Can't you feed that back into the same model? I believe most agentic pipelines just use a regular LLM to assess and review the answers from the previous step. At least that's what I've seen in these CoT examples. I believe training a model on rationality tests would be quite hard, as this requires understanding the reasoning, context, having the domain specific knowledge available... Wouldn't that require a very smart LLM? Or just the original one (R1) since that was trained on... well... reasoning? I'd just run the same R1 as "distillation" and tell it to come up with critique and give a final rating of the previous idea in machine redable format (JSON). After that you can feed it back again and have the LLM decide on two promising ideas to keep and follow. That'd implement the tree search. Though I'd argue this isn't Monte Carlo.
Is this a new thing? I thought that's how companies like Google and Meta operate for the last 10 years or so, minus a few poor people in underprivileged countries who have to sort the really bad stuff...
Does't seem too hard to me. I personally didn't. And it's kind of hard to track what happeded, with all the articles on DeepSeek.
I'd just take some prompt/agent framework like Langchain. That has Chain of Thought prompting built in for quite some time already. And then connect it to R1. That shoud do it. Maybe the thinking blocks need to be handled differently, idk.
Last time I checked, yunohost.org worked well. It either has the services you're interested in, or it doesn't... But it's really easy to get it running. Docker containers also usually work well. Though, you need some amount of technical knowledge. And I'd recommend to use something like docker-compose and not do everything manually... If you're a beginner, maybe just try YunoHost.
I've had the file system permissions wrong and that lead to all kinds of unrelated issues. But yeah the number of users and incoming posts are supposed to not increase unless you subscribe to users. Just adding instance URLs doesn't do anything. And it seems to federate new posts only, not the entire history of a user.
I think it's mainly the battery that isn't made for 24/7 charging. The other components are fine. And as a laptop is made not to waste electricity, it's efficient with the energy consumption, too. Just turn off the screen and it'll use as much as a Raspberry Pi or less...
Yeah, it really depends on the use-case. I've attached several external harddisks via USB to unsuitable hardware before. That kinda works, but isn't a good choice. But for some selfhosting of Bitwarden, home autiomation and calendar sync, an old laptop is more than enough. After that I bought an efficient mainboard, lots of RAM and built my own NAS for my files, and it does the other stuff as well.
I'm just flabbergasted by the fact that Email is now like 40+ years old(!) and nobody ever invented a worthy successor that saw some adoption. And we all know mail has some severe issues.
I also think it should be combined with messaging. But I mainly need it to be electronic letters. I want my bank statements, bills etc in that format. And be able to contact any business with it. Plus friends and family, the government etc and use it to coordinate things at work. And I think the internet made some advances since the 80s. We could include multimedia, sending large files, exchanging instant messages and maybe make it an identity provider so we also solve authentication, passwords and age verification on the internet. All of that isn't really hard and there is quite some demand for all of these issues being addressed by one uniform solution that is actually out there and usable.
But it seems to me not even the technology has been drafted/designed yet. So surely it can't be adopted when it doesn't even exist (yet).
That is correct. I've also observed Lemmy is a lot about commenting on the news. And I wish it was more about hobbies, conversations and interesting stuff. You're generally better off if you like Linux and AI, which are things I'm interested in and those are some pretty active topics... But I can see how that's not everyone's pair of shoes.
You're right. I finished reading their website only after commenting. Guess I'm not their target audience then. I prefer doing encryption on my device, as most of my regular mails come in unsecure and unencrypted anyways. And I guess we'd have to replace email in it's entirety with a more modern protocol to make it secure and usable. I mean my bank, internet service provider and almost everyone stopped sending me invoices, bills etc via mail already years ago. And it's not like they offer to do it via Tuta secure mail... So I guess all of these services are niche and limited by the trade-offs they need to invent.
My feed looks a lot different. And I've been blocking posts with these keywords for some time now, plus I've unsubscribed from world news and politics communities... But I have a lot of meme pictures in my feed, tech related stuff, casual talk and general questions, comics, animal pictures, computer hobby stuff, movies and TV... Maybe you're subscribed to the wrong communities?
Ah, alright. I was more thinking of getting a paid subscription anyways. At least the lower tier for 36€ a year seems fair. Though I wish they also offered chat and a few other services as well. Btw, if you're technically inclined... There are tools like imapsync which do this for any mail provider. Or with Evolution or Thunderbird as a mail client, I believe you can just select all your mails and drag and drop them into a different mailbox. That should get it done as well. (Edit: No IMAP)
Yeah, the first and last episode aren't even like the rest of the series, and include lots of weird stuff, the story which forms the framework... I've enjoyed them more than the middle part.
Good choice. I think people often invest too much into hardware and SBCs, when an old laptop does just fine. Just monitor the battery or remove it, if you run that for years and unsupervised in the broom closet.
Can't you feed that back into the same model? I believe most agentic pipelines just use a regular LLM to assess and review the answers from the previous step. At least that's what I've seen in these CoT examples. I believe training a model on rationality tests would be quite hard, as this requires understanding the reasoning, context, having the domain specific knowledge available... Wouldn't that require a very smart LLM? Or just the original one (R1) since that was trained on... well... reasoning? I'd just run the same R1 as "distillation" and tell it to come up with critique and give a final rating of the previous idea in machine redable format (JSON). After that you can feed it back again and have the LLM decide on two promising ideas to keep and follow. That'd implement the tree search. Though I'd argue this isn't Monte Carlo.