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 Cuba says it has run out of fuel, blames U.S. embargo
I’m curious how it’s legal to size a ship in international waters under any circumstances? We have a word for that - piracy.
mirzap··on Higher usage limits for Claude and a compute deal with SpaceX
I use Repoprompt's workflows for this. They are pretty good.
mirzap··on Zyphra releases the ZAYA1-8B MoE model optimized for intelligence density
Technical report: https://www.zyphra.com/zaya1-8b-technical-report

Announcement post: https://www.zyphra.com/post/zaya1-8b

mirzap··on Higher usage limits for Claude and a compute deal with SpaceX
Yeah, I do that too. Essentially, the system I described begins working on a task that is small enough and clearly defined. Each “slice” in a milestone usually have 5-10 subtasks (for instance, Slice E1 has P1...P6 subtasks). The orchestrator then receives the prompt to implement E1-P1.
mirzap··on Higher usage limits for Claude and a compute deal with SpaceX
I dedicate a significant amount of time to defining the precise actions that agents should perform (PRD/ADR). I break down the feature sets into Milestones and slices (tasks). These tasks are small, well-defined, and scoped. I have a prompt template that the “architect” agent prepares whenever I want to initiate a new feature. This ensures that the prompt structure remains consistent and standardized over time. The generated prompt is then pasted to the “orchestrator,” which performs context discovery (using Repoprompt) and finalizes the plan then proceeds to launch subagents to do the work.

Based on the size and complexity of the task, as well as any inter-task dependencies, the orchestrator deploys one or more subagents (sometimes 5 or 6 subagents) to work on these mini tasks. Once all tasks are completed, the orchestrator initiates verification and launches a review workflow. This workflow uses the original prompt, acceptance criteria, repository internal guidelines, and relevant skills to conduct a thorough review of the agents’ work.

Typically, there are one or two review iterations, during which the review agent identifies any issues. Sometimes, I may also notice issues and have to "steer" the orchestrator. The time required for a slice to complete ranges from 30 minutes to 4 or 5 hours, depending on its size, complexity, and the number of subtasks it contains.

Only if I run about 3 such orchestration in parallel I can reach hourly limit.

mirzap··on Higher usage limits for Claude and a compute deal with SpaceX
I'm on $200 Max plan
mirzap··on Higher usage limits for Claude and a compute deal with SpaceX
For me it's the opposite. I almost never hit hourly limit, but I hit weekly limit in about 5 days.
mirzap··on Higher usage limits for Claude and a compute deal with SpaceX
Doubling the five-hour rate limits is merely a marketing stunt if the weekly rates are not also doubled. It simply means that you can reach the weekly limits in three days instead of five.
mirzap··on DeepSeek v4
My thoughts exactly. I also believe that subscription services are profitable, and the talk about subsidies is just a way to extract higher profit margins from the API prices businesses pay.
mirzap··on Apple has removed most of the towns and villages in Lebanon from Apple maps?
Given the context, how it is not? These are not some random place on the map that have disappeared.
mirzap··on Apple has removed most of the towns and villages in Lebanon from Apple maps?
Those towns and villages will be rebuilt after the war. This is not excuse for what Apple did, it is justification for ethnic cleansing and occupation. Same as with Gaza City. It existed for 3500 years, it will be rebuilt and it will outlive the US/Israel for sure.
mirzap··on Show HN: Prawduct, a product development framework for Claude Code
It looks like AI-generated slop. Can't believe people would poison context with things like this.
mirzap··on RAM prices are plummeting after OpenAI failed to fulfill its commitment
https://wccftech.com/ddr5-prices-just-posted-their-first-dro...
mirzap··on OpenAI's latest repo has Claude as the third top contributor
Repo: https://github.com/openai/parameter-golf
mirzap··on Anthropic takes legal action against OpenCode
Why would they have that feature in claude code cli if it goes against the ToS? You can use Claude Code programatically. This is not the issue. The issue is that Anthropic wants to lock you in within their dev ecosystem (like Apple does). Simple as that.
mirzap··on Vite+ Is Now MIT
Official website: https://viteplus.dev/
mirzap··on I'm reluctant to verify my identity or age for any online services
And you shouldn’t verify. Many companies offering these identity verification services have ties to the intelligence networks of a country that shall not be named (similar to most VPN services that are supposedly there to protect your anonymity).
mirzap··on Anthropic officially bans using subscription auth for third party use
When you have 800–900 million active users, no matter how cheap it is, your costs will be in the billions.
mirzap··on Anthropic officially bans using subscription auth for third party use
For example, OpenAI’s agent (Codex) is open source, and you can use any harness you want with your OpenAI subscription. Anthropic keeps its tooling closed source and forbids using third-party tooling with a Claude subscription.
mirzap··on Anthropic officially bans using subscription auth for third party use
I’m not saying charging above marginal cost to fund R&D is weird. That’s how every R&D company works.

My point was simpler: they’re almost certainly not losing money on subscriptions because of inference. Inference is relatively cheap. And of course the big cost is training and ongoing R&D.

The real issue is the market they’re in. They’re competing with companies like Kimi and DeepSeek that also spend heavily on R&D but release strong models openly. That means anyone can run inference and customers can use it without paying for bundled research costs.

Training frontier models takes months, costs billions, and the model is outdated in six months. I just don’t see how a closed, subscription-only model reliably covers that in the long run, especially if you’re tightening ecosystem access at the same time.

mirzap··on Anthropic officially bans using subscription auth for third party use
They are not losing money on subscription plans. Inference is very cheap - just a few dollars per million tokens. What they’re trying to do is bundle R&D costs with inference so they can fund the training of the next generation of models.

Banning third-party tools has nothing to do with rate limits. They’re trying to position themselves as the Apple of AI companies -a walled garden. They may soon discover that screwing developers is not a good strategy.

They are not 10× better than Codex; on the contrary, in my opinion Codex produces much better code. Even Kimi K2.5 is a very capable model I find on par with Sonnet at least, very close to Opus. Forcing people to use ONLY a broken Claude Code UX with a subscription only ensures they loose advantage they had.

mirzap··on GLM-5: Targeting complex systems engineering and long-horizon agentic tasks
No, it's called hypocrisy https://en.wikipedia.org/wiki/Hypocrisy
mirzap··on GLM-5: Targeting complex systems engineering and long-horizon agentic tasks
Why are you all obsessed with this question when it comes to Chinese models? Here are some of the questions you should be asking Western governments and models instead: Who protects the pedophiles at the top of Western governments and corporations? How many people have been convicted in relation to the Epstein files? Who protects powerful politicians and Western oligarchs from pedophilia charges? Who did Epstein work for, and why (hint: it’s not Russia or China)?
mirzap··on Apple Creator Studio
How so? Apple's subscription cancellation is one click away, and you don't get overcharged when canceling.
mirzap··on Oh My Zsh adds bloat
I think convenience inherently comes with bloat. This is true for many things in life. Most people use cars simply to drive to work and back, yet they don’t use 90% of the features sold to them by salespeople.
mirzap··on Oh My Zsh adds bloat
Oh my god! How did I not know about this! Thank you! Are there any other useful tips like this one?
mirzap··on Oh My Zsh adds bloat
Oh thank you, I was looking for this. Will try it.
mirzap··on Phind Is Shutting Down
I’m really sad that Phind didn’t take off or at least get acquired by one of the larger players.
mirzap··on Oh My Zsh adds bloat
I tried using Starship, and it’s clearly faster than Oh My Zsh, but my issue was that I relied on some useful Oh My Zsh features that I didn’t know how to replicate in Starship. One of these is history filtering - for example, when I type source and press the up arrow, I only see previously run source commands, which makes it easy to find what I’m looking for. I tried to get this working in Starship, but had no luck.
mirzap··on Anthropic blocks third-party use of Claude Code subscriptions
I tend to think that their margins on API pricing are significantly higher. They likely gave up some of that margin to grow the Claude Code user base, though it probably still runs at a thin profit. Businesses are simply better customers than individuals and are willing to pay much more.
← PreviousPage 2 of 12Next →