bayesianbot

Just as FYI, for people currently using firefox or want to use firefox but found its keyboard control (or plugins like Vimperator) lacking, I really recommend glide[0] highly.

I've used qutebrowser for years as I feel the keyboard controlled web is much more convenient, and there hasn't been any reasonable competition to qutebrowser. The vim keyboard control plugins for chrome or firefox don't fit the bill for me, they feel slow, are often out of focus, and quite limited.

glide fixes all of those problems, supports firefox extensions and has a really powerful and approachable scripting API. It's alpha but feels quite ready, I've been running it a few weeks full time and loved the experience.

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

Something like LD2410 [0]. IIRC there's newer ones that report accurate position and even heart beat rate, but I've forgotten the names of those..

[0] https://dronebotworkshop.com/ld2410c-human-sensor/

I agree with the article and I hold 0 crypto right now. But I still think it's amazing that I can hold something limited, something I can exchange for real money, in my head, just based on math. Sure it is extremely inefficient database, and pretty much all the real value needs to be linked with real world banking, but it does have some really unique features that makes me sad that it (predictably) turned to just scams and speculation.

Edit: and the other feature I like is that I could just attach my code to the raw banking backend. People say that anyways everybody just uses exchanges, and that's true, but if you'd ever want to connect to banking backend, you'd get buried in paperwork. With crypto, you'd just run or connect to a node.

At the same time, the CO2 increase measured at Mauna Loa for 2024 was over 3.5ppm/yr, way up from the ~2.5ppm/yr seen previously this decade[0]

2025 State of the Climate report[1] said (on top of other horrible things)

> A dangerous hothouse Earth trajectory may now be more likely due to accelerated warming, self-reinforcing feedbacks, and tipping points.

I haven't seen hothouse earth mentioned in mainstream papers for a long time (decade+?), as it was deemed unlikely before.

Also The German Physics Society and the German Meteorological Society issued a joint statement warning about the possibility of 3 °C warming by the 2050s[2]

I am actually angry to people that they're irresponsible enough to vote for this without caring about others, but it feels like it was such a horrible timing for all this stupidity as well.

[0] https://www.carbonbrief.org/met-office-atmospheric-co2-rise-... [1] https://academic.oup.com/bioscience/advance-article/doi/10.1... [2] https://worldcrunch.com/focus/green-or-gone/global-warming-a...

I feel like democrats should make it clear that if there's still a fair election and they regain power, they'll go after both the corrupt people in this admin and entities buying favors. The current state can't be too good for the society, at least there should be a clear possible downside for being a part of it.

Not OP but

- Terminal search and focus (you can list kitty tabs and windows and get the window content from the socket, implementing a BM25 based search is quite easy)

- Giving the current terminal content for AI, so I can do things like run `ls` and then write "Rename the files (in some way)", and push the whole thing to LLM that replaces the command line without me having to write the full context

I even have a Codex session finder that uses codex session files to list and select the session I want, and then uses the kitty socket to find and focus the window which matches the session content

Trump also said he's gonna tell Australia's prime minister about the reporter, which is kinda nuts (and hilarious?)

Old track, but just hard to imagine what would have happened if Biden was asked about his corruption and answered like that. But it's hypothetical anyway, since no previous president would ever be rug-pulling crypto scams or selling watches and bibles.

I just can't believe how weekly, or sometimes daily, I share these wild stories and videos with some friends and they keep behaving like anything about this is normal. There are so many things that would make me go WTF even without the context of the constant grift it all comes with.

> SIDENOTE: If you want 2 way audio to work in frigate you must use the tapo:// go2rtc configuration for your main stream instead of the usual rtsp://. TP-Link are lazy and only implement 2 way audio on their own proprietary API.

Annoyingly when this is in use, I can't use ONVIF which seems like the only way to pan and tilt the camera using open tools. So if I want to use two way audio and also control the camera, I have to stop the process reading tapo:// stream, start onvif client and rotate, turn off onvif client and start streaming using tapo:// again

I definitely agree with all of those points. I just really prefer it completing steps and asking me if we should continue to next step rather than doing half of the step and telling me it's done. And the context degradation seems quite random - sometimes it hits way earlier, sometimes we go through crazy amount of tokens and it all works out.

I've been extremely impressed (and actually had quite a good time) with GPT-5 and Codex so far. It seems to handle long context well, does a great job researching the code, never leaves things half-done (with long tasks it may leave some steps for later, but it never does 50% of a step and then just randomly mock a function like Gemini used to), and gives me good suggestions if I'm trying to do something I shouldn't. And the Codex CLI also seems to be getting constant, meaningful updates.

I would never have bought one before but nowadays it could actually be useful. You could have Codex or Claude Code in your pocket, and every ~15min check the work and write a new prompt. Tablets are too big (for me) to constantly carry around for this, and phones annoyingly small for that use.

Usually you'd have some predefined widths, like 33%, 50% and 67%, and use a shortcut to cycle between them. And you can define window rules to start some applications with different width than your default.

edit: and as fellow niri user, I recommend people try it. I think it's one of the easiest tiling WMs to get into, it feels very natural within minutes.

So far I’ve just YOLO'd it. But if I install other software directly from git and the source isn’t fully reliable, I’ll usually at least check recent changes, or have codex take a look through the source, just like I read through PKGBUILDs when installing from AUR. It feels crazy that I then update LazyVim and suddenly pull in 150 new commits, some just minutes old, all with free access to my system.

Am I the only one who thinks the way plugins are updated in lazy.nvim (and probably others) is a bit insane? It seems to just pull the latest commits. Every time I update, I feel one rogue commit away from someone stealing my keys. It definitely feels like the riskiest thing I do on my system. Or have I misunderstood something?

This sounds strange? If I were to build this system (without really having time to think about it), I'd let the AI "build" the order, which would impose some hard limits - like not accepting 18 000 items. Then I'd have the user confirm it without any AI involvement, so you wouldn't end up with bacon in ice cream. This sounds like they just connected an AI directly to ordering functions and that's it?

Seeing this title, I really hoped they'd have a parameter to force the language for voice recognition. Being Finnish with a heavy rally-Finnish accent, the real-time modes quite often end up transcribing my English speech as Finnish text, and strangely, I get the correct Finnish for the words for what I spoke in English. It might not happen during the first sentence, but after a few queries and replies it usually does.

According to the OpenAI forums this is a common problem. I see they've addressed this in the post by prompting the model to stick to one language, but previously this didn’t work consistently, and in their Playground the newest `User transcript model` is still the same as before (`gpt-4o-transcribe`), so I don’t have high hopes. Must be hard to implement.

edit: Tried it again (with a prompt requesting English like always). By my 6th message it suddenly started transcribing to Finnish, and after that it became more common. Better than it used to be, but in many ways still useless. Though I'm sure it works better for people with lighter accents.

Had to look it up, here's a story about the missing report if anyone's interested: https://greens.org.au/news/media-release/if-minister-bowen-s...

And they included Flex pricing, which is 50% cheaper if you're willing to wait for the reply during periods of high load. But great pricing for agentic use with that cached token pricing, Flex or not.

I've made my own fork of Codex that always uses flex, or you can route agents through litellm and make it add the service_tier parameter. I haven't really seen native support for it anywhere.

I just saw another one, tuxagotchi[1], that seems similar but feeds on git commits to motivate the user.

[1] https://github.com/terpinedream/tuxagotchi

(just thought it's related, I haven't tried either)

OpenAI also has Flex processing[1] for o3. I've spent most of my time with Gemini 2.5, but lately been trying out a ton of o3 as it seems to work quite well and I get really cheap tokens (~95% of my agentic tokens are cached which is 75% discount and flex mode adds 50% for $0.25 / million input tokens)

[1] https://platform.openai.com/docs/guides/flex-processing?api-...

At least LazyVim[1] that I use does have visual help for key shortcuts - I can either press the first key (often prefix aka space) and get a popup of continuation keys, or use <space>sk to search keys.

[1] https://lazyvim.org

Honestly, I've used a fully set up Neovim for the past few years, and I recently tried Zed and its "edit prediction," which predicts what you're going to modify next. I was surprised by how nice that felt — instead of remembering the correct keys to surround a word or line with quotes, I could just type either quotation mark, and the edit prediction would instantly suggest that I could press Tab to jump to the location for the other quote and add it. And not only for surrounding quotes, it worked with everything similar with the same keys and workflow.

Still prefer my neovim, but it really made me realize how much cognitive load all the keyboard shortcuts and other features add, even if they feel like muscle memory at this point.

I feel like the optimal coding agent would do this automatically - collect and (sometimes) summarize the required parts of code, MCP responses, repo maps etc., then combine the results into a new message in a new 'chat' that would contain all the required parts and nothing else. It's basically what I already do with aider, and I feel the performance (in situations with a lot of context) is way better than any agentic / more automated workflow I've tried so far, but it is a lot of work.

Between (0, 2)s. Apparently the times are rounded down, so it could be :42.001 and :43.999, or :42.999 and :43.001

DeepSeek has gone a bit different route - they give automatic 75% discount between UTC 16:30-00:30

https://api-docs.deepseek.com/quick_start/pricing

Maybe not the best post to ask about this hehe, but what are the good open source LLM clients (and models) for this kind of usage?

Sometimes I feel like I would like to have random talks about stuff I really don't want to or have chance to with my friends, just random stuff, daily events and thoughts, and get a reply. Probably it would lead to nowhere and I'd give it up after few days, but you never know. But I've used extensively LLMs for coding, and feel like this use case would need quite different features (memory, voice conversation, maybe search of previous conversations so I could continue on a tangent we went on an hour or some days ago)

Might not (sadly) as directly affect the businesses but I hope we don't forget that his time in DOGE probably ends up causing ~14 million global deaths just from USAID: https://www.thelancet.com/journals/lancet/article/PIIS0140-6...

I thought democracy was doomed not long after I first learned about it as I had already seen how 'well' people check that their beliefs match with reality. I thought I was wrong (partly to keep my sanity), that there must be some system that keeps it stable. Nowadays it is clear I was wrong - the clown show of current global democracy is nothing like what I imagined. I thought it was going to be smart, greedy leaders with complex lies that are hard to check, but instead, people are being robbed and the processes keeping them alive and fed are being destroyed all in the open, there are mostly no complex lies or smart leaders involved.

It's hard to blame AI for something that was obvious three decades ago.

My first thought as well. But maybe at least people wanting to plug their apps to their AI forces developers to actually implement the interface, unlike APIs that are mostly unheard of in general population and thus not offered?