afiodorov

I keep working on

https://hn.fiodorov.es/ - an agentic search across HackerNews comments

https://climate.fiodorov.es/ - a climate agent with access to climatic data for many cities

https://lyrics.fiodorov.es - a way to search and translate song lyrics using an LLM

So far really wasn't an issue - I put $5 there and forgot about it. If traffic explodes will probably offer bring your own key or add credits/ads. Also just providing MCP for your agent to use is an option.

Queryable climate data for major cities. The main question I'm trying to answer is: where can I maximize the number of comfortable hours I can spend outdoors, whether doing light activity or just sitting around?

DeepSeek can query a processed climate dataset directly, and other agents can use the same interface:

https://climate.fiodorov.es

---

See also:

https://hn.fiodorov.es for RAG search across Hacker News comments

She was not offended but thought I could benefit from therapy. She didn’t share my opinion - she said it’d not bother her if AI could eventually solve problems for her.

I later realised that my motivation depended heavily on believing I was making a contribution that would otherwise go unmade. The prospect of AI doing that work undermined my motivation, but it needn’t undermine hers. I was trying to explain why I was struggling to continue, though I can see how calling the work “pointless” came across as dismissive.

I used to be a PhD student more than a decade ago, and I published a paper containing a solution to an open problem. Yet shortly after my first publication I became increasingly disillusioned, because I started to think that within my lifetime AI would reach and eventually surpass my ability to solve such problems—and that we only had a decade or two left.

So I started saying that it only made sense to focus on problems whose solutions would be useful immediately. I even emailed my supervisor about it, arguing that our efforts were “pointless” in the sense that AI-related problems were much more pertinent and had to be prioritised.

My supervisor thought I was bonkers. I still have the email, though. Quoting myself from April 2015:

> By 2030-2040 we will have enough computing power to simulate a human brain neuron by neuron. Once we manage to create a human intelligence we will be one little step away from super intelligence: just set the intelligence to modify itself and see the exponential growth in action. Our human intelligence is bounded by a number of biological factors (e.g. size of a skull) and even the smart human who has ever lived will appear to be a primitive ant to a supper intelligence (machine intelligence will also have perfect motivation). There is plenty of literature on this if you are interested in discussing this further. > > What does it have to do with research in pure maths? I can say that research in pure maths which won't come handy in the next 60 years is just wasted effort. The super intelligence will be able to do maths way better than humans. I believe a lot of current efforts should go into researching of artificial intelligence (or areas to do with AI) instead rather than the pure maths. I want to be proven wrong but most mathematicians I interact with are too narrow-minded to counter me and they just laugh about even contemplating the above. Frankly I am myself so perplexed that I take the above seriously, but I do and it's hurting my motivation.

I'm quite curious what my supervisor thinks of that email now.

I prefer getting rid of old devices that needing to store old cables ad-infinitum.

A RAG search accross all HackerNews comments is back online with a new design: https://hn.fiodorov.es/

The running costs are very low. Since posting it today we burned 30 cents in DeepSeek inference. Postgres instance though costs me $40 a month on Railway; mostly due to RAM usage during to HNSW incremental update.

Daily updates I do on my m4 mac air: takes about 5 minutes to process roughly 10k fresh comments. Historic backfill was done on an Nvidia GPU rented on vast.ai for a few dollars. If I recall correctly took about an hour or so. It’s mentioned in the README.md on GitHub.

I've been embedding all HN comments since 2023 from BigQuery and hosting at https://hn.fiodorov.es

Source is at https://github.com/afiodorov/hn-search

Apparently Persian and Russian are close. Which is surprising to say the least. I know people keep getting confused about how Portuguese from Portugal and Russian sound close yet the Persian is new to me.

I've found that building my side projects to be "scalable" is a practical side effect of choosing the most cost-effective hosting.

When a project has little to no traffic, the on-demand pricing of serverless is unbeatable. A static site on S3 or a backend on Lambda with DynamoDB will cost nothing under the AWS free tier. A dedicated server, even a cheap one, is an immediate and fixed $8-10/month liability.

The cost to run a monolith on a VPS only becomes competitive once you have enough users to burn through the very generous free tiers, which for many side projects is a long way off. The primary driver here is minimizing cost and operational overhead from day one.

RAG search that contains all HN comments since 2023

https://hn.fiodorov.es

I treat it more like a homework exercise for a Coursera course but I like the result.

> It was uncomfortable at first. I had to learn to let go of reading every line of PR code. I still read the tests pretty carefully, but the specs became our source of truth for what was being built and why.

This is exactly right. Our role is shifting from writing implementation details to defining and verifying behavior.

I recently needed to add recursive uploads to a complex S3-to-SFTP Python operator that had a dozen path manipulation flags. My process was:

* Extract the existing behavior into a clear spec (i.e., get the unit tests passing).

* Expand that spec to cover the new recursive functionality.

* Hand the problem and the tests to a coding agent.

I quickly realized I didn't need to understand the old code at all. My entire focus was on whether the new code was faithful to the spec. This is the future: our value will be in demonstrating correctness through verification, while the code itself becomes an implementation detail handled by an agent.

  sign of serious organizational disfunction.

You're not wrong, but it's a "dysfunction" that many successful tech companies have learned to leverage.

The reality is, most engineers spend far less than half their time writing new code. This is where the 80/20 principle comes into play. It's common for 80% of a company's revenue to come from 20% of its features. That core, revenue-generating code is often mature and requires more maintenance than new code. Its stability allows the company to afford what you call "dysfunction": having a large portion of engineers work on speculative features and "big bets" that might never see the light of day.

So, while it looks like a bug from a pure "coding hours" perspective, for many businesses, it's a strategic feature!

I live next to an abandoned building from the Spanish property boom. It's now occupied illegally. Hype's over yet the consequence is staring at me every day. I am sure it'll eventually be knocked down or repurposed yet it'd be better had the misallocation never happened.

The crux of the article is asking whether such a large investment is justified; downplaying the article saying it's only X% of the GDP compared to Y doesn't address the issue.

Nice clip yet it does not make it clear why 9% is the good value of GDP. Why not 7%?

My rule of the thumb is that Dataclasses are for compile-time type-checking, pydantic classes are for run time type checking.

You're right, they can try to manipulate you on a thousand tiny things. My counter-argument is that at a certain point, it's not worth the mental energy to fight over what amounts to pennies on the dollar.

Anecdotally, when I bought my car recently, they forgot to even offer me the extended warranty they'd planned to push. I find it funny to think it was so minor, even they forgot to care.

Let's be clear: it's entirely possible to leave the "herd". People can and do go completely off-grid and thus disengage from capitalism. The crucial point is that the vast majority of us choose not to. That choice is what makes your "slowest antelope" analogy so much more complex.

An antelope's greatest desire is to be in the herd, because while it may contain a lion, the world outside contains a thousand wolves.

We've built a herd—society—that is incredibly effective at holding those wolves at bay: famine, plague, and chaos. We willingly participate because it provides "shields" our ancestors could only dream of. The problem isn't the herd itself; it's the lion that we allow to stalk within it.

What I am suggesting isn't to abandon this safety and comfort brought by modern capitalism. It's to improve the herd—to enjoy its protections while finding ways to tame, cage, or evade the lion of exploitation. What we're discussing here aren't futile attempts to escape, but vital tactics for building a better, safer herd for everyone.

There's nothing an algorithm can do against disciplined, intentional engagement.

If you know which car you want to buy it doesn't matter what the salesman has to say.

We should not underestimate the timeless human response to being manipulated: disengagement.

This isn't theoretical, it's happening right now. The boom in digital detoxes, the dumbphone revival among young people, the shift from public feeds to private DMs, and the "Do Not Disturb" generation are all symptoms of the same thing. People are feeling the manipulation and are choosing to opt out, one notification at a time.

We've been visited by alien intelligence that is simultaneously fascinating and underwhelming.

The real issue isn't the technology itself, but our complete inability to predict its competence. Our intuition for what should be hard or easy simply shatters. It can display superhuman breadth of knowledge, yet fail with a confident absurdity that, in a person, we'd label as malicious or delusional.

The discourse is stuck because we're trying to map a familiar psychology onto a system that has none. We haven't just built a new tool; we've built a new kind of intellectual blindness for ourselves.

If you're hiring 16 year olds just because of their ability to write code sounds like you're bottlenecked by writing code. Your comment doesn't clarify why you disagree.

I think the "more capital than ideas" problem is highly contextual and largely a Silicon Valley-centric view.

There is immense, unmet demand for good software in developing countries—for example, robust applications that work well on underpowered phones and low-bandwidth networks across Africa or Southeast Asia. These are real problems waiting for well-executed ideas.

The issue isn't a lack of good ideas, but a VC ecosystem that throws capital at ideas of dubious utility for saturated markets, while overlooking tangible, global needs because they don't fit a specific hyper-growth model.

Even without LLMs, we were approaching a point of saturation where software development was bottlenecked by market demand and funding, not by a shortage of code. Our tooling has become so powerful that the pure act of programming is secondary.

It's a world away from when the industry began. There's a great story from Bill Gates about a time when his ability to simply write code was an incredibly scarce resource. A company was so desperate for programmers that they hired him and Paul Allen as teenagers:

  "So, they were paying penalties... they said, 'We don’t care [that they are kids].' You know, so I go down there. You know, I’m like 16, but I look about 13. They hire us. They pay us. It’s a really amazing project... they got a kick out of how quickly I could write code."

That story is a powerful reminder of how much has changed. Writing code was the bottleneck years ago. However the core problem has shifted from "How do we build it?" to "What should we build and is there a business for it?"

Source: https://youtu.be/H1PgccykclM?si=YuIFsUcWc6sHRkAg

I heard that actors & singers don't necessarily manage to fix the accent in the natural speech so they can only recite extracts perfectly well.

European Portuguese sounds very Slavic; I'm sure Russians have a blast with it. English is a phonetically isolated language, largely due to the Great Vowel Shift. Unlike English, most languages have a closer linguistic relative. This makes English challenging for most people to learn, and it also makes it difficult for native English speakers to learn a foreign language without a heavy accent.

When you spend some time transcribing live, impromptu speech, you'll notice that it often doesn't follow the rules of written grammar; speakers frequently abandon sentences midway through.

For example, in the linked clip[^1], the speaker says:

  "uh the European Union uh that's not a US creation that's a you guys creation so don't ex..[abandoned word] the strength of the west [abandoned sentence] and the west is a really I don't know what"

For a moment, she struggles to express herself. Yet, there's a qualitative difference between not knowing what to say because a thought is not fully formed, and knowing what you want to say but realizing you've forgotten the specific word you need. For instance, you might be about to say "cherry," only to find you've forgotten the word and instead say something more general, like "forest fruit (fruta de bosque)," which is still correct but less precise.

[^1]: https://youtu.be/_hBd8w-Hlm4?si=7-kvpUoeYo5ODPiI&t=787