I tried testing 4.5 opus and 4.6 opus both with “high” thinking. Same box, same repo. I had them plan a moderate complexity refactoring on a small codebase.
Observations:
4.6 had previously failed to the point where I had to wipe context. It must have written memories because it was referring to the previous conversation.
As the article points out, 4.6 went out of its way to be lazy and came up with an unusable plan. It did extra planning to avoid renaming files (the toplevel task description involves reorganizing directories of files).
4.6 took twice as long to respond as 4.5.
I’m treating this as a model regression. 4.6 is borderline unusable. I’ve hit all the issues the article describes.
Also, there needs to be an obvious way to disable memory or something. The current UX is terrible, since once an error or incorrect refusal propagates, there is no obvious recovery path.
Anyway, with think set to high, I see drastically different behavior: much slower and much worse output from 4.6.
> Also, there needs to be an obvious way to disable memory or something.
Memory files are stored in a path under ~/.claude somewhere. It's fairly easy to find (I'm just not typing this on a PC with Claude on it atm), and from memory (heh) it's in Markdown.
If you nuke the memory file(s) then you should be good. Oh, I think the memory files are project or directory scoped from memory (heh again) too, so you should be able to keep/remove things manually without losing important stuff if you want.
> Anyway, with think set to high, I see drastically different behavior: much slower and much worse output from 4.6.
Might be worth trying the CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING setting then?
I wonder how much adding a profiler to development flows would help modern apps.
JS is gross, but 16ms (time you get to render a frame at 60 fps) is an eternity on modern systems.
It’s tens of millions of
single-threaded CPU cycles.
Also, you probably can use GPU acceleration for client code. That’s enough time for a 2026 integrated CPU to do tens to hundreds of billions of tensor ops.
And yet, the iOS keyboard (presumably multithreaded and native) cannot reliably echo keystrokes in under a second. I regularly see webpages take multiple seconds to redraw a screen.
It ran at around 30 fps with a 320x200 screen. That's 64,000 pixels per frame, 1,920,000 pixels per second being rendered.
On a 66 Mhz CPU, that means less than 35 clock cycles per pixel, on a CPU architecture where a multiply or add instruction would take multiple clock cycles to complete.
I know DOOM was not a true 3D engine and it took a lot of shortcuts to look the way it did, but that makes it more amazing, not less. The amount of thought to go into it is just mind-boggling to me.
> I wonder how much adding a profiler to development flows would help modern apps.
Very much, but ideally you want telemetry on the user's device (assuming desktop app). Or your "optimization" might come back as a regression on the Snapdragons you didn't test on.
I literally just built a house for ~$60k a couple years ago. A burned out trailer even in a rural shithole with no jobs in my state is about $100k+. An actual functional house, $250k+. This is counter-intuitive but it makes sense in context of the recent COVID 0 real interest mania.
Meanwhile all the shithole land with no "dwelling" on it was never eligible for mortgages so people weren't able to bid it up to oblivion on debt that they locked in with 30 year mortgages so you get weird results like the cost of vacant land is way cheaper than the same piece of land with a house that can really only be bulldozed (latter would be cheaper in most times in history). End result is I built an entire house on property cheaper than a burned out uninhabitable trailer. Building on unmortgagable land is a way to bypass the fact houses are all locked up in 30 year loans at negative real interest rates.
End result is it's far cheaper to build a house than buy even a shitty burned out one because to do the latter you have to buy someone out of their money printing machine of a negative real rate loan, which obviously they are only willing to do for a king's ransom.
------ re: location ---------
I won't share my address but if you are looking to do this yourself: look up fishing canneries in Alaska, most of them are close enough to cheap plots you could do this on, often even without permits or property tax. These canneries are also usually desperate for workers and pay a livable wage to those with refrigeration technology certifications.
I'd like to see what kind of house you built for 60k. My assumption is its some small maybe 300 sqft box with no sewer you spent many hours yourself building. Not something something most people would do and while cheap in dollars certainly isn't affordable if you are putting a lot of work into it.
It's basically looks like a glorified rectangular shed but it does have sewage, electric, water, and hvac. Not very impressive but every single person I've had over who's lived in an apartment has expressed interest in learning how to do the same over paying rent out the ass for a similarly sized uninspiring shit-box and ending up with no equity.
The only people that have been over that have been unimpressed are people already living in an actual house, but that's not really the target audience for this kind of thing.
---- re: below [my account is throttled] ------
I speculated on an old well share that turned out to be good, so got a well for basically nothing. If you don't have such luck you can haul water.
I use septic, which in some counties (mine) no requirement you be licensed to build. It can be built with only a shovel and some pipes and concrete if you are on an extreme budget, although helps a lot more if you can get ahold of an excavator.
If I understood you correctly, land that has no building on it is not eligible for a mortgage, and is therefore cheaper to buy? With the downside that you have to pay cash, because you can't get a mortgage either?
Yes the land value is so insanely cheaper on un-mortgagable properties in my state, it's off the charts.
I have developed land in my county so I'm familiar with the costs to develop, buy land, place utilities etc. (I did not become a land developer on purpose, only because I realized this absolutely crazy arbitrage)
It would cost you about $200-$250k to buy a rural small acreage land with a manufactured home on it. If you pay cash for the land and drop the exact same manufactured home on it, it would only cost you about $150k, and you would get a brand new house instead of a "used" one.
There is huge pent up demand for someone to just buy a huge swath of small acreage properties and just drop the cheapest manufactured home you could on it as the non-luxury starter home market is currently not being met. You could pretty much double your money. I'm not sure why this isn't being done en masse although a few private actors seem to be doing it and making a killing.
This is not true. Building all types of housing increases the supply of affordable housing.
Build a new luxury apartment, and someone moves from a mid tier apartment into it, and someone moves from an affordable apartment into that, and so on.
Price is a function of constrained supply. The type of supply is not important to increase the numbers.
Yeah, that is big one. Perhaps we need to rethink housing. Shared restrooms, bathrooms, and kitchens. Keep the same restrooms on every floor. A gym with individual showers and a food court on specific floors.
> What will distinguish these structures from slums in 10 or 20 years?
The neighbourhood. When I moved to New York in my twenties, I had roommates. Everyone had roommates. That meant sharing a bathroom and kitchen. Not only did this breed camaraderie and teach me to not be a dick, it also freed up cash so I could enjoy the city and save.
Genuine question, who would actually want to share an intimate space like a kitchen or bathroom with dozens of other strangers on a daily basis? This is obviously a common setup in college dorms or prison, but that is specifically because it’s a temporary (and extreme) cost saving measure, or because you’ve lost the right to participate in society (i.e. prison, which is viewed by some societies to be cruel and inhumane). I lived with housemates for many years to save money and afford housing, but I could at least choose the few housemates with whom I shared those spaces.
I honestly do not know hoping that someone smarter than me figures it out. I suppose it will depend on the execution. If it is comfortable, looks nice, if it creates community, amenities, price, location, etc.
Except this goes against American individualism on every front. Americans really only fit one sort of mold in terms of what they want: single family home, owned outright (usually mortgaged though). You can extrapolate that out to cities as well: young urban professionals pine for polished condos or lofts with nice views and located in trendy neighborhoods, but their "unit" is still theirs totally, with no shared primary amenities (by that I mean kitchens and bathrooms, not features like pools or gyms).
It just so happens that America is luckily predisposed to this kind of living, with an abundance of space to accommodate lots of people in their own non-shared living spaces. The problem with that though is that you limit the opportunities for business, because space is cheap, so you have to implement regulations and zoning to create opportunities for moneymaking and before you know it you can't actually build housing anymore, despite the abundance of space sitting right there.
American cities were replete with dorm room style housing. These were especially popular with new migrants to the city.
An incredibly large percentage of apartments in cities like NYC are used as multi family housing with several housemates sharing them to save on rent.
The reality is that the reason such housing doesn’t exist/isn’t more widespread is because cities have passed laws eliminating them. Before the white flight to the suburbs, the attempt was to keep the poor out of cities where the rich lived by eliminating housing of this sort since the poor couldn’t afford single family housing.
This led to a proliferation of laws that required bathrooms and kitchens in every unit, etc.
It's easy to live in shareable spaces when you're young and unattached - it becomes a lot more difficult as you age and want to grow a family. I'm not sure I want the kind of life where I have to share a kitchen or a bathroom, spaces I consider very private, with people I'm not related to. Maybe this is a uniquely midwestern/American sentiment, I'm not sure. But I am confident that there are more people like me than there aren't. The picture of the American dream is familiar, it's a house with a car in the driveway. I feel that may just be who we are now, regardless of any way we used to be.
Edit cause I had more thoughts:
Honestly, probably one of the biggest mistakes we've made as a country have been not putting up enough resistance to RTO. The single family home is, I believe, probably one of the nicest standards of living in the world. Plenty of space for hobbies and activities, privacy, usually some community among neighbors. The only problem is that it's hard to square the circle when it comes to single family living and living close to an economic hub. To afford this standard you have to live close enough to a hub that you can afford one of the well-paying jobs that exist there, but not so far that your commute significantly eats into your life. With RTO, I think we lost a pretty good opportunity to weaken our dependency on the geographic economic hub. We could have had a diaspora of knowledge workers which gave people the opportunity to pursue a better life at a lower cost, and we sorta just threw all of that away.
Isn't it also the fact the almost no one wants to live like that? The expectations has changed and there's probably little demand for such type of housing.
Yes and there is fierce competition for that in many larger cities, with sky-high prices to rent out a room. But they can't be offered at scale commercially because you'll never get the permits, and the only reason why you can rent these is usually because they're either operating completely under the table or via some carveouts that let property owner rent to 1 or 2 persons.
The pent up demand for this is obvious to anyone who's tried to secure a room only to have a gazillion people competing with them to pay $1000+ to rent an oversized closet to sleep in.
Studio apartments seem like a better option. Also, from a property manager’s perspective, you generally want to minimize shared spaces because they’re a pain and annoying to deal with.
I absolutely disagree. Renting a room in a single family home vastly limits the number of people you have to share those intimate spaces like a kitchen or bathroom with. You also get the option to interview and pick who you’re sharing those spaces with. I lived with housemates for many years, and in dorms during university, and dorms are not even remotely the same from a social safety and privacy perspective.
When the choice is between $3000/mo for a proper apartment and $2000 for a flophouse room some people will take the flophouse. Right now the only choice we offer those priced out is a painfully long commute (with has its own time and car expenses that reduce the savings).
Commercial buildings can't be easily converted into housing that provides the same return. Once the current owners have gone out of business it'll be profitable to turn them into flats.
The arguments against conversion assume you care about the current owner's financial situation.
It's not so much the owner's financial situation, but rather that it'd be cheaper to build new homes than to retrofit a ten floor+ building's plumbing.
You'd also have to install a bunch of showers, which could be a significant problem on its own.
And then there's the increased amount of sewage, which the building might not be able to handle - even the local sewers might not be equipped to handle the uh... Load a large commercial building would generate with 24/7 occupancy vs 8/5 occupancy.
The reason you don't see folks converting commercial spaces into residential isn't because it's not wildly profitable, but because building new purpose-built residential buildings would be cheaper than a conversion for anything other than one or two floors.
> You'd also have to install a bunch of showers, which could be a significant problem on its own.
Compared to installing a new domestic water pipe riser and drains in an office tower (plus pumps, pressure tanks, etc), installing a shower in each unit is essentially free.
Connect the in-unit supply lines to the tap, core drill a hole in the floor to get to the floor below and connect to the drain piping, done.
You can also just raise the shower and toilet on a platform, plumb the waste directly to the wall while depending on the platform to buy you some vertical slope on the way out, and drill right through the side of the building and run the waste vertically down the side. It's not going to freeze on the way down unless you're in Yakutsk. The supply lines you might not even have to retrofit, just put a pressure tank on each floor for peak loads that are slowly topped off by the undersized supply lines.
More toilets, more sinks, plus laundry. A floor of apartments will use far more water than a floor used by an office tenant. A 20k sqft office tower floor might have 6-8 toilets and 6-8 sinks that see light usage for 40 hours a week.
> Commercial buildings can't be easily converted into housing - notably, plumbing is not designed for smaller units and can't be retrofitted
I still haven't seen numbers that show this is a physics problem versus zoning problem. Worst case, make some things (e.g. washers and dryers, maybe even showers) communal.
Having lived in some midwestern cities with a bunch of extra "not easily converted" warehouses, I've seen lots of "illegal" art collective operations where they just put all the extra plumbing on the ground floor where it is easily retrofitted (you can even raise the floor with a false bottom for plumbing if no other option) and then everyone shares a big kitchen, then the upper floors where retro fitting is more difficult are for habitation. Maybe the HVAC unit is undersized or something, or some other safety factors are substandard due to these people not having the money to improve it further, who gives a shit it is better and safer than living on the streets.
Obviously since it's illegal these aren't advertised but they're quite prevalent, and issues are rare enough that now decade past muh Ghost Ship Warehouse is the constant drum being beat by the brain dead building code worshippers who actually bought the line of bullshit that having people homeless and freezing and shitting in the streets was actually a 'written in blood' advantage.
> Does cp actually work on live sqlite files? I wouldn’t expect it to, since cp does not create a crash-consistent snapshot.
cp "works" but it has a very strong possibility of creating a corrupt copy (the more active the db, the higher the chance of corruption). Anyone using "cp" for that purpose does not have a reliable backup.
sqlite3_rsync and SQLite's "vacuum into" exist to safely create backups of live databases.
The containers would need to use a path on a shared FS to setup the SHM handle, and, even then, this sounds like the sort of thing you could probably break via arcane misconfiguration.
> The wal-index is implemented using an ordinary file that is mmapped for robustness. Early (pre-release) implementations of WAL mode stored the wal-index in volatile shared-memory, such as files created in /dev/shm on Linux or /tmp on other unix systems. The problem with that approach is that processes with a different root directory (changed via chroot) will see different files and hence use different shared memory areas, leading to database corruption. Other methods for creating nameless shared memory blocks are not portable across the various flavors of unix. And we could not find any method to create nameless shared memory blocks on windows. The only way we have found to guarantee that all processes accessing the same database file use the same shared memory is to create the shared memory by mmapping a file in the same directory as the database itself.
Pausing requests then running two sqlites momentarily probably won’t prevent corruption. It might make it less likely and harder to catch in testing.
The easiest approach is to kill sqlite, then start the new one. I’d use a unix lockfile as a last-resort mechanism (assuming the container environment doesn’t somehow break those).
If you are not ready to lock yourself in a bunker after reading the article and watching that short, I strongly suggest you consider the inclined plane.
You’d better do it now. Very few locks work in the absence of transformers, springs and inclined planes.
Does anyone have numbers for churn vs. cumulative code?
Most of my commits (hand written and AI) have delete counts that are 75-110% the added line count.
The point that many developers will probably forget to tell the LLM to run cleanup/refactoring paths is probably true though. (I’ve definitely found ghost-chasing bugfixes in all sorts of corners of LLM generated code).
Yeah /simplify is your friend. That and constrained prompts - “refactor x for simplicity - resulting diff must remove n lines of code. Dont change tests. “
Since the list of extensions they query targets certain religious groups and medical conditions, it's almost certainly in violation of US federal employment and hiring law.
Those being in the list doesn't mean that's what they're looking for. Take a look at the database of extensions, there's far more extensions that don't seem limited to any particular group. The author just called those out specifically because they're perfect for implying nefarious intent.
It does suggest that’s what they’re collecting. That is per se a violation in many jurisdictions. It should trigger investigations in most others to ensure it wasn’t mis-used.
The claim I replied to is “They try to profile for things like political beliefs”.
I wasn’t contesting that they query extensions that can be used for that purpose, or that they use query results for that purpose, but indicated that the fact that they make such queries doesn’t necessarily imply that they try to do such profiling.
Observations:
4.6 had previously failed to the point where I had to wipe context. It must have written memories because it was referring to the previous conversation.
As the article points out, 4.6 went out of its way to be lazy and came up with an unusable plan. It did extra planning to avoid renaming files (the toplevel task description involves reorganizing directories of files).
4.6 took twice as long to respond as 4.5.
I’m treating this as a model regression. 4.6 is borderline unusable. I’ve hit all the issues the article describes.
Also, there needs to be an obvious way to disable memory or something. The current UX is terrible, since once an error or incorrect refusal propagates, there is no obvious recovery path.
Anyway, with think set to high, I see drastically different behavior: much slower and much worse output from 4.6.
reply