nnurmanov

You have to have fixed commands that LLM could execute, just limit its universe. I don't think it is a good practice to give LLMs access to everything.

I believe this may impact on post training.

Training for instruction following, reasoning, reflection, and self-monitoring may have forced models to learn resource allocation: compress the currently useful concepts into a sparse shared workspace instead of keeping all computation equally active.

This is actually interesting observation, what if they add resource allocation constraints in post training, will it force to add more workspaces?

Wake me up in 100 years and ask me what EU politicians are doing.

My answer: regulating something:)

We will never know as the article is behind a paywall:)

Actually, my mornings start with reading HN; I am more of a reader than a writer :) But I see your point: if you find it useful, then repay it with a contribution.

I agree on the repo link as well; it was indeed selfish on my side. HN is one of the few places where you can read human content, so let's keep it that way. Lesson completely learned on my side.

Communicating with mods about issues could be a good idea too, but I shall keep it to a minimum. Apparently, they are super busy fighting AI :)

In this case, could coding it manually be faster?:)

1. Did not use the UI framework currently being used in the web app; 2. Did not see the pattern, I always use header, but it skipped; 3. Wrong placement of elements, e.g. one element should have been on the left pane, instead it put it on the top right and so on.

I agree that on pure code they are OK, but I thought maybe I am missing something?

I guess HN has become an AI/vs human battleground:)

I guess UI hasn't been solved yet.

My submission was killed immediately. It took me weeks to build the product and days to polish the documentation. At one point, I even went 24 hours without sleep working on it—only to have the submission killed instantly.

I emailed @dang at hn@ycombinator.com but never heard back. What should I do now? I’m scared to repost because I don’t want to risk a permanent ban.

But don't turn on the noprocrast option;) I did it today and had to wait for 3 hours:)

How is it with UI? Any good?

I’m not sure this growth is sustainable. Even if the barriers are adequate, companies may be reluctant to invest if they believe the bubble could burst soon.

Any startup would envy this rate of growth:)

Fixed, now Fabric works with subscription plans as well. It took some time to figure it out, but the issue is resolved now. I shall push the updated release later to GitHub

To the comment below, I have no idea why CC subscription did not work, apparently they have some different way to connect? Thanks to the leaked source code, I am going to find it out:)

This is basically why I built HailMary Fabric.

Claude Code is good, but a lot of the cost is repo discovery: finding files, reading candidates, following imports, etc.

Fabric indexes the repo locally and gives Claude Code the relevant context up front. Nothing goes to HailMary.

Tiny early test on httpx: 58% lower API cost across 2 tasks, both passed tests.

Repo is here if anyone wants to try it or poke holes in it: https://github.com/hailmarylabs/fabric/

@dang Could you please unblock https://news.ycombinator.com/item?id=48807184, users say they do not see the post! I spent so many days to build the product! Not quite sure why this happpened? Was I banned?

Basecheck.ai - a database auditing tool, supports Oracle, MS SQL Server, PostgreSQL, Supabase and SQLite. Happy to demo

IMO, Freelance marketplace could be a good pivot.

But what about SO's famous toxicity? Can agents be toxic? SO's - “Your question is stupid.” can save a lot of resources;)

Yes, they say we have backup procedures, but have they ever tested that the backups work? They write procedures to please auditors:)

There is always a disgruntled employee:) They will do anything to do maximum harm

You have no idea how indifferent security officers can be-even when you point out critical issues. The other day, we flagged that a customer’s database had users with excessive privileges. Their only question: “Can this be exploited from the outside?”

No, but most breaches today come from compromised internal accounts that are then used to break everything.

Usually, companies have procedures for such events. But most do not.

That’s how ecosystems start: by cutting out third parties.

What about EU Inc? Is there any timeline?

Do you have to restart your computer to exit telnet?:)

They claim: Since AI models are also prone to hallucinations, the company always has board-certified human doctors from top health institutions such as Stanford, Harvard, and UCSF review the final diagnoses, lab orders, and medical prescriptions., but is it economically viable?

I agree. Here is my thinking. What if LLM providers will make short answers the default (for example, up to 200 tokens, unless the user explicitly enables “verbose mode”). Add prompt caching and route simple queries to smaller models. Result: a 70%+ reduction in energy consumption without loss of quality. Current cost: 3–5 Wh per request. At ChatGPT scale, this is $50–100 million per year in electricity (at U.S. rates).

In short mode: 0.3–0.5 Wh per request. That is $5–10 million per year — savings of up to 90%, or 10–15 TWh globally with mass adoption. This is equivalent to the power supply of an entire country — without the risk of blackouts.

This is not rocket science — just a toggle in the interface and I believe, minor changes in the system prompt. It increases margins, reduces emissions, and frees up network resources for real innovation.

And what if EU/California enforces such mode? This will greatly impact DC economy.