(1): You don't have to be an Ed Zitron disciple to infer that OpenAI and Anthropic are likely overvalued and that Nvidia is selling everyone shovels in a gold rush. AI is a game-changing technology, but a shitty chat interface does not a company make. OpenAI and Anthropic need to recoup astronomical costs used in training these models. Models that are now being distilled[1] and are quickly becoming commoditized. (And frankly, models that were trained by torrenting copyrighted data[2], anyway.) Many have been calling this out for years: the model cannot be your product. And to be clear, OpenAI/Anthropic most definitely know this: that's why they've been aquihiring like crazy, trying to find that one team that will make the thing.

(2): Token prices are significantly subsidized and anyone that does any serious work with AI can tell you this. Go use an almost-SOTA model (a big Deepseek or Qwen model) offered by many bare-metal providers and you'll see what "true" token prices should look like. The end-state here is likely some models running locally and some running in the cloud. But the current state of OpenClaw token-vomit on top of Claude is fiscally untenable (in fact, this is why Anthropic shut it down).

(3): This is typical Dropbox HN snark[3], of which I am also often guilty of. I really don't think AI coding is a killer product and this seems very myopic—engineers are an extreme minority. Imo, the closest we've seen to something revolutionary is OpenClaw, but it's janky, hard to set up, full of vulnerabilities, and you need to buy a separate computer. But there's certainly a spark there. (And that's personally the vertical I'm focusing on.)

[1] https://www.anthropic.com/news/detecting-and-preventing-dist...

[2] https://media.npr.org/assets/artslife/arts/2025/complaint.pd...

[3] https://news.ycombinator.com/item?id=9224

    (1) OpenAI & Anthropic are absolutely cooked; it's obvious they have no moat
    (2) Local/private inference is the future of AI
    (3) There's *still* no killer product yet (so get to work!)

Nagel's point was quite literally the opposite[1] of this, though. We can't understand what it must "be like to be a bat" because their mental model is so fundamentally different than ours. So using all the human language tokens in the world can't get us to truly understand what it's like to be a bat, or a guppy, or whatever. In fact, Nagel's point is arguably even stronger: there's no possible mental mapping between the experience of a bat and the experience of a human.

[1] https://www.sas.upenn.edu/~cavitch/pdf-library/Nagel_Bat.pdf

Why not pay your debts then? I totally understand debt forgiveness for extenuating circumstances (and imo, it's a crime that student loan debt can't be forgiven, and the interest rates are often predatory—especially in the case of med school and law school), but this just sounds like stealing with extra steps.

Public model inference quality is almost at SOTA levels, why would anyone pay these VC-subsidized companies even a cent? For a shitty chat interface? Give me a break.

We have these things called systems: a "system" is anything that follows rules: a board game, traffic, the English language, math, C++, etc. Some systems are smart and they can talk about themselves, but others can't. For example, Tic-Tac-Toe can't talk about Tic-Tac-Toe, but English can talk about English.

Gödel is interested in smart systems because dumb systems are boring.

Some systems are useful: they are "useful" if they always say true things. So math is more useful than English. I can lie in English, but I can't lie in math. (Formally, this is what we call consistency).

So here's a problem for you: suppose we have a smart-useful-system, call it SUS. SUS should be able to say "SUS is useful." It can talk about itself and it can't lie, so we should have no problem, right?

Gödel showed that if our system can actually say that about itself, it wasn't useful to begin with. For a few centuries, philosophers and mathematicians were trying to come up with the "one perfect system": useful, smart, and also complete (it can say all true things), and a few more properties. Turns out such a system is impossible.

NB: I use the words "say" or "talk about" in a very hand-wavy fashion, sometimes I mean Prove(), sometimes I mean Entail(). The details are very nuanced, and this isn't meant to be a deep dive.

[1] https://inference-review.com/article/loebs-theorem-and-curry...

I mean, just like ray-tracing, SDF (ray-marching) is neat, but basically everything useful is expensive or hard to do (collisions, meshes, texturing etc.). I mean mathy stuff is easier (rotations, unions/intersections, function composition, etc.) but 3D is usually used in either modeling software or video games, which care more about the former than they do the latter.

Relocation: Case-by-case

I'm a CTO, expert engineer, and data professional interested in team-building, consulting and architecting data pipelines. At Edmunds.com, I worked on a fairly successful ad-tech product and my team bootstrapped a data pipeline using Spark, Databricks, and microservices built with Java, Python, and Scala.

At ATTN:, I re-built an ETL Kubernetes stack, including data loaders and extractors that handle >10,000 API payload extractions daily. I created SOPs for managing data interoperability with Facebook Marketing, Facebook Graph, Instagram Graph, Google DFP, Salesforce, etc.

More recently, I was the CTO and co-founder of a gaming startup. We raised over $6M and I was in charge of building out a team of over a dozen remote engineers and designers, with a breadth of experience ranging from Citibank, to Goldman Sachs, to Microsoft. I moved on, but retain significant equity and a board seat.

I am also a minority owner of a coffee shop in northern Spain. That I'm a top-tier developer goes without saying. I'm interested in flexing my consulting muscle and can help with best practices, architecture, and hiring.

Would love to connect even if it's just for networking!

Blog: https://ai.dvt.name/whos-david/ (under construction)

GitHub: https://github.com/dvx

Email: [david].[titarenco]@[gmail].[com]

This right here. Value prop quickly goes out the window when you're building anything novel or hard. I feel that I'm still spending the same amount of time working on stuff, except that now I'm also spending money on models.

> China is the world’s largest source of carbon emissions, and the air quality of many of its major cities fails to meet international health standards.

So tired of seeing this trope. Data center energy expenditure is like less than 1% of worldwide energy expenditure[1]. Have you heard of mining? Or agriculture? Or cars/airplanes/ships? It's just factually wrong and alarmist to spread the fake news that AI has any measurable effect on climate change.

[1] https://www.iea.org/reports/energy-and-ai/energy-supply-for-...

[1] https://microsoft.github.io/react-native-macos/docs/intro

[1] https://github.com/ykumards/eCLIP/commits/main/autoresearch

Which is fine, but please disclose it. Otherwise, like in this case, I'm going to assume the author is a moron that can't write for shit who thinks their readers are morons that can't read for shit.

This is not what they're getting at; I explained exactly what they're getting at. I mean, your equivalence of "loss" (what authors actually measured) and "performance" is just bizarre. We use benchmarks to measure performance, and the numbers there were like 1-5% better (apart from the GPQA-Diamond outlier).

Do people even read these papers?

This is not true. Authors claim that w.r.t. training, their method adds negigible overhead for AttnRes with no memory impact (but is way more complicated for Block AttnRes since we need to use pipelining for larger models, hence the O(Ld) & O(Nd) figures, with N ≪ L).

> WAY lower bandwidth requirements for inference.

Also not true. Paper has nothing to do with inference, apart from the benchmarks. If you're looking at the graph about "compute advantage," it's about training compute. They do some interpolation to get to the 1.25x number, basically answering the question "if non-AttnRes architecture were trained, how much compute would it take to get to the same loss as AttnRes?" (The answer being ~20% more compute.) It's an interesting claim, but there's all kinds of weird and unexpected convergence that can happen, so take it with a grain of salt.