* technical debt rather than reducing it.

Backwards Compatibility: Poor

AI models don't know *why* a legacy quirk or undocumented behavioral contract exists. They routinely suggest "cleaner" refactors that quietly break public interfaces or change serialization formats. Even more fun: when asked to update code for a newer library, they'll frequently invent non-existent API methods or misapply breaking-change patterns from older versions. Hallucinated upgrades are a real thing.

Edge Cases: Fair (with an asterisk)

If you ask an AI to implement a function, it writes the happy path and calls it done. Null pointers, boundary conditions, race conditions -- mostly invisible unless you explicitly ask. The flip side: if you paste existing code and ask "what edge cases does this miss?", it can be surprisingly sharp. The capability is there; it just won't volunteer it.

Regression Prevention: Very Poor

This is the killer. AI generation is probabilistic. A subtle prompt change or temperature difference can cause an agent to silently rewrite a secondary logic block in a way that passes syntax checks but fails under specific production loads. AI cannot minimize regressions on its own -- it's entirely dependent on a pre-existing human-written test suite and CI/CD pipeline. And it'll happily write tests that validate its own broken logic rather than your actual business requirements.

Bottom Line

AI is a great accelerator, a mediocre maintainer, and a terrible steward. It excels at greenfield generation and rubber-duck debugging. It fails badly at the place where most engineering cost actually lives: safely evolving systems where three changed lines might break a downstream dependency written a decade ago. Until AI can hold deterministic, holistic understanding of large evolving systems, human engineering oversight isn't optional -- it's load-bearing.

--- ChatGPT response ---

No.

AI code generation is useful, but it is not "excellent" at long-term software maintenance unless heavily constrained and reviewed.

It is usually good at:

- writing boilerplate
- implementing isolated features
- refactoring small, well-scoped code
- explaining existing code
- generating tests when told exactly what behavior matters
- finding obvious bugs

It is weak at:

- preserving undocumented behavior
- understanding why ugly legacy code exists
- maintaining backwards compatibility across real users
- detecting subtle regressions
- respecting hidden invariants
- handling rare edge cases
- making minimal, surgical changes
- knowing when NOT to change something

The biggest risk is that AI tends to "improve" code by rewriting too much. That often breaks compatibility, changes behavior, removes workarounds, or simplifies away edge cases that were there for a reason.

For maintenance work, the safest pattern is:

1. Give it the existing code.
2. State: do not refactor unrelated logic.
3. State the exact allowed change.
4. Require a diff.
5. Require tests or manual verification steps.
6. Review every change yourself.
7. Run the full regression suite.
8. Use version control aggressively.

For your kind of scripts , AI is best treated as a fast junior developer with good syntax knowledge but poor operational judgment. It can save time, but it must be boxed in tightly.

When quoting in Firefox for Android, the post ends up in a different thread than the one I hit reply on. This has happened repeatedly.

I have been taking Ozempic since last August. The effects were definitely not felt early on. My BMI has gone from 29.9 to 25.7, and is still going down. The effects are slow but steady.

My observatuon is that it reduces appetite. I can now be full on the same size plates i grew up with in France. Not the much bigger serving sizes I got used to in America the last 30 years. This is a very positive change.

I'm on the minimum dose. I tried a higher one in december. The nausea ruined my holiday meals. That still took a few weeks at that dose to show up, not days. And many more weeks before it went away after dialing down.

I can't really speak to braim changes, though.

What you describe makes perfect sense in a commercial setting. Our house really is that large, but we still don't need hot water 24/7 at all fixtures, and there are only 2 of us. Our showers and baths in the master bathroom are served by the larger 80 gallon water heater which does not have a recirculation pump. The delay is about 1 minute and tolerable.

The smaller 50 gallon water heater serves the other 3.5 bathrooms, but the showers are only used when we have guests. The rest of the time, the hot water from that water heater is used only by the dishwasher and clothes washer. We cook with filtered water, and thus boil it separately on induction. Therefore, the water heater is on the lowest setting, and the circulation pump is off 99.9% of the time. My previous Miele dishwasher actually specified colder water inlet. The newer Bosch says to connect to hot water, so that's what was done. The LG clothes washer requires both hot and cold inlets, and I don't know what would happen if both were cold. If they could really function on cold only, I could just turn off that water heater altogether.

There is no known safe level of PFAS. I don't know for sure that I "need" to filter my tap water. I do it anyway because the cost is fairly low, and there is no downside to filtering. I did a few searches on PFAS standards in water in Europe, including Netherlands, vs the US, and did not find significant differences. If anything, the standards appeared to be stricter in the US, and the Netherlands had a similar "advisory" which was not legally binding. In any case, the devils are in the details - enforcement. And various pipes between the utility and your faucet could still introduce additional stuff. As I said, I don't see much downside in filtering drinking water.

No such thing as a cheap 3000W electric kettle in the US due voltage being 120V on most outlets. Most plug-in appliances are thus 120V, and normal circuits are 15 amps, which means they max out at 1800W. 240V is available. My induction cooktop uses 240V/50 amp circuit, but is hardwired. Wall oven is 240V/40amp and also hardwired. The only plug-in 240V appliances are typically clothes dryers, and EV chargers that plug-in to dryer plug.

I don't think it's worth bothering with a kettle. Some day, I will upgrade my induction cooktop to one that has a zone auto-shut off cooking timer, since I can barely hear the beep now, and 1 minute too long in the pressure cooker with induction means the rice is completely burnt. The only model I could find with this feature is a $4000 Miele, so I'll wait ! I understand some newer induction cooktops support the entire surface, can ad can autodetect cookware. I don't know if any could work with a mug, if a ferromagnetic one existed. That Miele doesn't have that feature, though. I'll probably wait until my cooktop dies. It's 16 years old and shows no signs of failure.

In my kitchen, I have an undercounter reverse osmosis filter, to filter out things like PFAS. I only drink water from that faucet. So do my cats, who eat food and get enough minerals. Ice cubes from my fridge also come from the RO line. I use that water for cooking, too. My induction cooktop is extremely fast and efficient at boiling water. I cook rice under 5 minutes in my pressure cooker - less than one minute to boil the water before closing the pressure cooker, and 4 minutes cook time.

For beverages such as tea, I have to use the microwave oven, and that takes longer than to boil the rice water with induction. The smallest induction element is 6in and 1500W on boost. That would be at least twice as faster as my microwave oven. Alas, ferromagnetic induction compatible mugs don't appear to exist. The only mugs advertised as induction compatible come with transfer disks, which defeats the point. They would need to be quite wide also, minimum 5in diameter to be detected on the 6in element, otherwise the cooktop won't turn on the magnetic field.

Thanks. Recirculation also causes pipes to wear out faster. A repipe job would be enormously expensive, since they run on the slab foundation. Most likely would need to move pipes outdoors if that is ever needed.
Good thing it never freezes here.

She is a he.

It is not only the Supreme Court recent rulings. Even before, it was also DOJ policy, which is not law, and at the discretion of the executive branch. Not surprisingly, no matter who is in charge, they enjoy exempting themselves from prosecution.

I did not say how much I was spending. Only what the increase would be from not using a circulation pump vs using one.

What makes those heaters you mentioned more efficient than others ? I have a commercial grade unit.

I have already replaced it. Same issue. There may well be quieter models, but I'm just very sensitive to the pump vibrations. There is a bedroom adjacent to the utility room where the pump is located, and while it is a guest room, it can interfer with sleep, depending on how sensitive the guest is - probably not my mother, since she can just take her hearing aids off. The home theater is also adjacent, and I want silences in movie to really be silences.

100W constant load is 876 kWh per year. The average kWh cost in the US is 17.65 cents. That is $154 a year. PG&E charges about 45 cents/kWh, so that is closer to $438. This is ignoring the additional energy cost, which will depend on many factors, such as the size of your water heater, heat setpoint, whether your pipes are insulated, and the length of your pipes. My pipes are very long - likely as long as 300ft, which explains the 3 minute delay without the pump at some fixtures. Running the pump 24/7 might add 100 therms of gas a year, which is around $280/year with PG&E currently. It would be even more if the 50 gallon water heater was conventional electric - as high as 2930 kWh/year or $1318/year at PG&E's rates.

In my very large home, some of the fixtures take 3 minutes if the pump is off. It wasted a lot of water.

The pump is loud, and vibrations can be hard in the entire home. It also wastes a significant amount of electricity. Constant unnecessary recirculation also wastes a lo of energy in the water heater.

I put the pump on a smart switch. I added home automation, with motion sensors near showers, and energy monitoring plugs for appliances, that trigger the pump only when needed. There is still a delay to get hot water, but it is shorter, and I don't have to suffer the noise and energy waste.

There is also a second water heater which only serves the master bedroom with a huge tub, and has no recirculation pump. The delay with that one is about a minute.