Top all-time · #22867
Zbigniew Łukasiak
Personal blog · zzbbyy.substack.comaiprogramming
Zbigniew Łukasiak writes AI Adventures: A Programmer’s Journey, a personal blog at zzbbyy.substack.com about ai and programming. Since 2023, 3 different Hacker News readers have linked it 24 times, most often “You Need Rag, Not Finetuning”. It is still writing: the last post was on 2026-09-04, and you can follow it by feed.
Too few sampled comments (4) to say how often HN recommends it.?
Why HN links it
Posts on LLM retrieval and agent tooling: agentic RAG, an index of agent skills, and why RAG rather than fine-tuning is the way to teach a model new facts. Commenters link them as background only.
Summarised by Claude from a sample of 4 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
My notes on AI. Launched 3 years ago.
Most-linked posts?
- You Need Rag, Not Finetuning2 people · 6 mentions · first linked 2023-07-17
- Agentic RAG2 people · 3 mentions · first linked 2024-04-29
- Reinventing Daydreaming Machines2 people · 2 mentions · first linked 2026-07-16
What Hacker News says?
My wish list for LLM APIs to make them more useful for 'agentic' workflows: Finer grained control over the tools the LLM is supposed to use. The 'tool_choice' should allow giving a list of tools to choose. The point is that the list of all available tools is needed to interpret the past tool calls - so you cannot use it to also limit the LLM choice at a particular step. See also: https://zzbbyy.substack.com/p/two-ro…
Comment linking /p/two-roles-of-tool-schemas
on “Building Effective "Agents"”Submitted Separating Data from Instructions in Prompting as a story · 6 points
HN discussionLetting the LLM to decide what to do is a powerful technique. For example one pass RAG is very limited: https://zzbbyy.substack.com/p/why-iterative-thinking-is-cruc... To make it iterative you need the cede the control to the LLM.
Comment linking /p/why-iterative-thinking-is-crucial
on “Why we no longer use LangChain for building our AI agents”Submitted Reinventing Daydreaming Machines as a story · 1 points
HN discussionThese sampling based techniques is a rare occasion where experimenting with consumer hardware can let you improve on SOTA models. I don't think it will last - the end game surely will be a trainable sampler. But for now - enjoy tinkering: https://github.com/codelion/optillm implements a few of these techniques optillm authors suggest that the additional computations in Entropics don’t bring any better results in com…
Comment linking /p/o1-inference-time-turing-machines
on “Detecting when LLMs are uncertain”Submitted LLMs and Reasoning as a story · 1 points
HN discussionTraditional FTS returns the whole document - people take over from that point and locate the interesting content there. The problem with RAG is that it does not follow that procedure - it tries to find the interesting chunk in one step. Even though since ReAct we know that LLMs could follow the same procedure as humans. But we need an iterative RAG anyway: https://zzbbyy.substack.com/p/why-iterative-thinking-is-cruc…
Comment linking /p/why-iterative-thinking-is-crucial
on “Solving the out-of-context chunk problem for RAG”Submitted O1, inference time, Turing Machines and stuff as a story · 1 points
HN discussionWhat you describe sounds like Agetic RAG https://zzbbyy.substack.com/p/agentic-rag
Comment linking /p/agentic-rag
on “Show HN: How we leapfrogged traditional vector based RAG with a 'language map'”Submitted Agentic RAG as a story · 1 points
HN discussionSeems like the idea of Agentic Rag ( https://zzbbyy.substack.com/p/agentic-rag ). I have created a very bare bones implementation as an example at https://github.com/zby/LLMEasyTools/tree/main/examples/agent... (it uses whoosh for indexing) and I am working on a more complete one at https://github.com/zby/answerbot .
Comment linking /p/agentic-rag
on “Breaking up is hard to do: Chunking in RAG applications”Submitted You Need Rag, Not Finetuning as a story · 1 points
HN discussionThere is an article at the original site about that: https://www.anyscale.com/blog/fine-tuning-is-for-form-not-fa... Everybody new to this field thinks that he needs finetuning to teach the LLM of new facts. I made the same mistake initially, later I published a slightly ranty post on that: https://zzbbyy.substack.com/p/why-you-need-rag-not-finetunin...
Comment linking /p/why-you-need-rag-not-finetuning
on “Fine-Tuning Llama-2: A Comprehensive Case Study for Tailoring Custom Models”My review: https://zzbbyy.substack.com/p/quick-review-of-claudette
Comment linking /p/quick-review-of-claudette
on “Claudette, a new friend that makes Claude 3.5 Sonnet even nicer”This sounds like a paradox - but it is not. You don't give the bot the answer you expect - but only the ground facts, it generates the answer based on these facts by itself. This is a RAG system - and you need to treat it as a whole - it is an question answering machine that remembers the whole wikipedia. By the way just today I wrote a blogpost about the common misconception that to teach an LLM new facts you need…
Comment linking /p/why-you-need-rag-not-finetuning
on “Wikipedia-grounded chatbot “outperforms all baselines” on factual accuracy”This sounds like a paradox - but it is not. You don't give the bot the answer you expect - but only the ground facts, it generates the answer based on these facts by itself. This is a RAG system - and you need to treat it as a whole - it is an question answering machine that remembers the whole wikipedia. By the way just today I wrote a blogpost about the common misconception that to teach an LLM new facts you need…
Comment linking /p/why-you-need-rag-not-finetuning
on “Wikipedia-grounded chatbot “outperforms all baselines” on factual accuracy”I have made an experiment with 'idea generation' last year (with much worse llms): https://github.com/zby/DayDreamingDayDreaming The results were promising: https://zzbbyy.substack.com/p/reinventing-daydreaming-machin...
Comment linking /p/reinventing-daydreaming-machines
on “Can LLMs Perform Deep Technical Comprehension of Computer Architecture Papers”If intelligence is knowledge and the ability to apply it then maybe we should try to split AI into two systems: https://zzbbyy.substack.com/p/reasoning-machines
Comment linking /p/reasoning-machines
on “AI and the Frontier Paradox”The point is for the agent to have an index of available skills so that it can decide autonomously if it needs to load one. https://zzbbyy.substack.com/p/what-are-skills In my experience it doesn’t work too well with codex, but I expect llm providers to train them on that use case and improve the situation soon.
Comment linking /p/what-are-skills
on “Show HN: Agent Skills Leaderboard”> RAG is quite the hack. All you start with is the question, but why should the documents you need for the answer be similar to the question? I have always thought that a better RAG system would be to let the LLM use the search in the same way that a human would - think about good keywords and search the needed information iteratively until everything needed is retrieved. I have even wrote a proof of concept: https:…
Comment linking /p/design-principles-for-llm-based-systems
on “Infinite Context LLMs: Going Beyond RAG with Extended Minds”Finetuning is not useful for teaching new facts, the current solution for that is using RAG: https://zzbbyy.substack.com/p/why-you-need-rag-not-finetunin...
Comment linking /p/why-you-need-rag-not-finetuning
on “LLMs Unleashed: The Power of Fine-Tuning”Finetuning is not useful for teaching new facts, you need RAG for that: https://zzbbyy.substack.com/p/why-you-need-rag-not-finetunin...
Comment linking /p/why-you-need-rag-not-finetuning
on “LLMs Unleashed: The Power of Fine-Tuning”