drooby

This article is about silicon. But the other shortage that matters is meat-compute.

AI is trained on human intelligence. The hyper-scalers are squeezing every last drop of automatically verified reward, and that may get us very far. A compiler passes or fails in milliseconds for free, forever.

But.. "good design taste" has no compiler.

Of course, taste isn't unverifiable. But it's is expensively verifiable. Noisy, slow, and orders of magnitude lower throughput. People with deep domain knowledge often can't articulate well _why_ one design works and the other doesn't. So, judgment arrives as a verdict, and not a crisp rationale. I guess we'll see if sample efficiency outpaces the cost of human judgement.

In the mean time, leverage will sit with whoever holds this tacit knowledge (incumbents). I.e., hospital systems, law firms, chip designers, studios, SaaS that are dominating their niche.. and not with the labs training on it. To me, this is why valuations of companies like Palantir could potentially make sense.

Wouldn't "2-3 weeks on a workday" be a default hypothesis for this kind of thing?

All I want to say is that I absolutely love this essay. Thank you.

Hmm idk, this looks a bit more like serendipity for vitriolic trolls

Well, we take from the artist a motivation for a buyer to purchase their work.

There was a touch of hyperbole ;) we live in the Information Age after all.. but to answer your question,

Article I, Section 8, Clause 8 of the US Constitution

Which empowered Congress to "promote the progress of science and useful arts, by securing for limited times to authors and inventors the exclusive right to their respective writings and discoveries."

Scientists and the artists and their "exclusive rights" have built quite a lot over the centuries.

I mean.. this claim is just untrue. "Owning" something is a social construct defined by law. Our entire society exists because we own things we cannot hold, that is, intellectual property.

What this post is actually pointing out is that intellectual property that has transferrable physical representation has more value to the consumer.

And intellectual property that does not have transferable physical representation has more value to the producer.

Reselling or gifting a book you've read to a friend is wholesome.. it feels good. Truly.. but every time we do that we also take from the artist.

Should it also be their decision that they can gamble? Smoke cigarettes? Get a job? Have sex?

We draw the line somewhere because these things that "are the parents' decision" have consequences on broader society. They have consequences that impact you and me. And we also have a say.

You can make the argument that it's just the parents' decision. But you have to say why.

I put a little too much weight on "the".. sorry..

Reality is that was A bottleneck. Code review has historically been faster than writing the code.

That is no longer true for me. I can complete two to three PRs per day in a span of time that would have historically taken one to three days.

I now sit around doing code reviews and asking for code reviews.

Writing. Code. Is. No. Longer. The. Bottleneck.

Deciding what to build. Reviewing Code. And testing code. Are the new bottleneck.

So of course we don't see massive productivity gains. Because these parts of the SCLC were always bottlenecked but their capacity matched the throughout. We fired all the dedicated QAs years ago. Sr+ engineers that do all the code review are limited.

Teams have not re-organized to match the new code-input velocity.

Engineers don't want to do QA because it's "beneath them".. and most engineers don't like performing or are not Sr enough to do extensive or high quality code review.

Fable responded to that for me. Im nearly certain that blocking this class of prompt is a mistake of a classifier. No one at Anthropic thinks this kind of prompt should be gated. The classifier is still classifying. The model was released to the public yesterday.

What were the tasks?

What is the flaw in the problem's assumptions?

I get the sense that he is misidentifying the potential locus of consciousness..

In the same way that the sound waves and facial expressions I produce are not conscious, the output json of an LLM is obviously not conscious either.

The locus of consciousness and subjective experience may be in the computer, either at inference time or training time..

Has Chaing solved the hard problem of consciousness? I suspect not.

Lots of humans. Lots of companies. Random chance.

Of course, its not that simple. Some companies probably are great at scouting. Yegge mentioned a few ways in the post. Good internship programs, acquihiring, etc.

No, this is a totally fair response. That was probably the core error in my argument.

Some people, perhaps most, will enjoy AI art. Artists are at risk of displacement too.

But a subset of art appreciators hold firm that the creator is an essential aspect of the work, and I don’t see that conviction fading.

How large this subset is, I don’t know. But they also happen to be the world’s tastemakers and trendsetters. Their beliefs largely shape our aesthetic world, so their influence may only grow.

There is a class of human output that will retain value regardless of AI capability: art and sport. People care about the creator. The source defines the work, the awe, and the emotional response.

But almost all output outside that space is at risk of AI displacement. Corporations are amoral entities that optimize for profit, and they follow the law only as much as they must.

The law is our collective action. We socially construct what we value. We could fight to preserve the 5-day work week doing what machines can do. But.. I’d rather fight for collective ownership of the machines.

I can't tell if this is satire

"Augustine of Hippo was perhaps the greatest Christian philosopher of Antiquity and certainly the one who exerted the deepest and most lasting influence. He is a saint of the Catholic Church, and his authority in theological matters was universally accepted in the Latin Middle Ages and remained, in the Western Christian tradition, virtually uncontested till the nineteenth century."

.....

"...he nevertheless remained convinced that soul is an incorporeal and immortal substance that can, in principle, exist independently of a body"

https://plato.stanford.edu/entries/augustine/#AnthGodSoulSou...

At what point will we create micro tolls that AI must pay?

Foucault is taken seriously because his ideas are politically empowering

I think we’re agreeing. The point of the sleep parallel is exactly that the content doesn’t matter, and it’s the filtering process that does the work. Brains replay noisy, sometimes incoherent patterns during sleep and the value is in how that replay reshapes connection weights, not in whether the replay is accurate. That’s the same principle you’re describing with the steering signal

I.e sleep replays don’t need to replay Tuesday’s meeting accurately. They just need to activate the relevant pathways so that the strong ones fire and the weak ones don’t. The pattern of what fires versus what doesn’t is the signal. The “content” of the dream is basically irrelevant.

Fascinating...

This feels eerily similar to sleep consolidation or synaptic pruning

GraphRAG is for LLMS... markdown is for humans.. humans that exist in the meantime

A leading theory in neuroscience is that human brains are fundamentally prediction machines too, constantly predicting sensory input, other people’s behavior, the next word in a sentence. “it’s just prediction” isn’t the gotcha you think it is. Prediction and attention turn out to be a surprisingly powerful foundation for intelligence.

The “just a text predictor” framing was fair a couple years ago but hasn’t kept up. Current models can genuinely identify untested edge cases even when coverage is 100%. You're definitely using the latest and greatest models?

The architecture started as next-token prediction, sure, and yes, human judgment is still required, but that judgment is being captured and integrated too. Every time millions of people use these models, their feedback feeds the next round of improvements.

Also, these models don’t need to replace your best engineers to be disruptive. They just need to outcompete the bottom of the bell curve. For a lot of junior-level work, we’re already getting close.

This is a great point. The Netherlands is the cautionary tale of what happens when you don't do what Alaska and Norway did. A massive resource boom without proper public management hollowed out the rest of the economy.

If a handful of companies capture most of the value from AI while it simultaneously displaces workers across every other sector, that's Dutch disease applied to the entire knowledge economy. One sector booms, everything else withers.

Nobody is proposing that AI companies track down individual contributors and pay them based on their specific contribution. That's not how any of this works.

The Alaska Permanent Fund doesn't pay Alaskans based on how much oil was under their particular backyard. It pays a dividend from a public fund because the resource is collectively owned. The mechanism is simple: companies extract a public resource, a percentage goes into a fund, the fund pays out to the public.

And the infrastructure comparison doesn't quite work. The people who built computing hardware and internet infrastructure were paid for that work at market rates by the companies that hired them. The billions of people whose writing, code, images, and knowledge trained these models weren't paid anything for that specific use. That gap is the whole point.

Addition:

Also worth knowing: collective intellectual property already exists. ASCAP and BMI have been doing exactly this for decades. Individual songwriters can't enforce their rights every time their music gets played, so they pool their IP, license it collectively, and distribute the revenue. The problem they solved is almost identical to the training data problem. Each individual contribution is tiny, but the collective value is enormous. Applying this at the scale of the general public would be novel, but the underlying mechanism isn't. The concept works. It just hasn't been applied to training data yet.

That's not snarky at all, that's exactly the point. They did get it both ways.

The comment I was responding to argued that ownership of non-physical things is basically a "polite lie" and that information is just entropy that belongs to whoever can capture it. My point was that the AI companies clearly don't believe that when it applies to them. They patent their architectures, copyright their outputs, sue competitors for IP violations, and lock down their model weights. They fully believe in ownership of non-physical things.

But when it comes to the billions of people whose work they trained on? Suddenly information is free-flowing entropy that belongs to no one.

That's the asymmetry at the heart of this. The rules around IP apparently apply when it protects their profits, but not when it would obligate them to share those profits with the people whose work made them possible. Which is exactly why the public needs to assert a claim now, before that asymmetry gets any more entrenched.