bayesianbot

I think you mean "sandbox" like what is in the Codex TUI. Parent poster meant a VM or container (or bubblewrap/firejail) that doesn't let the agent to edit files outside of specific paths or run dangerous commands, so you can turn the whole confirmation off.

That is their Python SDK version, not Python version

CloudFlare AI Gateway was quite convenient for me - I wanted to give my zeroclaw instance a limited budget to services like Image generation/Replicate/Fal.ai, which would mean for each service I'd have to run my own proxy that stores the keys and cuts off the real calls if we go over the limits. Easy to do but still extra thing to build and maintain.

Instead I put my API keys to cloudflare, set limits, and gave the agent the CloudFlare token, and in minutes it could contact tens of services.

edit: not to mention instead of loading balance to each service I could just keep balance on cloudflare that covers them all

I know that's the way omp is often introduced but I don't think it's fair for either omp or pi - after lately switching my usage more towards omp I'd say by this point they're almost more different than Pi and Codex. And I like them both.

Interesting idea, but at the same time it is just so cheap that you can just run it with API pricing. I sometimes do even if I have available usage that I'm going to cap so I'll save it for bigger models

Cache reads priced the same as Opus 5.5? So there won't be that much price difference in agentic coding. Or is that a mistake in the table, that seems quite weird

I think gpt 5.6 family also dropped pricing but didn't give any more usage for the subscriptions. Maybe it's a way to silently lower the value given to subscriptions while keeping API pricing competitive

Wow those cache reads are quite reasonable - I think that's equal to 5.6 Terra. I might have to try Claude again after years of being priced out of it

1M cached tokens on deepseek is $0.006, the big labs can't sell anywhere close to this, they have funders expecting returns and huge overhead.

btw I've had a ton of fun with the new deepseek today, I was waiting for my OpenAI 5h limit reset and decided to give it some problems for fun, got pretty great results. Tried some harder problems and still got great results. I don't expect it to be Sol class or anything but I really didn't expect it to be anywhere near this good so we'll see where it ends up. And it's really fun throwing crazy amount of tokens at the wall for ~free instead of watching the subscription limits tick closer while your agents churn away.

This clearly is Codex-specific, not as much a feature of the model (though obviously they probably have trained it to be great at working with their own tools). Sounds somewhat similar to pi-observational-memory I'm using with Pi.

Maybe you could try go2rtc[0]? It can ingest multiple stream formats and output them in a different one.

Also has some other nice features, like I use it to ingest one RTSP stream and share that to multiple applications, and to push audio out to my Tapo camera

[0] https://github.com/AlexxIT/go2rtc/

Are they related? The twitch stream definitely didn't have nudes while I was watching (while the other link seems to be ~constant nudity). Also the twitch description was something like "With new/fixed content moderation" or something similar (so might be right that they've had moderation problems before)

Just saw it for few minutes and then it dropped offline. I wonder what terms they violated

And now you got 90% of the people saying the model sucks as it timed out while serving them

IIRC GPT 3 was priced at per 1k tokens, had to check, the biggest GPT 3 model from OpenAI was $0.06/1k, so $60 / 1M. gpt-3.5-turbo was the first model after ChatGPT and that was $2 / 1M. And no caching. So not really in the same ballpark

I think for me customizing my setup isn't time away from "real tasks" - I do the customization when I don't feel like doing the work or I'm doing some background thinking on how to proceed in the real task. It's an interesting way to keep my mind warmed up for programming through times when I'm not in the mood, and I'd guess 80-90% of the customization is waste of time, but the ones that I end up sticking with can be really useful and save time/effort dozens of times per day

DeepSeek's own API

That is exciting!

I don't understand how DeepSeek can be so cheap with their cache pricing - ~0.003 usd / 1Mtok. 100x less than Kimi K3, or similar numbers against pretty much any other decently sized model to my knowledge. I've been using it whenever possible as even longer agent sessions cost few cents.

And imo slskd is fantastic client for soulseek, I just found it few weeks back

I come from Finland where (in my childhood) we used to say that education in Finland is the best in the world and there's really no racism or corruption.

I'm not gonna say the claims used to be true, but at least they used to be much closer to truth then than today. And I've long felt that having these kind of national beliefs can cause people/voters to neglect those things - e.g. all the bad claims about education can be just brushed off with "well we have the best education here so no worries!" - and few decades later they're not even close to true anymore.

Reminds me of a saying in poker, when many good players changed their style radically when playing against worse players and started playing way too bad starting hands (thinking they can just outplay them on the later betting rounds), something to the effect "It isn't enough that you know how to play better than your opponents, you have to actually do it"

Once again I have to let people know about glide[0], fork of Firefox. I know a lot of people here like keyboard-driven and scriptable software, and glide shines at both. Keyboard control in it is so so much better than the extensions for Chrome or Firefox, and I'm quite happy how easily the API allows me do my own customizations. (as an example when I open HN in new tab with my bind ,sn , if I already have HN front page open it will refresh it, focus it and move it to the tab index where new tab would otherwise have opened. Really simple stuff in glide, not so much in others)

[0[ https://github.com/glide-browser/glide

My basis for critical thinking is like 80% scientific method, splitting arguments in the smallest possible pieces to help avoid bias, testing them thoroughly, trying to destroy my own arguments until I can't.

In school the scientific method, maybe the most important invention of human kind, was pretty much skimmed over on some science classes on 5th-6th grade. People didn't learn how to apply it and do their own tests in everyday life, at all. Didn't even understand they should, it was just a part of science (subject that nobody cared about anyways).

Even those essays and other projects it's crazy how much weight was put on "books great, reliable magazines ok, tabloids bad, internet mostly bad but .fi domain is the most reliable.." and later down the road "wikipedia bad". That does not help much in current media landscape, where you can buy a book to confirm whatever you want to believe.

I learnt my own critical thinking skills from scientific books and my own curious neurodivergent mind that was obsessed with being as correct as I could. I saw in real time how lacking it was in school, how it affected my peers, and thought this is gonna be a problem in the future, and here we are.

> In reality, people don't learn those critical thinking skills because they ignore school. Their parents taught them that education is liberal brainwashing and that it shouldn't be trusted. Their experience is that school is useless because that's what parents and media told them so they didn't pay attention and now as adults they blame the schools and more media insists that it is the schools fault.

This was absolutely not something that meaningful in my community. Yes there were people like that, but they were just a tiny slice of all the people that didn't really learn to think critically but would probably have wanted to if we just taught them.

I'd call what was taught in school "vibe critical thinking", it seems good on the surface but it really is not even close to good enough. Sure if you get to university it gets better, but that doesn't help with majority of voters.

I don't really know much about education, but even having something like weekly session where we'd talk about how we applied scientific method and other tools in our life to test our claims, and would have judged the tests/methods used by others to see the problems in them to improve, would probably have mattered greatly. At least people would see that you can and should do that in real life, while learning a good intuition on what makes a test good. Honestly people hearing me testing my claims thoroughly made everyone look at me like I'm a crazy person obsessed with the thing I'm trying to prove, when I just understood that if I don't do that my beliefs are worthless.

But you don't need to convince every voter, just give a nudge to the right direction. Also, a ton of people are falling for really easy questions, like "is human caused climate change true", there really isn't much propaganda should matter in those cases if people had the tools to find even the simplest, most clear facts.

But I'm not even sure if it was possible - lots of people seem hellbent on believing what they want, I've thought that probably comes from evolution from time when we didn't really have ways to study reality that well, and group cohesion was more important than individual thinking. But we should have at least tried. It was quite clear people would get corrupted when they're sent out of school with a bag of knowledge but no real ways (or will) to really judge any future information

The mistake was not spending the first years of school teaching people to care about what is true, and also the tools for critical thinking to figure out what is real and how sure we are about it.

I thought that already when I was a kid and saw that people do not care at all if their beliefs are correct, and because of that don't care about learning any critical thinking skills, but I wasn't convinced and thought maybe adults just know something I don't, but today it is clear the same problem is even worse and the most important factor to why we are so screwed up as a species

It's not completely unrealistic angle, you could pwn the speaker when someone is traveling with it in public and then exfiltrate data when it's plugged in a secure environment and you can't connect anymore

You can modify your regex to only match when it's not a shortened url - then the short one will redirect to the real www.reddit.com address, before the redirect matches.

(Don't have the correct regex on hand right now, as I changed browsers and decided to use Old reddit redirect extension instead of scripting, but it worked in my previous browser)

LiteLLM can serve OpenAI API endpoint IIRC and proxy that to other providers like DeepSeek, should work with Codex

Yep, if I use street name it attaches to wrong part of the street, if I use a place next to the street it attaches to the place, not the street. Couldn't figure out how to drag the origin point of the arrow

Feels quite clear Donut doesn't have much - no meaningful new tests released for many weeks already and some executive of Nordic Nano sued Donut Lab and said their claims were misleading.

I haven't really followed that closely myself, but I've noticed the people who I saw defending Donut before have gone really quiet about it lately.

Tried it again few days ago. I kinda get that currently you can only use AI on Helix through LSP, but on top of that it does not have auto-refreshing files when changed outside - makes it really hard to work with external AIs, as I'm just constantly worrying if I'm editing a stale file.