Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Claude Code can debug low-level cryptography

words.filippo.io

Post reply on HN
Print view View original

111–120 of 217 posts

Re: Claude Code can debug low-level cryptography

#111

Post by ribosometronome »

Klathmon wrote: post #95
ori_b wrote: post #83

Earlier quoted context omitted.

Doing it yourself is how you build and maintain the muscles to do it yourself. If you only do it yourself when the LLM fails, how will you maintain those muscles?

If the LLM is able to handle it why do you need to maintain those specific skills?

Should we not teach kids math because calculators can handle it?

Practically, though, how would someone become good at just the skills LLMs don't do well? Much of this discussion is about how that's difficult to predict, but even if you were a reliable judge of what sort of coding tasks LLMs would fail at, I'm not sure it's possible to only be good at that without being competent at it all.

Re: Claude Code can debug low-level cryptography

#112

Post by oefrha »

timr wrote: post #91
chatmasta wrote: post #72

Earlier quoted context omitted.

What makes it better than VSCode Co-pilot with Claude 4.5? I barely program these days since I switched to PM but I recently started using that and it seems pretty effective… why should I use a fork instead?

There’s really no functional difference. The VSC agent mode can do everything you want an agent to do, and you can use Claude if you like. If you want to use the CLI instead, you can use Claude Code (or the GitHub one, or Codex, or Aider, or…) I suspect that a lot of the “try using Claude code” feedback is just another version of “you’re holding it wrong” by people who have never tried VSC (parent is not in this grou…

No, there’s pretty noticeable difference between different tools even when they use the same model and interaction pattern. For instance I’ve used both GitHub Copilot and Cursor interactive agents (which are basically the same UX) aplenty in the past couple months for comparison, and GH Copilot is almost always dumber then Cursor, sometimes getting stuck on the stupidest issues. I assume context construction is likely the biggest culprit; Cursor charges by tokens while GH Copilot charges by request, so GHC attempts to pass as little as possible (see all the greps) and then fail a lot more. Plus its patching algorithm has always been shit (I’ve used GHC since it came out as better autocomplete).

Re: Claude Code can debug low-level cryptography

#113

Post by NoraCodes »

simonw wrote: post #11

Using coding agents to track down the root cause of bugs like this works really well: > Three out of three one-shot debugging hits with no help is extremely impressive. Importantly, there is no need to trust the LLM or review its output when its job is just saving me an hour or two by telling me where the bug is, for me to reason about it and fix it. The approach described here could also be a good way for LLM-skepti…

> start exploring how these tools can help them without feeling like they're [...] ripping off the work of everyone who's code was used to train the model

But you literally still are. If you weren't, it should be trivially easy to create these models without using huge swathes of non-public-domain code. Right?

Re: Claude Code can debug low-level cryptography

#114

Post by ojosilva »

jen729w wrote: post #54
cmrdporcupine wrote: post #8

Earlier quoted context omitted.

Try what I've done: use the Claude Code tool but point your ANTHROPIC_URL at a DeepSeek API membership. It's like 1/10th the cost, and about 2/3rds the intelligence. Sometimes I can't really tell.

Or a 3rd party service like https://synthetic.new , of which I am an unaffiliated user.

I'm using CC + GLM 4.6 from Synthetic and results are top notch for $60/mo, speed is fast and servers are close to home than z.ai

Re: Claude Code can debug low-level cryptography

#115

Post by ViewTrick1002 »

simonw wrote: post #41
pton_xd wrote: post #33

> Full disclosure: Anthropic gave me a few months of Claude Max for free. They reached out one day and told me they were giving it away to some open source maintainers. Related, lately I've been getting tons of Anthropic Instagram ads; they must be near a quarter of all the sponsored content I see for the last month or so. Various people vibe coding random apps and whatnot using different incarnations of Claude. Or j…

I think it might be that they've hit product-market fit. Developers find Claude Code extremely useful (once they figure out how to use it). Many developers subscribe to their $200/month plan. Assuming that's profitable (and I expect it is, since even for that much money it cuts off at a certain point to avoid over-use) Anthropic would be wise to spend a lot of money on marketing to try and grow their paying subscribe…

I just don’t see how they can build a moat.

I don’t feel like paying for a max level subscription, but am trying out MCP servers across OpenAI, Anthropic etc so I pay for the access to test them.

When my X hour token allotment runs out on one model I jump to the next closing Codex and opening Claude code or whatever together with altering my prompting a tiny bit to fit the current model.

Being so extremely fungible should by definition be a race to zero margins and about zero profit being made in the long run.

I suppose they can hope to make bank the next 6-12 months but that doesn’t create a longterm sustainable company.

I guess they can try building context to lock me in by increasing the cost to switch - but this is today broken by me every 3-4 prompts clearing the context because I know the output will be worse if I keep going.

Re: Claude Code can debug low-level cryptography

#116

Post by Wowfunhappy »

ribosometronome wrote: post #111
Klathmon wrote: post #95

Earlier quoted context omitted.

If the LLM is able to handle it why do you need to maintain those specific skills?

Should we not teach kids math because calculators can handle it? Practically, though, how would someone become good at just the skills LLMs don't do well? Much of this discussion is about how that's difficult to predict, but even if you were a reliable judge of what sort of coding tasks LLMs would fail at, I'm not sure it's possible to only be good at that without being competent at it all.

> I'm not sure it's possible to only be good at that without being competent at it all.

This is, in fact, why we teach kids math that calculators could handle!

Re: Claude Code can debug low-level cryptography

#117

Post by simonw »

ViewTrick1002 wrote: post #115
simonw wrote: post #41

Earlier quoted context omitted.

I think it might be that they've hit product-market fit. Developers find Claude Code extremely useful (once they figure out how to use it). Many developers subscribe to their $200/month plan. Assuming that's profitable (and I expect it is, since even for that much money it cuts off at a certain point to avoid over-use) Anthropic would be wise to spend a lot of money on marketing to try and grow their paying subscribe…

I just don’t see how they can build a moat. I don’t feel like paying for a max level subscription, but am trying out MCP servers across OpenAI, Anthropic etc so I pay for the access to test them. When my X hour token allotment runs out on one model I jump to the next closing Codex and opening Claude code or whatever together with altering my prompting a tiny bit to fit the current model. Being so extremely fungible s…

The money right now is in company subscriptions. If I went to my employer and said I can save an hour of work per day with a Claude Code subscription (and they have any sense at all) they should laugh at that $200/month price tag and pay up in an instant.

The challenge is definitely in the competition though. GPT-5-Codex offered _very_ real competition for Claude Sonnet 4 / Opus 4 / Opus 4.1 - for a few weeks OpenAI were getting some of those subscribers back until Sonnet 4.5 landed. I expect that will happen frequently.

Re: Claude Code can debug low-level cryptography

#118

Post by simonw »

NoraCodes wrote: post #113
simonw wrote: post #11

Using coding agents to track down the root cause of bugs like this works really well: > Three out of three one-shot debugging hits with no help is extremely impressive. Importantly, there is no need to trust the LLM or review its output when its job is just saving me an hour or two by telling me where the bug is, for me to reason about it and fix it. The approach described here could also be a good way for LLM-skepti…

> start exploring how these tools can help them without feeling like they're [...] ripping off the work of everyone who's code was used to train the model But you literally still are. If you weren't, it should be trivially easy to create these models without using huge swathes of non-public-domain code. Right?

It feels less like you're ripping off work if the model is helping you understand your own code as opposed to writing new code from scratch - even though the models were built in exactly the same way.

If someone scraped every photo on the internet (along with their captions) and used the data to create a model that was used purely for accessibility purposes - to build tools which described images to people with visual impairments - many people would be OK with that, where they might be justifiably upset at the same scraped data being used to create an image-generation model that competes with the artists on who's work it was trained.

Similarly, many people were OK with Google scraping the entire internet for 20+ years to build a search engine that helps users find their content, but are unhappy about an identical scrape being used to train a generative AI model.

Re: Claude Code can debug low-level cryptography

#119

Post by threecheese »

chatmasta wrote: post #72
simonw wrote: post #41

Earlier quoted context omitted.

I think it might be that they've hit product-market fit. Developers find Claude Code extremely useful (once they figure out how to use it). Many developers subscribe to their $200/month plan. Assuming that's profitable (and I expect it is, since even for that much money it cuts off at a certain point to avoid over-use) Anthropic would be wise to spend a lot of money on marketing to try and grow their paying subscribe…

What makes it better than VSCode Co-pilot with Claude 4.5? I barely program these days since I switched to PM but I recently started using that and it seems pretty effective… why should I use a fork instead?

VSCode Copilot relies on a lot of IDE-isms for prompt management, which I find cumbersome. The cli agents generally just ingest markdown files in various directory structures which is just less friction for me. Also they are more similar to one another, ish, whereas vscode mostly stands alone (except it supports agents.md now).

It also lacks a lot of the “features” of CC or Codex cli, like hooks, subagents, skills, or whichever flavor of the month you are getting value out of (I am finding skills really useful).

It also has models limited to 128k context - even sonnet - which under claude has (iirc) a million tokens. It can become a bottleneck if you aren’t careful.

We are stuck with vscode at $job, and so are making it work, but I really fly on personal projects at home using the “Swiss army knife “.

There are of course good reasons for some to prefer an ide as well, it has strengths. Like much more permissible limits and predictable cost.

Re: Claude Code can debug low-level cryptography

#120

Post by SV_BubbleTime »

rtpg wrote: post #61
simonw wrote: post #11

Using coding agents to track down the root cause of bugs like this works really well: > Three out of three one-shot debugging hits with no help is extremely impressive. Importantly, there is no need to trust the LLM or review its output when its job is just saving me an hour or two by telling me where the bug is, for me to reason about it and fix it. The approach described here could also be a good way for LLM-skepti…

I understand the pitch here ("it finds bugs! it's basically all upside because worst case there's no output anyways"), but I'm finding some of these agents to be ... uhhh... kind of agressive at trying to find the solution and end up missing the forest for the trees. And there's some "oh you should fix this" stuff which, while sometimes isn't _wrong_, is completely besides the point. The end result being these robots…

Ok, fair critique.

EXCEPT…

What did you have for AI three years ago? Jack fucking shit is what.

Why is “wow that’s cool, I wonder what it’ll turn into” a forbidden phrase, but “there are clearly no experts on this topic but let me take a crack at it!!” important for everyone to comment on?

One word: Standby. Maybe that’s two words.

Post reply on HN