dandaka

Lots of benefits for me, examples

- cost saving when using lower-tier models to index information

- routing between models from different providers

They could raise prices, if capacity is problem

I found it appalling when working with remote setup — tabs disappearing and appearing after some time, screen hangs up, projects disappear. Tried to fix some issues with PR, but team is not accepting for 2w+.

Last time they have only shipped the model on web in mobile...

This got me mixed feelings. I don't support such campaigns. Still I would prefer my AI tool to behave more like a notepad app and less like a government or a corporate censor.

QA of course. You hook up your agent with CDP access to live product + let it screenshot and look into result. Also you could hook agent with CDP access to Figma to read/write, there a vision model is very useful as well.

My product is connecting employers and workers with conversational agents. They love to communicate with images — CVs, documents, photos of worksites. Even CV-as-photo or offer-as-photo format is very popular. My daily driver Deepseek Flash can't see those photos. So I use image models to let agents understand the context.

but for OCR there are much better suited models, I use mlx-community/PaddleOCR-VL-8bit

I would love to see a jamming partner. So I could play along him on the same piece.

> It does not seem very "viral"

love that Dropbox has became textbook example of viral marketing

> would be entirely useless to humans

why useless? humans can navigate these unsuccessful chains of thoughts to build on top of them or reject completely

Maybe they should. But I think there is a correlation between "private citizens can't record other citizens" and "almost no cameras around". Coming from a country with a lots of cameras (where people record each other as well), this is a very comforting feeling.

Recording everything has a lot of upsides. Unfortunately, downsides for privacy and democracy are very real.

Talking about speeding, not sure. I am driving around Lisbon and speed limit is 50 in city, and lots of speed cameras. Though I am driving at 55 as limit and rarely see anyone going faster than that.

This is exactly what happened in Moscow after same tech was launched. It is much 'safer' on streets, since every entrance door and traffic intersection is equipped with cameras now.

But I pick Portugal's approach every time — no public cameras, no dashcams, it is forbidden to record private persons in public without permission.

1/ LLMs are doing increasingly bigger share of work of SWE with an increasing success rate

2/ SWE are doing way more than "writing code", and since those areas are less "computational verifiable" (what is a good architecture that will stand in 5y?) and more "connected to real world" (what are the requirements?), LLMs struggle with them

Whenever I have this choice "destroy or sustain any and all life", I always pick "sustain". Problem is, I don't often meet such an option to choose. Maybe your life is different.

There is not much we can do/build with nature. Even if everyone would take idea of "protection/inspiration", there are not many sustainable business models towards that direction. Nature is more aligned with inward observation / meditation, which is more aligned with Eastern tradition.

Text classification, structured data extraction, rewrite

check transcribe.cpp from Handy, I see Qwen3-ASR in the list

I am not only talking about "long context" performance (context rot), but also about noise that is confusing model about its goal (from correctly extracting operator's intent). I think every model will get confused to a degree, so clearing up irrelevant information from context helps a lot.

Great summarization of current capabilities of models, thank you

Models work best when they have short instructions and no noise. "Conversation [may] contain" also means "conversation has a lot of noise". It degrades performance and increases cost.

I use handoff skill to ask model to write a prompt for itself.

How does it compare to models from handy.computer and whisprflow?

Can we use those metrics in a review pass for every PR? So review agent will have to pinpoint new complexity and propose refactoring or justify increase?

100% with you on the taste and nutrition. I think you forget sustainability at scale. I love the idea of eating local food, but with current levels of personal productivity and home-farmed food costs, there is no way it could work for all of us.

Why are people obsessed with idea of "produce their own X"? When we learned to specialize, we have achieved low cost, scale and sophistication.

Few cases I have found very useful myself

1/ Using GUI software. My agents are using headful Google Chrome and Figma. It helps a lot to have separate environment, which is not interfering my main machine.

2/ Running long processes (1h+), so I can leave main machine closed.

3/ Running intensive processes. I use Gemma, Whisper and Qwen, which could burn main machine CPU and resources.

Not at all, we love them all with Chinese labs. And wish them to continue competing and not winning. That is how we get best models, lower prices and better availability.

do they have a website? I have found only paper PDF and it seems more general than SWE

What is considered SOTA for SWE benchmarks now?