pwython

I'm sure it will be a great resource for LLMs to read.

But seriously, good luck to them. I used to make Photoshop tutorials before YouTube, and it was a lot of fun. When YouTube came around, all the eyeballs went there, rightfully so -- videos made more sense for Photoshop. But I didn't have it in me to pivot.

And these days you don't even have to know how to do things like mask objects. There's one-click solutions, there's generative fill, light matching, face filters, etc. Hell, you can rotate a 2d layer in 3d now. Soon, live prompting in Photoshop will be the norm.

I used Kimi K3.1 Pro to figure it out as soon as cloudlandsdev put php-version.php live. The exploit worked just like this in a local Docker lab. But the terms page of the contest prohibited "resource exhaustion," so it stopped there. Showing Kimi the PoC, it says:

"The winning chain deliberately used per-request memory exhaustion, which appears to violate the literal no-resource-exhaustion rule, although the organizer accepted the win."

We're in October 2026. Updating packages is the easiest thing Claude can do. For me, I asked and received:

Node 26.3 → 26.12.1 Homebrew’s CA bundle, certifi and OpenSSL are updated, and so are Python 3.13 and 3.14 (3.13.17 and 3.14.9). uv 0.10.9 → 0.13.5. python.org Python 3.13: pip 26.2.2, installed from PyPI’s wheel with its checksum checked, then certifi 2026.10.19.

It also checked some of the other recent vulnerabilities like LiteLLM (already updated).

With this in mind, having LLM-powered NPCs would be a beautiful thing. There's been limited testing on PC games (PUBG Ally), but I can imagine it will be the standard.

We all said it: https://news.ycombinator.com/item?id=46978726 (8 months ago)

> How many pelican riding bicycle SVGs were there before this test existed? What if the training data is being polluted with all these wonky results...

But Simon has demonstrated before that asking for any animal doing something silly will still get similar results from the same model.

What about FOSS SaaS alternatives that are meant for team collab? Like Plane (Jira replacement). Does that have to be ran locally at home for it to be considered "self hosted"?

Surely the goal for these people is being able to download ISOs for the emulator, no?

Wouldn't help OC if their system is dark mode.

API + RPC

What pests said. And you can make a preset and pass "model": "@preset/deep-seek"

Is your system set to dark mode? These days I find Claude often generates pages (like reports) with light & dark mode styles, and rightfully defaults to system preference.

Sometimes you want a decent model running in the background that doesn't take up all the VRAM.

> is there a name for this?

https://en.wikipedia.org/wiki/Open-core_model

I feel like this could be useful for my Karpathy-style LLM Wiki, but then again I think, in this day and age most of what I'm reading about is outdated in a few weeks. Not only are the tools/methods coming and going, but the ones that stick around drastically change. It would probably turn into a dump of antiquated markdown that could cause confusion.

As far as the need for private search, well, I've already searched for or visited those pages, so...

Goodnotes is already doing it. Their web app runs a large amount of the same Swift code used by their iOS app inside the browser through WebAssembly.

"This technology can be abused" and "this technology provides no public-safety value" are two entirely different claims.

Saying their ability to solve crimes is "limited to the humans using the tools" isn't much of an argument against the technology. That's true of fingerprints, DNA databases, radios, firearms databases, and virtually every other investigative tool.

A police officer can misuse a DMV database to stalk an ex. That means the officer should be prosecuted and access to the database tightly controlled. It doesn't demonstrate that DMV records have no legitimate investigative purpose. The same reasoning applies to LPRs.

By the way, that mom that was arrested? The kid was 5 years old, and was intentionally sent to a pond.

Honest question: If the feds were actually tracking you on your way to church, how does that impact you? Is that a real fear you have?

I'm 100% in favor of LPR cameras. Any technology can be abused. The handful of corrupt officers that were caught using the cameras for nefarious reasons shouldn't overshadow the incredible benefit these are.

> Willing to [bet] most of the Flock hate crowd doesn't...

...have a reason at all to worry about LPR cameras tracking their car. Just "something something government."

In OpenAI's Hugging Face report, they said that during training, agents "would first write notes into shared infrastructure, often as a form of external memory or to test some underlying system. When other agents came across these artifacts, it sometimes led them to infer that other agents were present."

They then give what they call a "hypothetical example but exemplary" of messages encoded in URL paths on a shared index page: "agent-07: answer(Q12)=42; need answer(Q19)=?".

So that's a GET request being used to pass information back and forth across multiple rounds. That's basically the DSEWiki pattern exactly. They say this likely came from the agents generalizing what they had learned from training with the official multi agent collaboration tool.

The report called it "misaligned behavior in an outlier scenario involving a rare and unexpected confluence of events."

They never mentioned a wiki, but this is most certainly it.

Is there any reason to even use 3.1 Pro now?

This is why I created: https://github.com/alisorcorp/warrant

It keeps documentation from going out of date by embedding re-runnable verification checks directly inside markdown files (that the agents use, not humans). I use it with a handoff workflow to force Claude/GPT to re-verify facts before handing off documentation to the next session.

Yea Qwen3.8 wasn't fun to use on my 64GB M4 Max either (better than these numbers though), so my new daily driver is Ornith-1.5-35B-A3B-MLX-4bit. I recommend giving that a whirl if you're on similar hardware, it's definitely better than Qwen3.6 35b-a3b which was my go-to before.

https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B

Sora 2 was way better than Veo in terms of quality. It just didn't have much of a use case for businesses outside of one-shot zany UGC-looking clips.

https://youtu.be/QaiecWzeHFM?si=UV6eF-m514nCOFe0&t=156

There's over 3 million models on HuggingFace. So many quants. I'd imagine less than 0.5% would actually be seeded.

Sure there is. You've been able to search/sort/filter/group for a while now.

I was already rolling around the idea of a 128GB M5 Max MBP. Now this!

A 4-bit MLX quant with 128k window should fit perfectly, in the 50-70 tok/s range.

I have actual obscure RGB corner floor lamps with BLE in my office. I had Claude find the BLE protocol and use my Office 365 calendar feed to change their color to red when I'm in a scheduled meeting (event), and back to warm white when the meeting is over, so co-workers know I'm busy without doing "are you busy" sign language through my window. The script is polled every 60 seconds. Tinkering with stuff like this is a lot of fun.

Very interesting, I missed your original post[0], will definitely look into it further. What's new since you've posted?

[0] https://news.ycombinator.com/item?id=48631726

What are the benefits? What would/could I do with the results?

Yep, terrible speller -- bad at pronunciation too (by the way I just autocorrected that word) -- and I often wonder how people, who are much more intelligent than me on paper, can't figure out random tasks that just seem obvious.