HNHacker News
TopNewBestAskShowJobs

mirzap

2,317 karma · joined March 2, 2015

Curious developer.

[ my public key: https://keybase.io/mirzap; my proof: https://keybase.io/mirzap/sigs/5G0Sj20ied-ISqZWBbkxcQQ3pm3o5_yWUSkV9aHeRSk ]

submissionscomments
mirzap··on GitHub MCP exploited: Accessing private repositories via MCP
> An attack doesn’t have to be surprising to be an attack.

If you open an API to everyone, or put a password as plain text and index it, it's no surprise that someone accesses the "sensitive" data. Nor do I consider that an attack.

You simply can't feed the LLM the data, or grant it access to the data, then try to mitigate the risk by setting "guardrails" on the LLM itself. There WILL ALWAYS be a prompt to extract any data LLM has access to.

> Yes. That is exactly what the article recommends as a mitigation.

That's common sense, not mitigation. Expecting "security experts" to recommend that is like expecting a recommendation to always hash the password before storing it in the DB. Common sense. Obvious.

mirzap··on GitHub MCP exploited: Accessing private repositories via MCP
That's the thing. LLM or MCP is not a database. You can’t compare it. You simply can't set the permissions or guardrails within LLMs or MCPs. You always do it layer above (scoping to what LLM has access to).
mirzap··on GitHub MCP exploited: Accessing private repositories via MCP
Would you ever put a plain password text in a search engine and then complain if someone "extracted" that info with a keyword payload?
mirzap··on GitHub MCP exploited: Accessing private repositories via MCP
Bad analogy. It's more like indexing a password field in plain text, then opening an API to everyone and setting "guardrails" and permissions on the "password" field. Eventually, someone will extract the data that was indexed.
mirzap··on GitHub MCP exploited: Accessing private repositories via MCP
I read it, and "attack" does not make sense. If you grant access to MCP to access some data (data you want anybody has access to like public repos, and data that only you want to access to like private repos), you will always be able to craft the prompt that will "leak" the data you are only supposed to access. That's not surprising at all. The only way to prevent these kind of "leaks" is not to provide the data feed with private data to the agent.
mirzap··on GitHub MCP exploited: Accessing private repositories via MCP
I simply don't see how you could enforce a classic permission system on an MCP server. MCPs are API servers that allow LLMs access to context within the boundaries you set. You can set permissions for what an LLM has access to and define those boundaries. However, setting a permission on a context that an LLM has access to is futile. There will always be a prompt that will leak some "sensitive" data. This is like creating an index in a classic search engine with public and private data and then trying to enforce permissions based on certain keywords. There will always be a keyword that leaks something.
mirzap··on MCP: May Cause Pwnage – Backdoors in Disguise
I bet you used AI to write something that sounds smart, but trust me kid, it doesn't sound that way at all.
mirzap··on GitHub MCP exploited: Accessing private repositories via MCP
How is this considered an "exploit"? You give the agent a token that allows it to access a private repository. MCPs are just API servers. If you don't want something exposed in that API, don't grant them permissions to do so.
mirzap··on MCP: May Cause Pwnage – Backdoors in Disguise
Lol. Can't believe I've read this. It's like saying "REST: may cause pwnage". Everything the author found troubling can be said for any API server under certain implementation (or even some frameworks and their defaults), and everything is an implementation choice, not a protocol vulnerability.
mirzap··on Gumroad’s source is available
But that's the point he tries to make. When you "teach" LLM with some knowledge, you teach it a set of patterns. It won't necessarily drop the code that infringes copyright. Say you load Gumroad code into Gemini Pro context and say something like: "Check this app. Analyze the implementation of feature XY... I need you to help me implement feature XY... but in Go". Then, you can recreate an entire platform that will look nothing like the original but will have the same features and open source it.
mirzap··on Gumroad’s source is available
And how exactly are you going to prove derivative work?
mirzap··on US Administration announces 34% tariffs on China, 20% on EU
Let's imagine they do come back. Let's say you have a toaster factory in Vietnam, which allows the US company to produce low-cost toasters for about $100. The factory workers earn $1 per hour. Now, due to the tariffs, a toaster will cost $146. And you say, "Okay, they just need to move their factory to the US and then the problem will be solved - no more tariffs. " Fine. Do you really think a US worker will work for $1 per hour? Even if you automate 80% of the factory, you still won't offset the 46% tariff. You will never get a toaster priced at $100. More likely, it will be $300.
mirzap··on Show HN: Nue – Apps lighter than a React button
Honestly, after playing around with your example and tweaking a JavaScript example, I feel that my plain JS example is faster than the one with WASM/Rust on Nue (possibly due to the animations). Please disable animations; I think they make your WASM example feel slower than it probably is.

Here is a 1 million rows table in JS with filtering, sorting, and pagination: https://htmlpreview.github.io/?https://gist.githubuserconten...

mirzap··on Show HN: Nue – Apps lighter than a React button
It has nothing to do with architecture, but rather understanding how the DOM works. The DOM is notoriously slow, so you should never render a huge number of rows at once. You can render millions of rows in plain JavaScript without impacting performance.

Here, I have recreated your JS example with searching and filtering and it does not crash. It's trivial to reuse a similar approach with the real backend and real events from the event source.

https://htmlpreview.github.io/?https://gist.githubuserconten...

*Update: Here is 1 million rows table with search, filtering and pagination. In plain Javascript:

https://htmlpreview.github.io/?https://gist.githubuserconten...

mirzap··on Show HN: Nue – Apps lighter than a React button
Because he constructs a giant JSON by joining individual entries. Rendering that directly on the DOM will always cause the performance issues (even at the 10k entries). That's why you need to use virtualized list, it can be done in plain JS or using libraries like react-virtualized.

This works, plain JS 150k rows

    <style>
        #viewport {
            height: 600px;
            overflow-y: scroll;
            position: relative;
            border: 1px solid #ccc;
            width: 400px;
            margin: auto;
        }

        .item {
            position: absolute;
            left: 0;
            right: 0;
            height: 30px;
            padding: 5px;
            box-sizing: border-box;
            border-bottom: 1px solid #eee;
            font-family: Arial, sans-serif;
        }
    </style>

    <div id="viewport">
        <div id="content"></div>
    </div>


    <script>
        const viewport = document.getElementById('viewport');
        const content = document.getElementById('content');
        const itemHeight = 30;
        const totalItems = 150000;

        const items = Array.from({length: totalItems}, (_, i) => ({
            id: i + 1,
            name: `User #${i + 1}`
        }));

        content.style.height = `${totalItems * itemHeight}px`;

        function render() {
            const scrollTop = viewport.scrollTop;
            const viewportHeight = viewport.clientHeight;
            const start = Math.floor(scrollTop / itemHeight);
            const end = Math.min(totalItems, start + Math.ceil(viewportHeight / itemHeight) + 10);

            content.innerHTML = '';

            for (let i = start; i < end; i++) {
                const div = document.createElement('div');
                div.className = 'item';
                div.style.top = `${i * itemHeight}px`;
                div.textContent = items[i].name;
                content.appendChild(div);
            }
        }

        viewport.addEventListener('scroll', render);
        render();
    </script>
mirzap··on Show HN: Nue – Apps lighter than a React button
Make a demo with react-virtualized[0] and see if it crashes. Hint: It will not[1]. React can easily render 1 million rows with high performance without relying on WASM [2]

Here is the demo of react-virtualized[3], in which I entered 10m as the row count and scrolled to the bottom without crashing.

[0] https://github.com/bvaughn/react-virtualized

[1] https://www.youtube.com/watch?v=1JoEuJQIJbs

[2] https://medium.com/@priyankadaida/how-to-render-a-million-ro...

[3] https://bvaughn.github.io/react-virtualized/#/components/Lis...

*Update: Here I made a table with 1 million rows with search, filtering, and pagination. In plain Javascript:

https://htmlpreview.github.io/?https://gist.githubuserconten...

mirzap··on US tells European companies to comply with Donald Trump's anti-diversity order
Probably. They deny the fact that international commerce is mutually dependent. They think the world strictly depends on them and that companies would not be profitable if they sold products elsewhere. Sure, they will earn less money, but they will still be profitable, and the lifestyle around the world will improve because all the good products that were meant to go to the US will now be available in other countries.
mirzap··on US tells European companies to comply with Donald Trump's anti-diversity order
It's American exceptionalism at its best. They believe America is subsidizing the world when, in fact, it is the other way around. The world is subsidizing the American lifestyle.

That's why you see China jumping onto the global stage and following exactly the same playbook that the USA did after World War II—building connections and trading partners, investing in other countries, and subsidizing them because they see it as a long-term investment that will benefit China far more in the future than what they invest now.

mirzap··on Show HN: Nue – Apps lighter than a React button
> 150,000 records — far past where JavaScript (and React) would crash with a stack overflow error

I think react-virtualized and stack tables can easily handle 1 million rows client-side without a problem (I saw the demo somewhere).

Web development is about convenience, and the speed of development is far more important than ultra optimizations. People simply don't care about super optimizations because servers are so fast these days, and bandwidth is cheap, almost free.

But it's an interesting project. Good luck.

mirzap··on Entire staff at federal agency that funds libraries and museums put on leave
Every empire has fallen the same way - spending money on wars and games.
mirzap··on Charlie Javice convicted of defrauding JPMorgan in $175M startup sale
I don't think anyone is investing that kind of money without doing basic DD.
mirzap··on Charlie Javice convicted of defrauding JPMorgan in $175M startup sale
Do you know what zero knowledge proofs are? OpenAI could create ZKP that would be easy to verify.
mirzap··on OpenAI releasing new open model in coming months, seeks community feedback
OpenAI should release its frontier model as an open-weight model. There are already open-weight models that match OpenAI's best models (at the time of their release), so the idea that OpenAI would lose something by making its frontier models open-weight doesn't hold up. With an open-weight model, they would instantly kill any proprietary competitor, similar to what Google did to all its competitors with Android.

IMO, OpenAI should focus on tooling, infra, and setting standards for AI apps and profit from those. MCP is what "custom GPT" is supposed to be; OpenAI lost that battle, among many others.

Gopher started as a freely available protocol but later changed its licensing terms, requiring fees to be paid—similar to what OpenAI has done. We know how Gopher ended up: today, most people haven't even heard of it, despite its adoption at the beginning.

mirzap··on China's export boom sparks record number of WTO disputes
https://archive.ph/9lRNm
mirzap··on Electronic devices used for car thefts set to be banned
Stealing is already outlawed, so banning the devices is not comparable to murder. What would be comparable if the government decided to ban knives and hammers? Those "devices" are used to commit murders, and anyone possessing them could be prosecuted for the crime of murder.
mirzap··on Matt Mullenweg: I feel jdevalk stabbed me when I was down, betrayed confidence
Who attacked WordPress, exactly? Nobody. That's right. Nobody attacked Open Source. The only one who is doing wrong to the community and attacking OSS and Wordpress is Matt. He is mad that other people are using OSS and profiting from it more than he does. That's the real reason why Matt is so angry.

I'm sick of arguments, "Company X profits from OSS project Y and not making contribution to it". They ARE NOT required to contribute back. There's no such conditioning in OSS licenses. Deal with it when you are open-sourcing something. Make peace with the fact that somebody will probably make more money from that piece of code than you do. If you're not OK with that, don't open source it.

No OSS license has an issue with people making money from it. Only small greedy men who took 100s of millions from VCs have an issue with it.

mirzap··on Tesla's sales plummet across Europe
https://archive.ph/bclhi
mirzap··on OpenAI Furious DeepSeek Might Have Stolen All the Data OpenAI Stole from Us
Training costs are not the same as inference costs. DeepSeek (or anyone hosting DS largest model) will still need a lot of money and a bunch of GPU clusters to serve the customers.
mirzap··on Why OpenAI's $157B valuation misreads AI's future (Oct 2024)
Model capabilities are not plateauing; in fact, they are improving exponentially. I believe people struggle to grasp how AI works and how it differs from other technologies we invented. Our brains tend to think linearly; that's why we see AI as an "app." With AI (ASI), everything accelerates. There will be no concept of an "app" in ASI world.
mirzap··on Musk bashes $500B AI project Trump announced, backers don't have the money
People with massive egos, like Musk, often don’t meet a good end. He supported an AI "pause"—but only until he launched xAI. In his view, anything done without his involvement is inferior. Ultimately, people like this tend to make powerful enemies who eventually bring them down.
← PreviousPage 5 of 12Next →