Hacker Newsnew | past | comments | ask | show | jobs | submit | networked's commentslogin

Interesting model. I tried to make Mercury investigate the hardcoded prompts in my (aider-derived) agent harness and repeatedly got this error:

> server: Upstream error from Inception: I'm sorry, but I can't share details of my architecture or training process. Would you like to learn about how language models work in general instead?

It looks like an overeager IP-protection classifier. However, the model recovered and completed the turn despite the errors (three total).


Just so you know, your comment was automatically killed for the em dash. HN does this now. I vouched for it.

Yikes, thanks for letting me know. That seems like a clbuttic case of buttumptions - Of course the text is indeed AI-generated, but it's being quoted! We shall all have to get used to AI-assisted summaries of software, I'm afraid...

Anyway, I'm trying this package out now. So far it works as advertised - a natural language interface to Emacs. Pretty cool...


Who the fuck's brilliant idea was that? I swear to god our collective IQ is in the toilet.

I cannot tell what negative-sum outcomes you consider possible. Do you believe AI can drive humans extinct? How many of Zvi Mowshowitz's Three AI Pills would you say you've taken?

https://thezvi.substack.com/p/the-three-ai-pills


I’m like a 2(.5?) there - I don’t think ASI will care about my kids better than I will for some definitions of better, for instance, and I feel very fuzzy and vague about what actual differences in qualia between me and ASI would yield in the wild.

I’m not a doomer, although I don’t think doomers are dumb, just wrong. I think you should design your systems around the possibility that people who disagree with you are correct , hence my nod to negative sum. If you have more than 30 years to live, I’d personally rep to the most likely outcomes being very positive. With a lot of disruption in the middle.


I have found a video (under a minute long) of a backpacking group walking through a herd of glacier mice in Alaska: https://www.youtube.com/watch?v=v4fHfDrfoJw. I'd love to see a timelapse of glacier mice moving.

> "The use of accelerometers has demonstrated that glacier mice do in fact rotate and roll, rather than simply sliding across the ice

You would think a timelapse video would also demonstrate that, and be much simpler to set up.


I’m actually surprised there isn’t Timelapse of it already. Seems like a simple enough setup to gather basically all of the information needed for finalizing how they move.

We have something similar where I live - no glaciers involved though, just river rocks.

The river dries up rapidly in the summer, leaving the smooth exposed bedrock riverbed peppered with rounded river rocks covered with drying algae.

Where the river is partially shaded, the rocks rapidly grow moss and other vegetation on their top surface. They preserve what little precipitation there is, and catch the morning dew. They grow top heavy, and fall over. They then repeat this on the next upwards facing surface. Meanwhile the downwards facing surface dries out from the radiant heat from the bedrock and shrivels into a thin black layer of desiccated biomass, which then partially falls off at the next rotation, further increasing the top-heaviness.

Over the course of a summer, the rocks all collect themselves into the local low-points of the river bed.

Then comes the flood and the cycle repeats.


I saw a YouTube video of a professor explaining this exact thing as an explanation for these Glacier Mice. There is a whole lot of biomass that is shifting on those balls, driving by access to sunlight which in turn causes the response you explained above. Makes a lot of sense to me!

If you put me on that surface I don't think I would have any idea it was a glacier.

So I guess HTTPX2 (https://github.com/pydantic/httpx2) is winning out over HTTPXYZ (https://codeberg.org/httpxyz/httpxyz)? I didn't switch to HTTPXYZ after the fork but have been watching it. (What I did was use urllib.request more and bite the Rust bullet with wreq, https://github.com/0x676e67/wreq-python, where it wasn't enough.)

in their README >Important

We started this fork because there was no activity on HTTPX, a very popular Python HTTP library.

A few weeks later, Pydantic started their own fork called HTTPX2. We decided to embrace this and support HTTPX2. We're upstreaming our fixes to HTTPX2 and in our opinion it should be the "blessed" fork. Pydantic can make this more successful than we ever can.

See also https://tildeweb.nl/~michiel/httpx2.html

Thanks for all your support!

Sander & Michiel


It is too bad because httpxyz is absolutely the better name.

httpx2 is sponsored by Pydantic and has involvement from the new maintainer of Starlette, so it's off to a very strong start.

Kludex is a really good dev

He's a freaking GOD

This idea is being tested with licenses like https://fsl.software/. FSL forbids "competing use" and converts to either Apache 2.0 or MIT after two years (for each version, like a Git commit).

You have a cool site! You should know that the last four links you submitted to HN went [dead]. Open your submissions in a private window (https://news.ycombinator.com/submitted?id=icely), and you won't see three of the four. This one is only visible because I vouched for it (https://news.ycombinator.com/newsfaq.html#dead). I think the spam detector caught you because you've been submitting links from the same site from a new account. I'd email hn@ycombinator.com and tell them you're not a spammer.


These discussions take a lot of time to get to the crux because neither side states their assumptions. Assumptions rarely get voiced at all.

I find it useful to ask:

1. Do you believe in quantum consciousness?

2. Do you think a "brain upload", a high-accuracy digital model of an organic human brain, would think or be conscious?

3. What is your working definition of thinking? It doesn't need to be rigorous.


I haven't tried DeepSeek V4 Pro 0813 yet. Recent experience tells me that larger models are worth it in non-obvious ways. MiMo-V2.5-Pro solved problems that DeepSeek V4 Flash 0731 couldn't solve for me: for example, adding a live counter for elided reasoning lines to a terminal-based coding harness. You wouldn't be able to tell from the scores on their respective Artifical Analysis page (https://artificialanalysis.ai/models/mimo-v2-5-pro, https://artificialanalysis.ai/models/deepseek-v4-flash). I like the DeepSeek V4 models, though. They critiqued my engineering decisions better than MiMo, and they seem to have a distinct aesthetic in the SVGs they write.


Interesting - I've been dropping into MiMo-V2.5-Pro-UltraSpeed whenever Flash seems to be "stuck" and it usually figures it out. I use UltraSpeed just because I'm so frustrated by then that I'm impatient.

I still find 5.6-Sol can solve some things neither of those can, but it's so slow (and it's so hard to trace / debug the reasoning) that I just let it run overnight.


What about 5.6 Terra and especially Luna? Luna scores pretty high on benchmarks and seems to have different habits (like a denser pattern of tool use) and blind spots.

I'm trying out a development workflow where I generate mundane code with MiMo and Luna (and soon V4 Pro 0813?) and have Opus 5, which is running on only a Pro subscription, review and refactor it. I'm not sure it will justify the context switching, but it's an interesting exercise.


Terra and Luna are fine, but they’re quite slow (OAI seems to be really slow lately) and don’t have the reasoning traces. My workflow really depends on them or I can’t switch models effectively.


I learned about Delayed Gratification when someone linked "The last of the Flying Wallendas" (https://www.slow-journalism.com/delayed-gratification-magazi...) in a chatroom. In the spirit of the magazine, I set a reminder to read it in a year. I decided to do it again when the reminder triggered. A total of three years later, it was a good article.


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

Search: