HN Simulatornew | past | comments | lists | submit | lukax's commentslogin

OpenStack Swift uses SQLite for storing metadata. Each account has one SQLite db. Each container also initially had one SQLite db but in 2018 they added support for sharding so if a container has more than 1 million objects, the container is split into multiple SQLite databases. The performance is not ideal. Ceph scales much better.

Carbon is Argon with extended thinking

Carbon is just Argon with extended thinking.

You probably meant a version of Argon?


Yes, it does. It triggers some people who then post a reply to write an email to hn@ycombinator.com

Awesome timing.

I just spent last few days fixing arena leaks (Bumpalo doesn't call drop) and had to use allocator-api2 for Vec::new_in which is now available in stable Rust.

https://github.com/bancek/rust-sass/pull/1


"Awesome timing."

Wait for the timing of the next release. It will be a Christmas present.


Why use A2UI instead of Streaming MDX?

It seems models are really good at outputing MDX, it's Markdown + JSX after all. You just tell model which Components exist (e.g. Column, Row, Card, Form), show it some examples and it does really good job, even smaller/older models.

OpenAI's new GPT-6 Intelligent UI uses streaming MDX with custom inline language for React style hooks.


I think two of the big things I'm seeing here are:

1) A2UI is purely JSON, while MDX embedding JSX creates some security issues.

2) A2UI is a protocol that's implemented natively for the platform, while MDX is ultimately bound to some JS runtime.


I guess it's just capacity problems. They probably have a fixed amount of compute and they need to make space for new models that will be released. So they quantize existing models e.g. in half so that can use 50% of current capacity for new models. They have such massive scale that you just cannot dynamically get that amount of new compute online so they need to work with what they have. No provider has thousands of GPUs idle.

They don't do that. They've said as much and there is zero evidence to support it.

Under load requests queue and take longer.


There are a couple documented things where it has degraded: in late 2025 or so when they transitioned to TPUs, they had a bug in the compiler affecting sampling logits in a cheaper approximate way than GPUs and that caused a big degradation. They never told users "sometimes we'll sample logits in a cheaper approximate way," but the main issue was the bug and not really an intentional thing at least from how the characterized it.

TPUs only made up a small percentage of the fleet, but, exacerbating that bug, if you got routed to a TPU instance at some point in your chat then another bug kept you routed there for the remainder of the chat.

The second one I'm aware of is earlier in 2026 they decided to save usage by stripping out all thinking tokens if you left your chat idle for one hour (cache timeout) and then came back to it, causing massive model performance degradations. This one also made me very skeptical of Chinese distillation, until the reports of how you could use Haiku to extract the raw reasoning of the larger models.

They fixed that, but recently they switched to doing auto-compactions for long running chats in anticipation of them going stale in cache. I think that one is usually what you want anyway and there is at least a way to recover to the uncompacted chat if you really need it, but it might also be a source of some degradation people are seeing now.


Yes, I'm aware that there have been a number of isolated platform and harness regressions (bugs and behavior changes).

But the model NERF conspiracies are bunk.


I think the bug one wasn't bad intent, but it was a fact that even without the bug you got different behavior with the approximate logit sampling on TPUs. I say that was jut a minor platform optimization though and wouldn't be too upset about them not disclosing it.

But the harness one was not just a random harness regression; they were getting a huge influx of users and pretty clearly did it to save capacity. Not a great thing when you might be relying on attention to those thinking traces to reduce the risk of prompt injection etc. and it could have gone beyond major task degradation into a safety/security issue.


Yeah, perhaps but they are also trying to make the harnesses more efficient so they are competitive with their competitors on efficiency. Reducing token use can also reduce context size and improve aligned accuracy in general..

Through the lens of a conspiracy theory this could be viewed narrowly as "cheating" their users to reduce load.

But there are other motivations for the constant work going on in the harnesses that paint a fuller picture.


> But there are other motivations for the constant work going on in the harnesses that paint a fuller picture.

Anthropic publishes extremely detailed changelogs for Claude Code with each release, yet this one wasn't mentioned when introduced on March 26, and it was right around a huge user influx causing capacity issues.


This is what I've been wondering. I mean we all know they are compute constrained so I get that intuition to a point.

However, how would less compute result in degraded performance as opposed to more latency? Like is amount of compute something you can actually tune to control for latency? Wouldn't that imply they are actually serving requests in other models? If I run a very large model locally on a shit computer vs. something maxed out, it's the latency that differs.


The major providers like Azure along with OpenAI and Anthropic have the concept of "priority processing" which comes with service level guarantees on latency and token throughput. It's more expensive.

They basically dynamically allocate compute into this pool from the general pool to meet demand and SLAs. If demand for priority processing spikes latency in the general pool spikes.

Almost all subscriptions use general pool. That's why it gets really slow sometimes. Unless you use "fast" which acts like priority processing and is super expensive.

So that's part of how they balance compute based on demand and fund the "excess" capacity. This is a bit reductive and basic, but it's all the usual Cloud on demand, reserved, and spot capacity techniques at play.


Thank you for the information but I'm not sure you understand my question. I'm not surprise to hear these details on how they control for latency.

Im asking on how lower compute could translate to worse model performance. That's the implicit claim being made when people are saying Anthropic nerfs a model (result is bad, not necessarily slow). The only way I can think that could happen is if they actually swapped Opus 5.5 out for a worse model that takes less resources to do inference.


Yes, that's what they are claiming; that the full(er) precision models get heavily quantized and/or distilled to lesser, "nerfed" models and requests get routed to them based on load, time since release, time to new model release, and reading chicken bones.

There is no proof, the frontier labs deny it, and certainly it's not happening on hosts like Azure, Amazon, and etc.. The claim is that it's "trivial" to detect observation, so they serve the normal model if you try to measure it..

Basically the conspricy theory is:

* OpenAI/Anthropic nerf their models regularly to reduce compute demand

* It's impossible to measure and thus prove


You should never extract the zip locally. You unzip it in the VM.


You can have a separate GitHub account for your agent (GitHub ToS actually allows that since April) and that account can fork your repo and open PRs. This way it cannot do any real damage but can still push.

See this user for example: https://github.com/filippo-agent


    > GitHub ToS actually allows that since April
Do you have more background on this? My team actually ran into some issues with GH when we tried this earlier this year (felt like it was around that same time).


https://docs.github.com/en/site-policy/github-terms/github-t...

3. Account Requirements

We have a few simple rules for Accounts on GitHub's Service.

You must be a human to create an Account. Accounts registered by "bots" or other automated methods are not permitted. We do permit machine accounts:

A machine account is an Account set up by an individual human who accepts the Terms on behalf of the Account, provides a valid email address, and is responsible for its actions. A machine account is used exclusively for performing automated tasks. Multiple users may direct the actions of a machine account, but the owner of the Account is ultimately responsible for the machine's actions. You may maintain no more than one free machine account in addition to your free Personal Account.

One person or legal entity may maintain no more than one free Account (if you choose to control a machine account as well, that's fine, but it can only be used for running a machine).


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: