gardnr

It's hard to say what their motivation is. The Chinese firms seem to be working hard to commoditize intelligence which may be the most effective way to debase American frontier labs. And yeah: it also happens to be really good for humanity.

No GH contributors. New HN account. Vibe coded in 5 commits over the last two week.

Interesting choice of model:

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

2 weeks ago

---

Opus 4.5 was introduced in November of 2025 (8 months ago)

Opus 4.6 was introduced in February of 2026 (5 months ago)

Opus 4.7 was introduced in April of 2026 (3 months ago)

Opus 4.8 was introduced in May of 2026 (2 months ago)

From: https://github.com/Planwright/planwright/commit/c746cdcaf7e6...

This is like Google Ultra for personal accounts. I signed up to see what it was like and then assumed I would be able to disable training on my data as a paid customer. The only way to disable training on paid personal accounts is to disable history (no chat logs) which makes the service much less useful for me.

For Google Workspace accounts that use the Ultra plan you can disable training while retaining history. I didn't bother signing up again. It is user-hostile.

I too was nervous until I clicked through and had a read.

It has a JSON mode that you need to enable in settings and then you can create a simple python script to interact with it or have the agent use `curl` and `jq` to interact with it.

It's at the bottom of this page: https://docs.searxng.org/admin/settings/settings_search.html

Different bubble than the one I was hoping for.

This appears to be different than the recent "Speculative Pipeline Decoding" paper: https://arxiv.org/abs/2605.30852

Dwarkesh Patel has AI/ML guests on his podcast. BoorishBears may have been referring to the Jensen Huang episode where they discuss TPUs: https://youtu.be/Hrbq66XqtCo?t=982

The training and deployment of LongCat-2.0 are built on large-scale clusters of tens of thousands of AI ASIC superpods. Compared to the mature Nvidia GPU ecosystem, the supporting software community is still less developed. We have therefore put significant effort into building a stable, secure, and scalable infrastructure.

This is the real news story. It looks like they may have used Huawei Ascend 910C chips: https://nitter.net/teortaxesTex/status/2071708141037781407#m

It really depends on your personal perspective on life. Consider the two statement and which one aligns more with yourself:

1. "It's better to regret something that you have done instead of something that you haven't done."

2. "I want to retire comfortably with a high level of confidence."

People in my life always advised me to follow the highest paying job. I'm not sure that was the right advice given my circumstances. It's really up to you.

This really sucks. Given how bad their regexes were in their leaked code, I am guessing this will get triggered all the time when I am fine tuning a model or doing work with datasets. The fact that there's no feedback means I can't trust the tools.

How snobbish was the CEO acting?

Just in case anyone is thinking about trying it: 25g seems pretty high. It’s worth it to review what that means for the rest of your body before starting this regime. Kidneys are really useful organs to have working.

Are you using Opus? Sonnet remains as useful as it was while Opus efficacy and token burn rate has soured over the last 4 months.

I have a suspicion most Libertarians are actually something else but who haven't realized it yet. When you really dig into it: idea of private property (as in owning land and charging people rent for using it) is akin to slavery (owning people and extracting labor from them). The pre-colonial indigenous structures of managing societies were much more aligned with my internal values but they are poorly defined in modern vernacular and we don't have a good "common vocabulary" to talk about how we might want to do things in a different way.

This is straight outta 2023:

Agents aren't reliable; use workflows instead.

Here is one source that agrees: https://artificialanalysis.ai/models/capabilities/coding

Same throughput with much less heat. Not sure what that extra 175w is going towards but it's diminishing returns.

They keep adding “fancy” UI and hijacking standard browser behaviour that is infuriating on a daily basis.

Please consider a lofi version for people that want to select text without navigating to a different page.

Looks like it may not have a package manager like apt or dnf:

Can you please add wget, nano, $my_fav_app_omg_i_love_it to the root filesystem?

No, not likely.

I am guessing the way to use software not already in the image is to use `docker run`.

It can be damn near impossible to break them out of some loops once they've committed. Gotta trim the context back to before the behaviour started.

865 GB: I am going to need a bigger GPU.

If you buy a new MacBook Pro be prepared to ditch all of your duckhead adapters.

If you enjoy using a "Power Adapter Extension Cable" to extend your distance from the wall, then be prepared that you cannot buy one that fits the new prongs as Apple does not sell them.

I just got off the phone with their sales support and they recommended that I go to a hardware store and buy a regular extension cable.

I wouldn't label that arrogance. In my experience outside the USA, my GP has been unaware of new research / advice / guidelines published by the MoH. They generally respond to new info from reputable sources when you print it out and bring it to them.

I would expect specialists to be subscribed to journals and reading the latest articles in their field. When I saw a specialist at UCSF this was definitely the case; while my GP still has gaps where their current knowledge on a specific subject is from their time at med school.

An equivalence would be a front-end engineer being naive to the happenings on the Linux kernel mailing list. They could likely understand what's going on if they took the time to read it, but that is not their focus.

The GLM-4.7 model isn't that great. I was on their $200/month plan for a while. It was really hard to keep up with how fast it works. Going back to Claude seems like everything takes forever. GLM got much better in 5.1 but Cerebras still doesn't offer that yet (it's a bit heavier). I have a year of Z.ai that I got as a bargain and I use GLM-5.1 for some open source stuff but I am a bit nervous about sending data into their API.

It's like the author handed the copy to the editor who then added a new broken sentence after each original sentence that somehow jams "agents" in there.

The llama weights were leaked. It open sourced itself.

You are right though. Meta could have been in lockstep releasing ChatGPT features into some chat bot on Facebook.com but instead it seemed like their FAIR arm was hell bent on commoditising this stuff by publishing their research models before the Chinese companies took the lead in that.

It’s hard for me to be mad at FAIR even though I general disagree with the outcomes that Meta produce for their users.

It’s personal…

I was really excited until I realised that “personal” meant “owned by meta“.

I’m trying to decide is I find the doublespeak a bit offensive or not.

There is this 3rd party tracker: https://marginlab.ai/trackers/claude-code/

A genuinely good-for-the-world project. The data is really useful for science and for machine learning. You can export all the research-grade identifications of fungi to train a classifier; if that’s what you’re into.