Ashwin Bharambe

Software Plumber variously.

[Ashwin Bharambe @ashwinb]

Every once in a while I can elicit insane insights from Claude (or GPT). And sadly, somewhat more frequently I get volumes of stupidity. And I can't tell what about my own behavior led to it.

[Ashwin Bharambe @ashwinb]

What a wonderful, thoughtful and constructive post. Very rare these days.

[Yishan @yishan]

The US is going to lose the AI race to China because of our cultural polarization.

You imagine that because the US leads in frontier models that all the US has to do is reach AGI first, and then the exponential curve will be higher, and some Patriotic US AI-bot will materialize and smite America's enemies.

For anyone who doesn't believe in the "fast take-off doomer" scenario, it should be self-evident that this is not how it happens.

The way the AI victory is won is that AI gets progressively more powerful, and is gradually integrated into every part of society and industry, providing the real massive economic uplift that leads to global domination.

When it comes to facillitating that, China is far ahead of the US, not only in terms of energy infrastructure and hardware ecosystem, but also because its culture broadly accepts and is eager to adopt more AI into their lives. This is a critical US weakness.

This is entirely self-inflicted, and tech needs to take responsibility for it and fix it.

Most people in tech prefer to sneer and look down on the anti-AI people, like coastal elites looking down on flyover country residents, and they are making the same mistake: failing (or refusing) to see that there are real and valid grievances, and they need to be solved.

I am going to describe two big ones that the tech world needs to wake up to and solve:

1) Datacenters do cause water and electrical shortages

I was part of the crowd that sighed condescendingly about the poorly-informed "datacenters use a lot of water" criticisms. I'm a water guy (see my other work), I get it.

It's true that datacenters are not likely to cause a global freshwater shortage. It's not a planet-scale problem (vs the way fossil fuel usage is). Datacenters don't "use" that much water, in a planetary sense of the word.

But datacenters absolutely do cause regional water shortages. Water is not perfectly fungible: you need to bring it to places. So every place has a certain amount, and a certain inflow rate. Datacenters use up a lot of it, and this makes water scarce, and raises local utility prices.

Many of the areas where datacenters are sited had marginal utility infrastructure to begin with, and a sudden new consumer will drive up local utility prices. Ordinary people see this directly on their utility bills!

They are also noisy and in the instances where generators have been brought into provide immediate power, they pollute the air.

I used to be able to tell people "No, no, datacenters are among the greenest users of electricity, because they're often powered by huge solar and other renewable buildouts!" but Elon, the great electric-car / solarcity entrepreneur, made it impossible by hyperscaling his datacenter build-out with gas generators. 15-20 years ago, everyone regarded Elon as a tech climate hero (I know, hard for the young people today to believe), but that reputation is never coming back. Yeah, he built his datacenter in 100 days or whatever, record speed, but the local population pays the cost.

You can't deny these experiences! They make life harder and more expensive! And then these people tell their friends and family elsewhere, and their friends and family listen.

There are solutions available to us, but I will get to them at the end, after I outline the second problem - because in the meantime all those people look on the internet and they think "for this??"

2) AI is flooding all of our online spaces with slop

You probably think, "that's just dumb people using it, AI does a lot of amazing things!"

And you'd be right, it does. It does genuinely astounding and helpful things. But that does not solve the slop problem.

If I were served a perfectly marbled ribeye steak on a platter that also had literal shit on it, it wouldn't matter how good the steak was. That is what is happening to the internet. The Good Works of AI are being drowned out by the slop we allow to surround it, to say nothing of crowding out the quality human content we initially came for.

Social media, online forums, and discussion boards used to be fun. They were a way for people to connect with other people, in many cases people you wouldn't otherwise have ever met. Now that is impossible, because you have to wade through slop.

Ordinary people on the internet experience this! It makes every tech product less fun! It's why there is less and less enthusiasm for every new tech launch that happens.

You might think this problem will solve itself as AI gets better. I don't agree.

The slop is produced by using low-cost models. No one uses frontier models to produce their marketing slop. The spammer needs to minimize costs, so they are always going to use the cheapest possible model that can output something passable. That means every spammer is always going to use Qwen3-27B for all their spam no matter how good the frontier models get if we are lucky. And we cannot just auto-filter it out using more expensive frontier models because then spammers are asymmetrically using a cheap model to force usage of a more expensive model. The arms race cannot be won that way: slop will always be created by low-quality models.

The everyday experience of the median internet user (a millennial who grew up in the halcyon days of Facebook) is one where everything they enjoyed about the internet has been overrun by the worst possible thing - not polarized trolling or disinfo - no, those would be far preferable - but relentlessly and aggressively stupid content.

There are solutions to this that tech can implement but before I get to them, I will say that they are not hard solutions (not compared to the things we are building, like rockets to Mars and supersonic civilian jetliners), but the hurdle is an insular tech culture that refuses to acknowledge these as problems.

An industry that consistently does things that makes the lives of ordinary citizens worse and polarizes a large segment of the voting public against them will not succeed on a global level against a competitor that is supported broadly by its people. This is an American self-own that doesn't need to be happening.

====

Solutions:

1: Datacenters need to be built holistically to benefit and ameliorate the costs they impose on surrounding communities.

Neither water nor electricity need to be scarce. New water supplies can be built, and new solar (cheaper than ever) can be installed. The mad rush to do everything as quickly as possible is unnecessary. Because the water and electricity footprint of a datacenter is often so large compared to that of the surrounding community, only a small fraction of overbuilding would be necessary to generate a surplus of water and electricity. Build that fraction out first, slashing the utility costs in the surrounding region, and then build the datacenter.

The technology for emissions trapping, carbon capture and storage (CCS) for natural gas plants has existed for nearly a decade now. Natural gas generators used to power datacenters can be outfitted so that they don't end up dumping loads of pollutants into the surrounding air. This is a tech implementation problem, and the demand should be huge!

As for noise, why aren't we just constructing huge sound baffles around the datacenter? (Really, just the generators). The same arguments about "it's not that loud" cut both ways: if it's not a huge amount of noise, it can be muffled. This is a tech implementation problem!

Datacenters could be infrastructure delivery vehicles for water, electricity, and even internet. Their huge data interconnects could easily serve high-speed traffic for their local areas, providing free wifi in all public spaces. All of this is possible, and reducing all of these costs for local residents would spur economic growth the same way a tax break would. Instead, right now, we are doing the opposite: causing regional shortages, increasing ordinary peoples' cost of living, and stunting economic growth when we need it the most.

2: Control AI slop with this one weird trick

Require all AI content to be clearly labeled as such, and demand that all online platforms enforce it.

This will solve the problem.

The common objection is "that would be impossible." No, it wouldn't. Sayiung "someone would still try to do it and it'd be an arms race" is wrong. It's not untrue, it's that it doesn't cause the solution to fail.

Plenty of content is already banned or otherwise regulated, and compliance comes from enforcing the actions of large-scale actors, and pushing the implementation down to individual platforms. Once a behavior becomes the norm, most people will follow it and explicit enforcement only needs to occur at the margins. It's enough to improve the online experience for most people, which is what we want.

Once it's labeled, people and platforms can decide if they want to allow AI content or not allow it. It creates the most powerful market incentive: individual choice.

Right now, no one has the choice. This gives everyone the choice of how much AI content they see. Once users can choose, this creates the conditions that incentivize production of higher-quality content. Right now everyone is equally forced to endure everything, so no incentive to "pay more to produce better" exists. If the stuff on the other side of the slop filter eventually becomes good or desirable, we'll actually win.

Far-seeing much-maligned Sam Altman foresaw this is 2019, that the world would need a way to verify true human content from AI content. His solution was just unfortunately dystopian: an orb that scans everyone's eyeball. Something lower-tech and more socially-oriented has already been shown to work better:

China already has such AI regulations. It's the greatest irony that after all the "we mustn't regulate ourselves, China will win" justification, China ended up implementing comprehensive AI regulation - and this is the important part for Americans to understand: not because the Communist Party wanted control over AI, but because it was in the public interest to do so - with the result that AI products are being much more readily accepted and adopted across China in both society and industry.

====

These solutions have two big advantages:

1) They play to tech's strengths: build infrastructure, come up with clever solutions that make things that suck, suck less.

2) They can progress and yield benefit in piecemeal fashion: every datacenter whose plan lowers utility costs in the surrounding area and provides free digital access will be more readily accepted, and every online platform that implements an "AI labeling rule" and enforces it well will eventually trend towards increasing quality, decreasing slop, and more user success even if others don't.

US tech can start doing these things now, or it can believe that all the AI-hate is unfounded and misinformed, and not worth addressing. If it does so, go ahead and bookmark my prediction now that China wins the AI race.

[Ashwin Bharambe @ashwinb]

Ban this flopper

[BrickCenter @BrickCenter_]

SGA’s signature move 🔥

[Ashwin Bharambe @ashwinb]

@ankrgyl In what ways?

[Ashwin Bharambe @ashwinb]

@radoshi Huge congratulations, Rushabh!

[Eric S. Raymond @esrtweet]

Is it weird that AI coding assistance is not giving me identity fracture?

A lot of software developers are feeling disoriented and threatened these days. Programming by hand is clearly going the way of the buggy whip and the hand-cranked auger. Which is how we're finding out that a lot of people have their identities bound up in being good at hand-coding and how it feels to do that.

That's not me. It's not me at all. Rather to my surprise, I don't miss coding by hand, not any more than I missed writing assembler when compilers ate the world and made that unnecessary. (That was in a couple years back around 1983, for you youngsters.)

Maybe the fact that I'm not feeling any of this disorientation disqualifies me from having anything to say to people who are. On the other hand...if you can learn to emulate my mental stance and be completely unbothered, maybe that would be a good thing?

So. If you're a programmer, and you're feeling disoriented, try this on for size:

I like being a wizard. I like being able to speak spells, to weave complex patterns of logic that make things happen in the world. Writing code is a way to manifest my will.

Yes, I've piled up a lot of arcane knowledge over the 50 years I've been doing this. But languages of invocation, they come and they go. Been a long time since I've had any use for being able to program in 8086 assembler, and that's okay. I have better spells now, and these days some rather powerful familiars.

What I'm inviting you to do is think of yourself as a wizard. Not as a person who writes code, but as a person who is good at assuming the kind of mental states required to bend reality with the application of spells.

And if that's who you are, does it matter if the spells are painstakingly scribed in runes of power, versus being spoken to an obedient machine spirit?

It's all one; it's all the manifestation of will. Arcane languages come and go, machine spirits appear and then diminish to be replaced by more powerful ones, but you? You are the magic-wielder. Without you, none of it happens.

Same as it ever was. Same is it ever was. And so mote it be.

[Ashwin Bharambe @ashwinb]

The things you think about determine the quality of your mind. Your soul takes the color of your thoughts.

- Marcus Aurelius

[blue @bluewmist]

i like luxury things like:
resting without guilt
slow mornings
watching sunsets
being present

[Ashwin Bharambe @ashwinb]

@Vjeux Yuge!

[Ashwin Bharambe @ashwinb]

Why is no big lab building a competitor to Cursor Tab?

[Ashwin Bharambe @ashwinb]

@charlieholtz @conductor_build I can’t get Claude to work because the Claude binary is an org private wrapper which injects its own separate auth etc. What’s Conductor trying to do which makes this not work? Does it have its own build of Claude?

[Ashwin Bharambe @ashwinb]

Cursor Tab has become so unbelievably bad. This is what happens when you try to write "cool" blog posts about online RL. You lose all taste.

[Patrick Collison @patrickc]

Maybe a very prosaic observation, but I've been reflecting on just how much the pandemic changed the world in ways that are completely unrelated to the pandemic itself. I think I've underestimated it 'till now.

In a recent interview, I was struck by the comment that so many of the shops that we associate with the best of France—the poissonneries and the fromageries—closed during the pandemic, to be replaced by take-out pizza shops and the like.

College professors almost uniformly describe big changes in student behavior: lecture attendance and willingness of students to complete reading assignments are both way down.

A UK government official recently told me that British economic statistics have become much less reliable since the pandemic: data on trade, employment, and population is suspect. (The true GDP per capita figures are probably worse than what is indicated by the published data, since the 2021 census is believed to be an undercount.)

In the West, there are far fewer bustling workplaces than there used to be. In recent conversation with a well-traveled friend, he bemoaned how so many cities—places like Madrid, Buenos Aires, and Bali—have lost so much of their erstwhile vibrant nightlife.

Immigration accelerated enormously across many countries, including the US, the UK, Canada, and Australia.

In China, I hear descriptions of how fear, caution, and conservatism have persisted since the COVID lockdowns. (And Western travel to China remains massively depressed.)

Lots of the changes are neutral, or even good. Retail participation in the US stock market almost doubled overnight, say, and has persisted at that elevated rate. Firm creation in the US increased by around 50%, which is probably a very good thing.

Overall, the number of time series (either literal or figurative) that jumped discontinuously during COVID and then didn’t return to baseline is just very striking.

Which are the best historical analogs? Are there any apart from major wars?

I want to read this book!

[Ashwin Bharambe @ashwinb]

No AGI is coming. These models be dumb as F*

[Ashwin Bharambe @ashwinb]

typing on Cursor is so slow. ugh what is eating all the CPU? first principle of an editor is to render the character ASAP.

[Ashwin Bharambe @ashwinb]

gpt-5-codex: main brain
claude sonnet: main executor, reasonably sharp but super fast
cursor-agent: F** that it is terrible ugh, so much regression
cursor-tab: best if you actually code but so rare now

[Ashwin Bharambe @ashwinb]

and terminal UIs are beating the crap out of IDEs. maybe cursor could release a neovim integration for cursor-tab.

[Ashwin Bharambe @ashwinb]

Claude Code is so pretty I love using it even though it can be _so_ dumb a lot of the times.

[Ashwin Bharambe @ashwinb]

@sawyerhood why is that? faster but too sloppy in thought?

[The Humanoid Hub @TheHumanoidHub]

This slapstick sequence happened in the 4x100m relay prelims at the World Humanoid Robot Games.