Biggest question I have is maybe... just maybe... LLM's would have had sufficient intelligence to handle micropayments. Maybe we might not have gone down the mass advertising "you are the product" path?
Like, somehow I could tell my agent that I have a $20 a month budget for entertainment and a $50 a month budget for news, and it would just figure out how to negotiate with the nytimes and netflix and spotify (or what would have been their equivalent), which is fine. But would also be able to negotiate with an individual band who wants to directly sell their music, or a indie game that does not want to pay the Steam tax.
I don't know, just a "histories that might have been" thought.
Love it, we can finally make the libertarian paradise of a patchwork of private roads possible by having your agent negotiate a path to where you want to go and make the appropriate micro payments.
Is that 415:1 the rate of return of an audit, or the expense:revenue ratio of the IRS as a whole? I remember hearing some time ago that the expense ratio was 11% for the IRS? But 415:1 is way way less than 11%.
Captured revenue : cost to capture (could be an audit, billing for interest/fees due, etc. lots of avenues to capture revenue that is being missed).
The problem is these metrics aren't really scalable productivity metrics. If you doubled cost, it might go to 100:1, if you tripled cost, it might go to 0.5:1
Each dollar generally gets more expensive to capture.
A key point is that there are large indirect costs that scale up rapidly that are not accounted for in these direct costs. These costs show up on the balance sheet somewhere else in the government, which makes the ROI for the auditors look much better than it actually is.
This is well-understood by the Federal government. When they set their targets they fully account for the growth of indirect costs created by the audit activity that don’t show up in the ratio.
Good point, and kind of interesting in that as we keep cutting funding to the IRS, this ratio will probably get wider (which looks good, but is actually bad for what it implies).
I also worked at Google (on the original Gemini, when it was still Bard internally) and my experience largely mirrors this. My finding is that Gemini is pretty great for factual information and also it is the only one that I can reliably (even with the video camera) take a picture of a bird and have it tell me what the bird is. But it is just pretty bad as a model to help with development, myself and everyone I know uses Claude. The benchmarks are always really close, but my experience is that it does not translate to real world (mostly coding) task.
Gemini interesting with Google software gives me the best feature of all LLMs. When I receive a invite for an event, I screenshot it, share with Gemini app and say: add to my Calendar.
Yeah, as evidenced by the birds (above), I think it is probably the best vision model at this time. That is a good idea, I should also use it for business cards as well I guess.
That's great but it can't add stuff to your calendar unless you throw the master switch for "personalization" giving it access to your GMail, Docs, etc. I tried that and it went off the rails immediately, started yapping in an unrelated context about the 2002 Dodge Ram that I own, which of course I do not own, but some imbecile who habitually uses my email address once ordered parts for one. I found that to be a pretty bad feature so I had to turn it off, and now it can't do the other stuff like make calendars or add my recipes to Keep.
Gemini is pretty hit-or-miss with tool calls. Even when I explicitly ask for a code block, it tends to break the formatting and spill the text everywhere.
I don't know ... as of now I am literally instructing it to solve the chained expression computation problem which incurs a lot of temporary variables, of which some can be elided by the compiler and some cannot. Think linear algebra expressions which yield a lot of intermediate computations for which you don't want to create a temporary. This is production code and not an easy problem.
And yet it happily told me what I exactly wanted it to tell me - rewrite the goddamn thing using the (C++) expression templates. And voila, it took "it" 10 minutes to spit out the high-quality code that works.
My biggest gripe for now with Gemini is that Antigravity seems to be written by the model and I am experiencing more hiccups than I would like to, sometimes it's just stuck.
People's objections are not the quality of code or analysis that Gemini produces. It's that it's inept at doing things like editing pieces of files or running various tools.
As an ex-Googler part of me wonders if this has to do with the very ... bespoke ... nature of the developer tooling inside Google. Though it would be crazy for them to be training on that.
Can't argue with that, I'll move my Bayesian's a little in your direction. With that said, are most other models able to do this? Also, did it write the solution itself or use a library like Eigen?
I have noticed that LLM's seem surprisingly good at translating from one (programming) language to another... I wonder if transforming a generic mathematical expression into an expression template is a similar sort of problem to them? No idea honestly.
It wrote a solution by itself, from the scratch, with dozens of little type traits, just as I would do. Really clean code. And the problem at hand is not the mathematical, linear algebra one. I gave that example just for easier understanding of the problem at hand. The problem is actually about the high-performance serialization. Finally, I instructed it to build complex test cases with multiple levels of nested computations to really check whether we are making any copies or not. Did it in a breeze.
Not sure about the other models. I'd guess that Claude would do equally good but I don't have the subscription for other models so I can't really compare. I for sure know that the ones from the free-tier are not worth spending time with for tasks like this. I use them mostly for one-shot questions.
So yeah, I think I have a pretty good experience. Not perfect definitely but still looks like a SF to me. Even to a highly trained C++ expert it would take probably like a day to build something like this. And most C++ folks wouldn't even know how to build this.
I think this is neat. I use org-mode for pretty much everything, which has all of these features I think, but sometimes there is nothing more motivating than a quick responsive UI to actually do something. This looks motivational.
My only pushback is using sqlite. I am a big fan of just using simple (structured) text files that can be edited by hand when needed. Your computer is more than capable of doing all the joining/querying/aggregating/whatever with the text file itself rather than relying on a database. I personally find these sort of file structures comforting as it means they can be easily modified in unsupported ways.
Really seems to me that there should be no exemption for land tax for non profits or religious reasons. It is just far too subject to abuse, and it means that we have large churches in the middle of incredibly dense cities that pay almost nothing in taxes.
The issue is that, most of the time, "incredibly dense cities" are not the places where this is hitting the hardest. It's the smaller towns where the impact of hospital rollups hits hardest on the property tax rolls.
Problem is, of course, that if we don't get one of the hospitals in, say, Houston, to put a facility in, say, Nacogdoches, on its books; then that facility may go away entirely. In which case you'd have issues in the market with inequity of access for the very populations who may need that access most. (Elderly and poor.) But if you do allow it, well, you have issues with property tax rises.
So local leaders are put in a position of having to weigh the value of having a hospital or clinic be available locally, against any potential decrease in property tax revenues. Now you hope they get that cost-benefit analysis correct, but there's no guarantee.
In my metro area it irks me to see the churches with large empty parking lots empty most of the week. We have a housing shortage and they seem to have no little incentive to convert their parking to more productive use.
I agree, the whole ruse that these 501s meaningfully does charitable work for our communities is laughable and their tax exemption should be revoked, at least with regard to land taxes.
There are almost no places where a housing shortage is due to a lack of land. Housing shortages have all sorts of reasons, from constructions cost, to zoning, to restrictions on what can be built, but it's virtually never a lack of land.
And parking is a productive use - they have services once a week, and parking means people can come to the service. That's the definition of productive use. Something does not need to be used 24/7 to be productive.
Church goers using parking lots like this is a use, but I doubt it's a productive charitable use that should to be subsidized by localities.
Every other contemporary development in my area that faces real economic reality is ground floor retail, commercial/residential on top, and optionally underground parking.
There are certainly productive religious charitable efforts using facilities like this: homeless shelters, community low-cost/free clinics, soup kitchens. I think these uses should be tax subsidized, but other mystical efforts should not be whether they generate a profit or not.
I think a good reform to the 501c3 system would be to make non-profits like these churches and hospitals classify their actual charitable activity and separate it from their other activity, just like individuals with a mix of personal/small business income/expenses are required to do.
Why should churches get great real estate in central locations but not housing? If people only come to church once a week, surely they can spend the extra time driving further.
The idea is that we give up the land tax revenues in exchange for the services the non-profit provides. (And of course the government does not decide which services are useful or not, the people do.)
One thing I might agree with is land tax for non-profits that charge fees for services, as opposed to those who work off of donations. I think that would fix the issue without destroying non-profits.
Do you think a government should be able to seize property under eminent domain if they believe that selling it to a third party to commercially develop would lead to higher tax revenue?
The government already has and does do exactly this. Is this suppose to be a gotcha? If you have very valuable property, you should pay taxes on it. Claiming that you have ownership over land on this planet is odd, you didn't create the land and governments change overtime.
Property taxes are the most evil of taxes because they force you out onto the street if you're unable to pay them. Qualifying it with the words "very valuable" to solve the problem creates an arbitrary two-tier system that is inherently unfair.
>Claiming that you have ownership over land on this planet is odd, you didn't create the land and governments change overtime.
Property taxes are the most just of all taxes because they are the most correlated with your consumption. Speficially, the land value tax portion of property tax (ideally, that is the whole component).
>The government didn't create the land either.
The government did create the peace and order that allows you to sleep at night on your land without having to worry about another tribe taking your land from you. Without an ability to defend it, "your" land is a tenuous label.
The government, and the rest of society, also pays a hefty price routing utilities, police, ambulances, and people around your property's borders. The more property you have, the more it costs the rest of society, not just in money, but in time.
Earned income taxes are the most evil of all taxes. Why would you have to pay for the act of providing value to society?
There's no such thing as a free lunch. Because it is politicaly unpalatable to tax landowners, we tax economic activity instead.
The result is that return on effort are reduced. That mean labor, entrepreneurs, and capital bear the burden of supporting government budgets as opposed to landowners who benefit from the economic activity making their land valuable.
Taxes as a rule discourages whatever get taxed. The exception to this is land, because land isn't created. It already exists in nature.
Historically speaking, I am not sure if humans argued that they have created the land and therefore they should be allowed to use it. Ownership of the land and its use is, rather, simply tied to one's ability to retain it ( possession being 9/10ths of the law and all that ).
Yes, you are correctly identifying that all land rights stem from one's ability to claim nature's productive power as his own and monopolize all output from it.
This was self-evident in the feudal era, when landlords (Lords) had to at least raise their own militaries to assert this monopoly right. But the modern State and the landlords reached a compromise: the State will provide security to protect the lords' monopoly on nature so long as the landlords don't raise armed forces.
It may be absurd, but do you have a workable framework that can replace it? If not, it makes zero to no difference whether it is absurd or not. It works for the society in place.
Yes. A high land value tax prevents the capture of unearned wealth by owners of land without introducing market inefficiencies or price distortions.
The current arrangement demonstrably does not work for society in place, and as AI (whether in this wave of innovation or the next) increases productivity further, it will work less and less by virtue of further increasing land rents, thereby pricing out larger and larger swaths of society from a place to live, work, or otherwise exist.
Huh? Last time I checked, municipalities big and small fight for every bit of investments they can get and they typically get it by offering a swath of incentives at the cost of the actual taxpayer. That high value land ends up being tax free for the actually wealthy while a schmuck like me get his bill increased and argues with otherwise well-meaning people that akshually high taxes are good for me.
I think that what I am saying is that, in practice, the well-intended solutions like the one that was listed above are effectively nullified as they do not seem to anticipate real world human reactions. What ends up happening is that it is only a subset of the people, who own land that pay higher taxes. The solution is to remove any and all subsidies. Governments of all levels have not exactly proven to be a reliable steward over the past few centuries..
Well it's pretty easy to have a useless conversation if you're going to act as if the words your interlocutor are saying are "effectively nullified."
My solution does anticipate real-world human interactions: don't give rich landowners tax breaks. This is baked into the premise of having a high tax. A tax that is effectively not-high is by definition not a high tax, ergo is not the solution I am proposing. If I proposed a solution of "have a tax that is claimed to be high but actually is not," then your response would be valid. But my solution was: have a high land value tax.
Your solution is dismissible by your same logic. "While removing any and all subsidies is well-intended, in practice real-world human interactions dictate that will not occur."
<< My solution does anticipate real-world human interactions: don't give rich landowners tax breaks. This is baked into the premise of having a high tax.
If it fails to address those now ( because those are already high ), what, exactly makes you think, it will work better if we increase those taxes? If anything, increasing those taxes will become an incentive to find ways to mitigate their impact..
The solution to remove those for everyone across the board, but we can't do that. We can't have an even playing field.
Or property taxes should be eliminated because they are subject to abuse, and instead sales tax should be the primary source of income for all governments.
I guess it would be a consumption-based tax. The usual argument against it is that it's regressive: the poorer people spend higher percentage of their income on consumption, and therefore end up with a higher tax burden relative to their income. This can be counterbalanced by e.g. not taxing groceries/food but it becomes a whackamole tax breaks game quickly---should we also exclude fuel/housing/educational expenses/etc.
Strong disagree. If something has value, then the community should decide to preserve it as a group or the state should preserve it for us. I suspect that most of these schemes are some form of tax avoidance for wealthier people. The idea that some politically connected and likely wealthy group of people need some sort of help "preserving" historic buildings seems... dubious.
What do you think the community deciding to preserve it looks like? The government is the community. It's made out of the community. It's elected by the community. What mechanism are you suggesting?
Has anyone contrasted gas town to Stanford's DSPY (https://dspy.ai/)? They seem related, but I have trouble understanding exactly what Gas Town is and so can't myself do a comparison?
let me take a shot. i have thought about both for a while.
dspy is declarative. you say what you want.
dspy says “if you can say what you want in my format, I will let you extract as much value from current LLMs as possible” with its inference strategies (RLM, COT; “modules”) and optimizers (GEPA).
gas town is …
given a plan, i will wrangle agents to complete the plan. you may specify workflows (protomolecules/molecules) that will be repeatedly executed.
the control flow is good about capturing delegation. the mayor writes plans, and polecats do the work. you could represent gas town as a dspy program in a while loop, where each polecat loops until its hooked work is done. when work is finished, its sent to the merge queue and integrated.
gas town uses mostly ephemeral agents as the units for doing work .
you could in theory write gas town with dspy . the execution layer is just an abstraction . gas town operates on beads as state . you could funnel these beads thru a dspy program as well.
the parallels imo are mostly just structured orchestration .
i hope this comes off as sane. 2026 will be a fun year.
Haha, yes, when read out loud, all the new terms do come off as a bit unhinged. :]
It sounds like the major difference is that DSPY is more of a "define a node in a graph of computation, flow data through those nodes". While Gas Town is ideally more of "Tell me what you want, I will spin up a graph of nodes that you can have some input on to complete your work".
I can't make up my mind if you meant this the other way around? If you meant as stated, I'm genuinely curious why you would rather take and uber/lyft vs a (seemingly much safer and more pleasant) waymo?
One interesting point was the original PageRank algorithm greatly benefited from the fact that we kinda only had "text matching" search before Google (my memory was AltaVista at the time).
Because text matching was so difficult to search with, whenever you went to a site, it would often have a "web of trust" at the bottom where an actual human being had curated a list of other sites that you might like if you liked this site.
So you would often search with keywords (often literals), then find the first site, then recursively explore the web of trust links to find the best site.
My suspicion has always been that Google (PageRank) benefited greatly from the human curated "web of trust" at the bottom of pages. But once Google came out, search was much better, and so human beings stopped creating "web of trust" type things on their site.
I am making the point that Google effectively benefited from the large amount of human labor put into connecting sites via WOT, while simultaneously (inadvertently) destroying the benefit of curating a WOT. This means that by succeeding at what they did, they made it much more difficult for a Google#2 to come around and run the exact same game plan with even the exact same algorithm.
tldr; Google harvested the links that were originally curated by human labor, the incentive to create those links are gone now, so the only remaining "links" between things are now in the Google Index.
Addendum: I asked claude to help me think of a metaphor, and I really liked this one as it is so similar.
```
"The railroad and the wagon trails"
Before railroads, collective human use created and maintained wagon trails through difficult terrain. The railroad company could survey these trails to find optimal routes. Once the railroad exists, the wagon trails fall into disuse and the pathfinding knowledge atrophies. A second railroad can't follow trails that are now overgrown.
```
> I am making the point that Google effectively benefited from the large amount of human labor...
This is exactly right, but the thing most people miss is that Google has been using human intelligence at massive scale even to this day to improve their search results.
Basically, as people search and navigate the results, Google harvests their clicks, hovers, dwell-time and other browsing behavior to extract critical signals that help it "learn" which pages the users actually found useful for the given query. (Overly simplified: click on a link but click back within a minute to go to the next link -> downrank, but spend more time on that link -> uprank.)
This helps it rank results better and improve search overall, which keeps people coming back and excluding competitors. It's like the web of trust again, except it's clicks of trust, and it's only visible to Google and is a never-ending self-reinforcing flywheel!
And if you look at the infrastructure Google has built to harvest this data, it is so much bigger than the massive index! They harvest data through Chrome, ad tracking, Android, Google Analytics, cookies (for which they built Gmail!), YouTube, Maps and so much more.
So to compete with Google Search, you don't need just a massive index, you also need the extensive web infra footprint to harvest user interactions at massive scale, which means the most popular and widely deployed browser, mobile OS, ad tracking, analytics script, email provider, maps, etc, etc.
This also explains why Google spent so many billions in "traffic acquisition costs" (i.e. payments for being the Search default) every year, because that was a direct driver to both, 1) ad revenue, and 2) maintaining its search quality.
This wasn't really a secret, but it (rightfully) turned out to be a major point in the recent Antitrust trial, which is why the proposed remedies (a TFA mentions) include the sharing of search index and "interaction data."
Like, somehow I could tell my agent that I have a $20 a month budget for entertainment and a $50 a month budget for news, and it would just figure out how to negotiate with the nytimes and netflix and spotify (or what would have been their equivalent), which is fine. But would also be able to negotiate with an individual band who wants to directly sell their music, or a indie game that does not want to pay the Steam tax.
I don't know, just a "histories that might have been" thought.
reply