Yea a surprisingly small number of people don’t know a git remote can literally be any folder outside of your tree, over almost any kind of connection.
I thought about doing a forge but realized that if I was the only one working on this stuff then I could do the same thing by setting my remote to a folder on my NAS.
This seems cool but also a gateway for RSI
Most users are still on Twitter.
The people that have moved to Misskey are mostly erotic artists cause Misskey has less “international interference” (read:lolicon rules)
I assume at some point most mastodon instances will block Misskey for the same reason they blocked the pixiv mastodon instance(pawoo).
I assume China will put somebody on the moon before Musk’s meme rocket gets there.
They’ve also blown their entire development budget and have received another billion dollars in development funds.
The sheer number of people looking at starship’s delays and cost overruns and not seeing the exact same issues SLS had with Boeing are kind of staggering.
It’s strange that Niantic is doing this when google has been doing the same thing with its own street map data.
Seems weird to make a competing product when (I thought) they are on good terms.
It’s interesting axios specifically calls out the Homicide vs Violent crime statistic reliability:
The big picture: Homicides are more straightforward to compare year-to-year from pre-2021 to the present because the criteria for classifying them have remained the same while police have changed their methods of recording other violent crimes. Beginning in 2021, the FBI and police departments started shifting to the National Incident-Based Reporting System (NIBRS) from the decades-old Summary Reporting System (SRS). That allowed law enforcement agencies to submit more details on crimes like aggravated assaults but resulted in reported surges in violent crime in cities like Chicago and Minneapolis.
I read a (weirdly antagonistically conspiracy-themed) article about the FBI recently having to revise their statistics in 2022: https://www.realclearinvestigations.com/articles/2024/10/16/stealth_edit_fbi_quietly_revises_violent_crime_stats_1065396.html
It makes me wonder if we need to go back and revisit the last few years of crime statistics since they switched reporting structure to get a better idea of what’s going on…
The fact they pulled ROCM support for older cards boggles the mind.
This is why I like playing games through proton on Linux and the steam deck. Being able to “pause” an unpausable game by hitting the power button is great.
Reminder that all these Chat-formatted LLMs are just text-completion engines trained on text formatted like a chat. You’re not having a conversation with it, it’s “completing” the chat history you’re providing it. By randomly(!) choosing the next text tokens that seems like they best fit the text provided.
If you don’t directly provide, in the chat history and/or the text completion prompt, the information you’re trying to retrieve, you’re essentially fishing for text in a sea of random text tokens that seems like it fits the question.
It will always complete the text, even if the tokens it chooses minimally fit the context, it chooses the best text it can but it will always complete the text.
This is how they work, and anything else is usually the company putting in a bunch of guide bumpers to reformat prompts into coaxing the models to respond in a “smarter” way (see GPT-4o and “chain of reasoning”)
All LLMs are text completion engines, no matter what fancy bells they tack on.
If your task is some kind of text completion or repetition of text provided in the prompt context LLMs perform wonderfully.
For everything else you are wading through territory you could probably do easier using other methods.
You don’t need to run the LLM on the device. It supports 13 different protocols for both text completion and chat completion.
There’s also RisuAI which has some nicer features like better integration with function calling. Sillytavern is working on Function calling more but it’s not quite there yet.
The nice thing about Sillytavern is that it also has plugins for Alltalk for TTS and ComfyUi/A1111 for image generation directly from the roleplay interface.
It also has support for RAG through upload of documents and web scraping and a shitload of other features it would take awhile to list here.
Mostly decking about.
Zelensky was elected in 2019 specifically on an anti-corruption platform and in the past 4 years even with the war going on he’s made (somewhat) steady progress towards that end.
Admittedly the pressure from the US to clean up the administration in exchange for weapons has given Zelensky a lot more political leeway to oust corrupted officials.
Also Statista is notorious for cherry-picking data and not presenting the whole story. If you dig deeper into The actual report you’ll see that Ukraine has been making steady gains against corruption since 2013. The organization even Specifically commends the country for making inroads into corruption:
Although it still scores low, war-torn Ukraine (33) is one of few significant improvers on the CPI, having gained eight points since 2013. The country has long struggled with systemic abuse of power, but has taken important steps to improve oversight and accountability.
What do you mean by personal package manager?
What do people use for command line utilities? The selection on flatpak is a bit sparse
It depends on whether those dependencies are shared with other programs.
Somebody hit the token chain jackpot
Bluesky is not great, but it’s at least (for now) a better platform than X and the AT protocol is actually very well written. (For instance having a moderation service separated from the service that provides the posts I think is a hands-down better way to handle it than most ActivityPub servers having their admins handle all incoming and outgoing moderation)
Bluesky federation is just now getting started so it’ll be interesting to see if it goes anywhere/where it goes.
Seeing as how the full unquantized FP16 for Llama 3.1 405B requires around a terabyte of VRAM (16 bits per parameter + context), I’d say way more than several.