Hacker News: msvana

New comment by msvana in "LLMs work best when the user defines their acceptance criteria first"

msvana — Sat, 07 Mar 2026 18:29:13 +0000

I think there is one problem with defining acceptance criteria first: sometimes you don't know ahead of time what those criteria are. You need to poke around first to figure out what's possible and what matters. And sometimes the criteria are subjective, abstract, and cannot be formally specified.

Of course, this problem is more general than just improving the output of LLM coding tools

New comment by msvana in "Labor market impacts of AI: A new measure and early evidence"

msvana — Fri, 06 Mar 2026 07:13:12 +0000

I work as an ML engineer/researcher. When I implement a change in an experiment it usually takes at least an hour to get the results. I can use this time to implement a different experiment. Doesn't matter if I do it by hand or if I let an agent do it for me, I have enough time. Code isn't the bottleneck.

I also heard an opinion that since writing code is cheap, people implement things that have no economic value without really thinking it through.

I built an RGB controller with Arduino

msvana — Sat, 28 Feb 2026 14:44:23 +0000

Article URL: https://svana.name/2026/02/i-built-an-rgb-controller-with-arduino/

Comments URL: https://news.ycombinator.com/item?id=47195936

Points: 9

# Comments: 1

Do LLMs hallucinate more in Czech than in English?

msvana — Mon, 16 Feb 2026 08:22:53 +0000

Article URL: https://svana.name/2026/02/do-llms-hallucinate-more-in-czech-than-in-english/

Comments URL: https://news.ycombinator.com/item?id=47032358

Points: 2

# Comments: 0

Hallucinations in LLMs: What are they and what causes them

msvana — Thu, 08 Jan 2026 09:26:45 +0000

Article URL: https://svana.name/2026/01/hallucinations-in-llms-what-are-they-and-what-causes-them/

Comments URL: https://news.ycombinator.com/item?id=46539050

Points: 1

# Comments: 1

Managing GPU Rentals with Rsync: Workflow for Volatile Cloud Resources

msvana — Thu, 30 Oct 2025 10:32:01 +0000

Article URL: https://svana.name/2025/10/managing-gpu-rentals-with-rsync-workflow-for-volatile-cloud-resources/

Comments URL: https://news.ycombinator.com/item?id=45758372

Points: 2

# Comments: 0

Implementing a local AI coding agent is hard

msvana — Sun, 28 Sep 2025 13:00:23 +0000

Article URL: https://svana.name/2025/09/implementing-a-fully-local-ai-coding-agent-is-hard/

Comments URL: https://news.ycombinator.com/item?id=45404023

Points: 2

# Comments: 1

Show HN: Read-only AI coding assistant

msvana — Wed, 24 Sep 2025 08:47:56 +0000

I am a bit conservative when it comes to using AI to write code for me. However, I want to use AI to chat about my code - something like ChatGPT that sees my local files. So I started working on FileChat.

FileChat is still quite new, and any feedback is appreciated. Please file an issue if you find a bug or have a feature suggestion.

Comments URL: https://news.ycombinator.com/item?id=45357827

Points: 1

# Comments: 0

How I solved PyTorch's cross-platform nightmare

msvana — Mon, 08 Sep 2025 04:59:35 +0000

Article URL: https://svana.name/2025/09/how-i-solved-pytorchs-cross-platform-nightmare/

Comments URL: https://news.ycombinator.com/item?id=45164762

Points: 73

# Comments: 29

Lessons from AI Safety for Businesses

msvana — Fri, 22 Aug 2025 09:31:46 +0000

Article URL: https://svana.name/2025/08/reading-club-lessons-from-ai-safety-for-businesses/

Comments URL: https://news.ycombinator.com/item?id=44982531

Points: 1

# Comments: 0

Ask HN: How do you imagine an AI-driven utopia?

msvana — Mon, 28 Jul 2025 08:45:48 +0000

Imagine that the development of artificial superintelligence goes extremely well and we solve any possible obstacle. How do you imagine the ideal future, say in 50 to 100 years?

Comments URL: https://news.ycombinator.com/item?id=44708709

Points: 1

# Comments: 2

New comment by msvana in "A non-anthropomorphized view of LLMs"

msvana — Tue, 08 Jul 2025 04:40:04 +0000

This reminds me of the idea that LLMs are simulators. Given the current state (the prompt + the previously generated text), they generate the next state (the next token) using rules derived from training data.

As simulators, LLMs can simulate many things, including agents that exhibit human-like properties. But LLMs themselves are not agents.

More on this idea here: https://www.alignmentforum.org/posts/vJFdjigzmcXMhNTsx/agi-s...

This perspective makes a lot of sense to me. Still, I wouldn't avoid anthropomorphization altogether. First, in some cases, it might be a useful mental tool to understand some aspect of LLMs. Second, there is a lot of uncertainty about how LLMs work, so I would stay epistemically humble. The second argument applies in the opposite direction as well: for example, it's equally bad to say that LLMs are 100% conscious.

On the other hand, if someone argues against anthropomorphizing LLMs, I would avoid phrasing it as: "It's just matrix multiplication." The article demonstrates why this is a bad idea pretty well.

LLMs Are Simulators

msvana — Wed, 25 Jun 2025 04:43:26 +0000

Article URL: https://www.alignmentforum.org/posts/vJFdjigzmcXMhNTsx/simulators

Comments URL: https://news.ycombinator.com/item?id=44373676

Points: 1

# Comments: 0

New comment by msvana in "Ask HN: Why study anything if AGI is (supposedly) coming?"

msvana — Wed, 18 Jun 2025 15:15:54 +0000

I had this exact analogy in mind when writing point no. 3

New comment by msvana in "Ask HN: Why study anything if AGI is (supposedly) coming?"

msvana — Wed, 18 Jun 2025 15:08:16 +0000

Thanks, sounds kinda similar to my first point.

Ask HN: Why study anything if AGI is (supposedly) coming?

msvana — Wed, 18 Jun 2025 14:46:07 +0000

I had this question in my head for a few days. So far, I've been able to come up with 4 reasons:

1. It’s about optimizing the worst case. Maybe AGI isn’t coming that soon; we don’t know. So it might be worth being prepared for multiple scenarios.

2. AGI might replace average humans, but we would still need the top performers to be human. If this is the case, things still need to change. It might make less sense for most people to pursue careers in, say, software development, unless they are exceptionally talented, motivated, or passionate.

3. We can learn new things just because they are fun. Most people who learn how to play an instrument won’t ever make money from it. In the future, programmers and other mainly white-collar workers might end up in a similar situation. It will be a hobby, not a job.

4. Related to the previous point, learning is also a social activity. It’s about doing something together, meeting new people with similar interests, making new friends, and possibly even finding spouses.

Would you add anything else to this list? Or do you think studying anything is pointless?

Comments URL: https://news.ycombinator.com/item?id=44310369

Points: 5

# Comments: 10

The value of commercial coding courses

msvana — Thu, 12 Jun 2025 19:31:38 +0000

Article URL: https://svana.name/2025/05/on-the-value-of-commercial-coding-courses/

Comments URL: https://news.ycombinator.com/item?id=44262167

Points: 1

# Comments: 0

New comment by msvana in "Getting AI to write good SQL"

msvana — Sat, 17 May 2025 06:44:32 +0000

Problem no. 2 (Understanding user intent) is relevant not only to writing SQL but also to software development in general. Follow-up questions are something I had in mind for a long time. I wonder why this is not the default for LLMs.

Would I do a PhD again?

msvana — Fri, 16 May 2025 06:29:15 +0000

Article URL: https://svana.name/2025/01/would-i-do-a-phd-again/

Comments URL: https://news.ycombinator.com/item?id=44002356

Points: 2

# Comments: 0

AI-LieDar: Examine the Trade-Off Between Utility and Truthfulness in LLM Agents

msvana — Fri, 02 May 2025 19:45:09 +0000

Article URL: https://arxiv.org/abs/2409.09013

Comments URL: https://news.ycombinator.com/item?id=43873925

Points: 1

# Comments: 0