Hacker News: c7b

New comment by c7b in "The unreasonable difficulty of time series forecasting"

c7b — Tue, 21 Jul 2026 19:06:18 +0000

Well. Lots of math that boils down to 'predicting the future is hard'. Especially when the future is one of social construction, that's what gets lost a bit here. Predicting the future is easier for planetary motions than for Bitcoin.

New comment by c7b in "Hacker wipes Romania's land registry database"

c7b — Mon, 20 Jul 2026 17:34:46 +0000

git as-is is completely unfit for purpose and a huge security risk for this use case, because it uses MD5, which is no longer collision-resistant.

If your proposal is 'something like git, with a few modifications to make it suitable for this use case', then that's exactly what I'm proposing. What you will need, once you've considered all requirements to the best extent feasible, will be a blockchain. Eg you do need the ability to make protocol updates, no matter the time scale of transactions.

New comment by c7b in "Hacker wipes Romania's land registry database"

c7b — Mon, 20 Jul 2026 17:23:41 +0000

Clearly, you would scope the features and choose implementation options so that it makes sense for the use case. Mining (ie PoW and/or Nakamoto consensus) is clearly inferior to any deterministic consensus here. A cryptocurrency isn't necessary and would be a distraction. But at least some limited programmability could make sense (eg for escrow).

New comment by c7b in "Hacker wipes Romania's land registry database"

c7b — Mon, 20 Jul 2026 15:43:28 +0000

That is what I'm saying. And you may be right, but the security guarantees are math and math is patient. If people decide to ignore it today, it'll still be true any time they decide to take another look.

New comment by c7b in "Hacker wipes Romania's land registry database"

c7b — Mon, 20 Jul 2026 15:20:41 +0000

Agreed that you don't need total ordering and hence consensus for pure asset transfers. But if you want to make any changes at all, like updating the list of nodes, you do. So in practice, you do need it. Every other proposal will fall short in a critical and avoidable way. Amending your git proposal with the necessary parts will turn it into a blockchain.

New comment by c7b in "Hacker wipes Romania's land registry database"

c7b — Mon, 20 Jul 2026 15:16:55 +0000

You need that anyway. It's not technology that's stopping notaries from falsifying transactions in the current system.

New comment by c7b in "Hacker wipes Romania's land registry database"

c7b — Mon, 20 Jul 2026 15:08:12 +0000

You can put an index on your blockchain DB, what are you talking? Every node can manage it as it pleases (within the limits of the consensus algorithm).

Yes, if you have a centralized model you don't need it. Just saying that this incident is the exact risk a blockchain is meant to mitigate through redundancy. You can say you don't care about this risk, but it doesn't change the truth of the statement.

New comment by c7b in "Hacker wipes Romania's land registry database"

c7b — Mon, 20 Jul 2026 14:56:50 +0000

2/3 attack. Read up on BFT.

New comment by c7b in "Hacker wipes Romania's land registry database"

c7b — Mon, 20 Jul 2026 14:56:28 +0000

Internally, you don't need it, agreed. But that is centralization risk exemplified in this incident. Which isn't necessary. Every notary could be a node. It would be a lot more resilient, and it would look a lot like a blockchain (not necessarily with a crypto currency though) if you do it properly.

New comment by c7b in "Hacker wipes Romania's land registry database"

c7b — Mon, 20 Jul 2026 14:53:38 +0000

A blockchain solves the problem of consensus under Byzantine faults. Payments are an incidental use case, and not even a good one because managing payments including avoiding double spending can be achieved with a weaker primitive than consensus.

New comment by c7b in "Hacker wipes Romania's land registry database"

c7b — Mon, 20 Jul 2026 14:43:44 +0000

A blockchain isn't stopping you from doing any of those things, if anything it's facilitating different backup models by different parties. A paper record is so much more susceptible to various kinds of risks, adversarial and otherwise. I recommend to look past the hype or hate at the technology. A blockchain is the logical extreme of where you end up when you think about how to sync backups, and you recognize that it comes with certain mathematical limitations on how much disruption you can handle.

New comment by c7b in "LM Studio Bionic: the AI agent for open models"

c7b — Thu, 16 Jul 2026 23:54:53 +0000

Neural machines were always going to be an alternative computing paradigm to von Neumann machines. Had it not been for Minsky we would arguably have gotten to a point where they're useful sooner. But why do you say that as if it's a small thing?

New comment by c7b in "Hands-On with the AMD Ryzen AI Halo"

c7b — Fri, 10 Jul 2026 19:45:58 +0000

Agreed, the Strix Halo doesn't feel good enough for $4k. It was supposed to be $2k (and was so until 6-12 months ago), which felt like a great deal. Not the best AI chip but you get what you pay for. A tinkerer's dream that could maybe even fit into a birthday gift budget for a lucky teenager. I hate to say it, but I hope they fail with their $4k box.

New comment by c7b in "The revenge of the philosophy majors"

c7b — Tue, 07 Jul 2026 16:00:03 +0000

Bit of a tangent, but it's fun to think about how much it takes to become a -er, -ian or -ist in a given field. Philosophy is probably one of the hardest, you need to be seen as up there with the all-time greats. In history or physics you probably need to be faculty, in economics you need to have a PhD, in engineering you don't even need a degree but you need to be practicing,...

New comment by c7b in "YC CEO says he ships 37K LoC AI code per day. A developer looked under the hood"

c7b — Tue, 07 Jul 2026 10:01:59 +0000

I think we're on the same team, I find the general attitude towards security in the current AI scene scary. I was just hoping for a bit more ammunition than what that article gave us.

New comment by c7b in "YC CEO says he ships 37K LoC AI code per day. A developer looked under the hood"

c7b — Tue, 07 Jul 2026 09:52:22 +0000

Hmm the list is a bit underwhelming. Basically, it's unnecessary requests, bloated JS, unoptimized images and generally poorly structured code. I would hate if that was where the average website is headed, but realistically, we were already headed there before LLMs. From the headline I was expecting CVEs, broken UX flows / business logic, leaked secrets.

New comment by c7b in "Price per 1M tokens is meaningless"

c7b — Mon, 06 Jul 2026 21:03:14 +0000

Even more important in a local context is the difference between token generation and prompt processing speed. We tend to focus on the former, but for multi-turn/agentic workflows the latter can dominate.

New comment by c7b in "AMD Ryzen AI Halo – $4k AI Dev Kit"

c7b — Mon, 06 Jul 2026 18:26:16 +0000

You're the one making things up. An M3 Ultra with 128GB RAM doesn't exist, the M3 Max has 410GB/s bandwidth [0]. I was of course talking about the M4 Max with 546GB/s, which was closer to twice the price of a Strix Halo mini PC in a typical configuration when it was still available. And memory bandwidth isn't everything, NVidia's lead in software is substantial, look up any tests comparing them side-by-side.

[0] https://en.wikipedia.org/wiki/Apple_M3 [1] https://en.wikipedia.org/wiki/Apple_M4

New comment by c7b in "AMD Ryzen AI Halo – $4k AI Dev Kit"

c7b — Mon, 06 Jul 2026 18:14:53 +0000

The M4 Max with 128GB RAM has 546GB/s memory bandwidth [0], compared to Strix Halo's 250 (on the label, I've yet to see a benchmark that tops 220). It's not available at 128GB RAM anymore, at least in my shop, but when it was not so long ago it was about 4,7k, or a little over twice the price of a cheaper Strix Halo PC (around 2,2k a few months ago).

[0] https://en.wikipedia.org/wiki/Apple_M4

New comment by c7b in "AMD Ryzen AI Halo – $4k AI Dev Kit"

c7b — Mon, 06 Jul 2026 18:08:38 +0000

But ideally they would be competitive, right? If your goal is LLM or Diffusion inference or - god forbid - training, you're going to get way better performance on DGX Spark. The difference is more stark than 250 vs 273 GB/s bandwidth delta would suggest.

Now I think it's totally fine to have a less capable offering, and the Strix Halo is still a mighty capable machine for inference on mid-size MoEs. At 2k it was a tinkerer's dream. But the performance difference should be reflected in the price. This is roughly a doubling of the price compared to less than a year ago without adding any notable features, it's appalling.