55
45

Terminal Trove showcases the best of the terminal, Discover a collection of CLI, TUI, and more developer tools at Terminal Trove.

6
submitted 6 months ago by daredevil@kbin.social to c/vgmusic@lemmy.world

イニシエノウタ/デボル · SQUARE ENIX MUSIC · 岡部 啓一 · MONACA

NieR Gestalt & NieR Replicant Original Soundtrack

Released on: 2010-04-21

117

On Monday, Mistral AI announced a new AI language model called Mixtral 8x7B, a "mixture of experts" (MoE) model with open weights that reportedly truly matches OpenAI's GPT-3.5 in performance—an achievement that has been claimed by others in the past but is being taken seriously by AI heavyweights such as OpenAI's Andrej Karpathy and Jim Fan. That means we're closer to having a ChatGPT-3.5-level AI assistant that can run freely and locally on our devices, given the right implementation.

Mistral, based in Paris and founded by Arthur Mensch, Guillaume Lample, and Timothée Lacroix, has seen a rapid rise in the AI space recently. It has been quickly raising venture capital to become a sort of French anti-OpenAI, championing smaller models with eye-catching performance. Most notably, Mistral's models run locally with open weights that can be downloaded and used with fewer restrictions than closed AI models from OpenAI, Anthropic, or Google. (In this context "weights" are the computer files that represent a trained neural network.)

Mixtral 8x7B can process a 32K token context window and works in French, German, Spanish, Italian, and English. It works much like ChatGPT in that it can assist with compositional tasks, analyze data, troubleshoot software, and write programs. Mistral claims that it outperforms Meta's much larger LLaMA 2 70B (70 billion parameter) large language model and that it matches or exceeds OpenAI's GPT-3.5 on certain benchmarks, as seen in the chart below.
A chart of Mixtral 8x7B performance vs. LLaMA 2 70B and GPT-3.5, provided by Mistral.

The speed at which open-weights AI models have caught up with OpenAI's top offering a year ago has taken many by surprise. Pietro Schirano, the founder of EverArt, wrote on X, "Just incredible. I am running Mistral 8x7B instruct at 27 tokens per second, completely locally thanks to @LMStudioAI. A model that scores better than GPT-3.5, locally. Imagine where we will be 1 year from now."

LexicaArt founder Sharif Shameem tweeted, "The Mixtral MoE model genuinely feels like an inflection point — a true GPT-3.5 level model that can run at 30 tokens/sec on an M1. Imagine all the products now possible when inference is 100% free and your data stays on your device." To which Andrej Karpathy replied, "Agree. It feels like the capability / reasoning power has made major strides, lagging behind is more the UI/UX of the whole thing, maybe some tool use finetuning, maybe some RAG databases, etc."

Mixture of experts

So what does mixture of experts mean? As this excellent Hugging Face guide explains, it refers to a machine-learning model architecture where a gate network routes input data to different specialized neural network components, known as "experts," for processing. The advantage of this is that it enables more efficient and scalable model training and inference, as only a subset of experts are activated for each input, reducing the computational load compared to monolithic models with equivalent parameter counts.

In layperson's terms, a MoE is like having a team of specialized workers (the "experts") in a factory, where a smart system (the "gate network") decides which worker is best suited to handle each specific task. This setup makes the whole process more efficient and faster, as each task is done by an expert in that area, and not every worker needs to be involved in every task, unlike in a traditional factory where every worker might have to do a bit of everything.

OpenAI has been rumored to use a MoE system with GPT-4, accounting for some of its performance. In the case of Mixtral 8x7B, the name implies that the model is a mixture of eight 7 billion-parameter neural networks, but as Karpathy pointed out in a tweet, the name is slightly misleading because, "it is not all 7B params that are being 8x'd, only the FeedForward blocks in the Transformer are 8x'd, everything else stays the same. Hence also why total number of params is not 56B but only 46.7B."

Mixtral is not the first "open" mixture of experts model, but it is notable for its relatively small size in parameter count and performance. It's out now, available on Hugging Face and BitTorrent under the Apache 2.0 license. People have been running it locally using an app called LM Studio. Also, Mistral began offering beta access to an API for three levels of Mistral models on Monday.

3
submitted 6 months ago* (last edited 6 months ago) by daredevil@kbin.social to c/tech@kbin.social

On Monday, Mistral AI announced a new AI language model called Mixtral 8x7B, a "mixture of experts" (MoE) model with open weights that reportedly truly matches OpenAI's GPT-3.5 in performance—an achievement that has been claimed by others in the past but is being taken seriously by AI heavyweights such as OpenAI's Andrej Karpathy and Jim Fan. That means we're closer to having a ChatGPT-3.5-level AI assistant that can run freely and locally on our devices, given the right implementation.

Mistral, based in Paris and founded by Arthur Mensch, Guillaume Lample, and Timothée Lacroix, has seen a rapid rise in the AI space recently. It has been quickly raising venture capital to become a sort of French anti-OpenAI, championing smaller models with eye-catching performance. Most notably, Mistral's models run locally with open weights that can be downloaded and used with fewer restrictions than closed AI models from OpenAI, Anthropic, or Google. (In this context "weights" are the computer files that represent a trained neural network.)

Mixtral 8x7B can process a 32K token context window and works in French, German, Spanish, Italian, and English. It works much like ChatGPT in that it can assist with compositional tasks, analyze data, troubleshoot software, and write programs. Mistral claims that it outperforms Meta's much larger LLaMA 2 70B (70 billion parameter) large language model and that it matches or exceeds OpenAI's GPT-3.5 on certain benchmarks, as seen in the chart below.

The speed at which open-weights AI models have caught up with OpenAI's top offering a year ago has taken many by surprise. Pietro Schirano, the founder of EverArt, wrote on X, "Just incredible. I am running Mistral 8x7B instruct at 27 tokens per second, completely locally thanks to @LMStudioAI. A model that scores better than GPT-3.5, locally. Imagine where we will be 1 year from now."

LexicaArt founder Sharif Shameem tweeted, "The Mixtral MoE model genuinely feels like an inflection point — a true GPT-3.5 level model that can run at 30 tokens/sec on an M1. Imagine all the products now possible when inference is 100% free and your data stays on your device." To which Andrej Karpathy replied, "Agree. It feels like the capability / reasoning power has made major strides, lagging behind is more the UI/UX of the whole thing, maybe some tool use finetuning, maybe some RAG databases, etc."

Mixture of experts

So what does mixture of experts mean? As this excellent Hugging Face guide explains, it refers to a machine-learning model architecture where a gate network routes input data to different specialized neural network components, known as "experts," for processing. The advantage of this is that it enables more efficient and scalable model training and inference, as only a subset of experts are activated for each input, reducing the computational load compared to monolithic models with equivalent parameter counts.

In layperson's terms, a MoE is like having a team of specialized workers (the "experts") in a factory, where a smart system (the "gate network") decides which worker is best suited to handle each specific task. This setup makes the whole process more efficient and faster, as each task is done by an expert in that area, and not every worker needs to be involved in every task, unlike in a traditional factory where every worker might have to do a bit of everything.

OpenAI has been rumored to use a MoE system with GPT-4, accounting for some of its performance. In the case of Mixtral 8x7B, the name implies that the model is a mixture of eight 7 billion-parameter neural networks, but as Karpathy pointed out in a tweet, the name is slightly misleading because, "it is not all 7B params that are being 8x'd, only the FeedForward blocks in the Transformer are 8x'd, everything else stays the same. Hence also why total number of params is not 56B but only 46.7B."

Mixtral is not the first "open" mixture of experts model, but it is notable for its relatively small size in parameter count and performance. It's out now, available on Hugging Face and BitTorrent under the Apache 2.0 license. People have been running it locally using an app called LM Studio. Also, Mistral began offering beta access to an API for three levels of Mistral models on Monday.

[-] daredevil@kbin.social 18 points 6 months ago

I'd imagine this will also be very problematic for non-celebrities from all sorts of backgrounds as well. The harassment potential is very concerning.

[-] daredevil@kbin.social 14 points 6 months ago* (last edited 6 months ago)

kbin has this -- the feature is called collections. https://kbin.social/magazines/collections

you can make public ones that others can follow, or private ones to make curated feeds for yourself.

[-] daredevil@kbin.social 33 points 6 months ago

Pandora's Box is already opened, unfortunately. The Streisand Effect is only going to make this worse.

[-] daredevil@kbin.social 51 points 6 months ago

What scares me more than the fact that this guy exists are how many people chose to and continue to support him

16
submitted 7 months ago* (last edited 7 months ago) by daredevil@kbin.social to c/vgmusic@lemmy.world
3
submitted 7 months ago by daredevil@kbin.social to c/gaming@kbin.social

Resident Evil 4 Remake has been crowned PlayStation Game of the Year at The Golden Joysticks 2023 powered by Intel.

Capcom's third Resident Evil remake was released in March of this year and took players back to rural Spain to confront the mysterious, and deadly, Los Illuminados cult - 18 years after we originally did on the PlayStation 2. Fans clearly loved revisiting the classic survival horror game as it managed to beat out other games in the category including Final Fantasy 16, Street Fighter 6, and Star Wars Jedi: Survivor.

The other Golden Joystick Awards 2023 nominees in this category can be found below:

  • Final Fantasy 16
  • Resident Evil 4 Remake (winner)
  • Street Fighter 6
  • Humanity
  • Armored Core 6: Fires of Rubicon
  • Star Wars Jedi: Survivor
2
submitted 7 months ago by daredevil@kbin.social to c/gaming@kbin.social

Final Fantasy 7 Rebirth has won the Most Wanted Game category at the Golden Joystick Awards 2023 powered by Intel.

Due in February of next year, Square Enix's much-anticipated follow-up marks the second part of a planned three-part modern-day reimagining of its 1997 source material.

Hot on the heels of 2020's Final Fantasy 7 Remake, Final Fantasy 7 Rebirth extends the legendary story beyond Midgar – with a recent trailer teasing familiar spots such as Cid's Rocket Town, Red XIII's Cosmo Canyon, and the indelible Gold Saucer theme park.

Add flashes of an introspective Sephiroth, Jenova, Junon Harbor and that thoroughfare-dominating parade, and it's easy to see why people are looking forward to this one, and, indeed, why it's come out on top of this year's Golden Joysticks' Most Wanted category.

Throw in the teasiest of Emerald Weapon teasers, and… yeah, February 29, 2024 really can't come soon enough. Full credit to Final Fantasy 7 Rebirth rising to the top of its 20-game-strong category.

Here's the full list of Most Wanted Game Golden Joystick 2023 nominees, and as you can see Final Fantasy 7 Rebirth beat 19 other games to come out on top:

  • Death Stranding 2

  • Star Wars Outlaws

  • Final Fantasy VII Rebirth (Winner)

  • Tekken 8

  • Vampire: The Masquerade - Bloodlines 2

  • S.T.A.L.K.E.R. 2: Heart of Chornobyl

  • Hades 2

  • Fable

  • Hollow Knight: Silksong

  • EVERYWHERE

  • Frostpunk 2

  • Ark 2

  • Metal Gear Solid Δ: Snake Eater

  • Persona 3 Reload

  • Bulwark: Falconeer Chronicles

  • Suicide Squad: Kill the Justice League

  • Pacific Drive

  • Black Myth: Wukong

  • Banishers: Ghosts of New Eden

  • Warhammer Age of Sigmar: Realms of Ruin

Discover the best games of 2023 at the best prices by checking out the Golden Joystick Awards Steam sale page

[-] daredevil@kbin.social 26 points 7 months ago* (last edited 7 months ago)

One of my favorite things about /kbin is that it utilizes threads and microblogs. In my experience thus far, users here seem rather shy. I don't hold it against anyone though, because I totally understand.

Federating content from the likes of Mastodon is very helpful for having discussions trickle in from the fediverse. I think it's also really helpful for establishing an ongoing daily discussion space so the thread feed isn't as cluttered. IMO, there's more potential beyond that, too (Think: drawing everyday for a month, photography-based posting/challenges while using tags for content organization, language-learning exercises, the list goes on...).The combination of threads with microblogs has shown me the power that lies behind content federation. As a result, /kbin is by far my favorite of the fediverse platforms so far.

I still have some minor issues with how it currently works. Currently, I believe the name of a magazine causes hashtags with the exact same string to federate content to that magazine. The magazine that matches the desired hashtag also takes priority, even when the hashtag isn't assigned in the magazine's settings. An issue with this is that if any subsequent magazines try to federate content using that hashtag, it won't be able to do so.

It seems as though microblogs can only federate content to either the magazine that matches the hashtag in question, or the magazine that uses the hashtag first. There's also an issue where a microblog that uses multiple hashtags will only federate content to the magazine with the first available tag. E.g. if someone writes an unused tag for the first, followed by #kbinmeta, then #fediverse third, the post would only go to the kbinmeta microblog section. It would be lovely for microblogs to be federated, or even mirrored across magazines (as in child comments/replies) that implement the same tag. Hopefully, this could also be done without adding excessive overhead to Ernest/the server. Perhaps even offer the ability to have a magazine choose to refuse federating tags that match the magazine's name.

There are also some minor issues with moderation federation, but I don't exactly want to specify here, because I'm worried it could be used maliciously.

That being said, I can't wait to see how /kbin will mature.

[-] daredevil@kbin.social 21 points 7 months ago* (last edited 7 months ago)

I've taken care of it. 🙂

14
submitted 7 months ago by daredevil@kbin.social to c/AskKbin@kbin.social

First one that comes to my mind is having to travel with an NPC and our walk/run speeds don't match.

68
submitted 7 months ago by daredevil@kbin.social to c/kbinMeta@kbin.social

@Ernest has pushed an update which allows users to request ownership/moderation of abandoned magazines. Ghost/abandoned magazines were fairly prevalent after the initial wave of hype due to users either squatting magazine names or becoming inactive for other reasons. Now is your chance to get involved, if you were waiting to do so.

To request ownership/moderator privileges, scroll down to where it says "MODERATORS" in the sidebar. There will be an icon of a hand pointing upwards that you can click on, then make the request. Cheers, and thank you for your hard work Ernest, as well as future mods.

2
submitted 8 months ago by daredevil@kbin.social to c/vgmusic@lemmy.world

Title: Let the Battles Begin!
Name: Final Fantasy VII
Year Released: 1997
Composer: Nobuo Uematsu
Developer: Square Enix
Platform: PlayStation

3
submitted 9 months ago by daredevil@kbin.social to c/vgmusic@lemmy.world

Title: Green Hill Zone
Game Name: Sonic the Hedgehog
Year Released: 1991
Composer: Masato Nakamura
Developer: Sonic Team
Platform: Sega Genesis

[-] daredevil@kbin.social 18 points 9 months ago

This does not spark joy.

[-] daredevil@kbin.social 15 points 9 months ago

I hope you're not refusing to clean the cat's litter box because neglect is a form of abuse. I feel sorry for the cat.

[-] daredevil@kbin.social 14 points 9 months ago* (last edited 9 months ago)

A lot of social engagement through social media is driven by impressions such as up votes, favorites, likes, etc. Unfortunately, an easy way to promote engagement and such lies in rage bait. This is likely due to the visceral emotional response generated by rage baiting. I would also extend this issue to how ubiquitous instant gratification is to the internet and social media. People tend to acquire clout through reacting to something quickly, which isn't always well-thought out. Add in the notion of mob mentality, and you have a recipe for the rapid exponential propagation of negative words, thoughts, and emotions. People also tend to not have productive ways of channeling their frustrations and issues, so they often see other entities on the Internet as just a name, sometimes less than that.

There's also a heavy amount of tribalism across a variety of domains which allows one to take refuge from this rage baiting by finding other like-minded individuals to identify with. In some cases, the stress of everyday life or what have you removes a sense of agency or power in one's life and sometimes people cope with this by developing a sense of superiority through whichever group or ideal that they identify with. This cycle repeats itself until there is a constant battle between any given groups where people attempt to elevate their self-worth by putting those that they dont agree with down, while emphasizing the superiority of their own ideal, IMO. I could be totally wrong ofc. I'm hardly perfect.

It's been a pretty exhausting experience. I'm tired of it as well; my fondness for engaging with people has diminished greatly.

[-] daredevil@kbin.social 12 points 10 months ago

AMD has served me well since I've started actively using Linux.

[-] daredevil@kbin.social 26 points 11 months ago

Even though this is a nice development, I'm pretty disappointed in the resources directed at Linux support. I'm considering dropping Proton soon.

[-] daredevil@kbin.social 23 points 11 months ago

Are you shore?

view more: next ›

daredevil

joined 1 year ago