Hacker News: nahco314

New comment by nahco314 in "Ask HN: How do LLMs run code described in prompts?"

nahco314 — Fri, 14 Feb 2025 17:46:47 +0000

I don't know exactly how OpenAI does it, but I think it's probably something like this: the OpenAI server that generates the responses reads and executes the ‘code that LLM wants to execute’ that exists in the LLM output in some way, and then passes the execution results as internal conversation history. As a concrete method, for example, the code to be executed could be enclosed in tags such as , or it may be achieved using metadata at a lower layer.

However, although GPT-4o should have a code execution function, it is possible that the code is not actually being executed and that the LLM is simply generating output that looks like (but is not actually) correct.

New comment by nahco314 in "Show HN: Transform your codebase into a single Markdown doc for feeding into AI"

nahco314 — Fri, 14 Feb 2025 14:10:13 +0000

I have always used o1 pro and deep research, but these are only available through the web UI. there is no doubt that cursor and others have a better UI, but the demand for this type of tool exists because OpenAI does not release an API

New comment by nahco314 in "Show HN: Transform Your Codebase into a Single Markdown Doc for Feeding into AI"

nahco314 — Fri, 14 Feb 2025 13:58:29 +0000

This is self-promotional, but https://github.com/nahco314/feed-llm has TUI to choose what to give to llm. There are many similar tools out there, but I think this approach is relatively effective for larger code bases.

New comment by nahco314 in "OpenAI Roadmap Update for GPT-4.5 and GPT-5"

nahco314 — Thu, 13 Feb 2025 01:26:42 +0000

I think this means that the cost of cutting-edge models such as o3 pro has become too high, and that ordinary users can no longer use them freely. I fear that, even though it is clear that smarter models are optimal, OpenAI will use cheaper models for its own convenience.

Show HN: feed-llm – A TUI-based file selection tool for feeding code to LLMs

nahco314 — Tue, 11 Feb 2025 18:06:21 +0000

Article URL: https://github.com/nahco314/feed-llm

Comments URL: https://news.ycombinator.com/item?id=43016000

Points: 1

# Comments: 0

New comment by nahco314 in "Ask HN: What inspires you about computing?"

nahco314 — Tue, 11 Feb 2025 16:56:51 +0000

In my experience, I think it’s primarily a mental health issue. My passion for computers has always been rooted in a fundamental curiosity, rather than any deep reason or meaning. Simply writing code and watching software run used to excite me—it was fun, pure and simple.

There was a time, though, when it all felt utterly meaningless and made me depressed. I found myself thinking things like, “Even if I build this, it won’t make any money,” or “This job may pay, but it doesn’t solve any real, fundamental problems.” Intellectually, I knew I used to do it just because I enjoyed it, yet I couldn’t stop those thoughts. However, as my depression went into remission, I started to feel that simple joy again (though I’m not fully recovered yet).

I suspect you’re going through something similar. While it’s important to think, I believe that rather than overthinking, focusing on your mental health first is the better approach—though of course, I understand how hard that can be.

New comment by nahco314 in "Ask HN: Examples of O1 pro DeepResearch proving its worth?"

nahco314 — Tue, 11 Feb 2025 13:15:22 +0000

They are very good at searching for niche results that are not often found by simply looking them up on Google. It is often depressing to find a better alternative after building software, but if you ask deep research first, they will find all the competition.

example: https://chatgpt.com/share/67ab4d04-4470-8011-a67d-fd7effc923...

Ask HN: Are there any LLM-based translation sites with excellent UI?

nahco314 — Sun, 09 Feb 2025 16:39:04 +0000

google translate or deepl are useful. The translation accuracy and speed are perfect, and the UI is excellent. However, deepl pro is a bit expensive and I would like to switch.

I think modern LLMs are better and some models can combine a response as fast as google translate and accuracy like deepl. However, it is tedious to go to sites like ChatGPT, type “please translate the following text” and wait for the tokens to be generated bit by bit. Do you know of any translation site with an architecture that directly wraps GPT or claude that combines UI, accuracy, and speed?

In particular, what we need is a UI that displays the translation immediately after copying a sentence (no translation button), and a feature that says, “If you paste an English sentence, it automatically becomes an EN->JA translation, and if you paste a Japanese sentence, it automatically becomes a JA->EN translation”.

Comments URL: https://news.ycombinator.com/item?id=42991705

Points: 1

# Comments: 0

New comment by nahco314 in "Ask HN: How are Small Language Models used commercially?"

nahco314 — Sun, 09 Feb 2025 15:58:47 +0000

I think jina's Reader-LM is a good example. It's useful (I use it) and LLM is too much for this use.

Show HN: cb – The simplest CLI clipboard tool

nahco314 — Fri, 07 Feb 2025 13:22:08 +0000

Article URL: https://github.com/nahco314/cb

Comments URL: https://news.ycombinator.com/item?id=42972211

Points: 1

# Comments: 0

New comment by nahco314 in "Show HN: lscat – ls for directories, cat for files"

nahco314 — Thu, 03 Oct 2024 17:56:54 +0000

I created independent binary for the sake of using the same thing in bash/fish/powershell and for the ease of adding functions, but for simple use, that way would be better.

Show HN: lscat – ls for directories, cat for files

nahco314 — Thu, 03 Oct 2024 13:26:07 +0000

Article URL: https://github.com/nahco314/lscat

Comments URL: https://news.ycombinator.com/item?id=41730585

Points: 3

# Comments: 3