#better than Nvidia

[zman1175]

This Stock Is Up Over 6x More Than Nvidia in 2024 !

[rene-hl-trashcan]

( ˘͈ ᵕ ˘͈♡)

I gave up hoping for something more than the in-game animations' dead eyes.

[transmasc-tabris]

yeeeesssssssssss

[plushchimera]

welcome to dawntrail gunbreaker, finally

gnb is very fun it turns out, like 2 dps in tank's trench coat

[isa-ah]

<- first world problems alert.

bro I see all these live2d showcases of BEAUTIFUL art with the WORST rigging every DAY and I want to chew on my controller and yell I can rig better than that just give me a CHANCE!!!!

my biggest flaw is that i can't make a showcase of my rigging so I can't pitch my hat into the ring. even when I stream, I'm using the lumpy cousin of face tracking softwares so my model is glitchy and stiff. I'd do anything for vbridger tracking ..... shy of getting an actual iphone to use it. ughh

[awkward-teabag]

No but seriously, this is big problem that keeps getting worse and worse.

It used to be you could expect to be able to run new games for years after buying new hardware. Not even top-of-the-line hardware, high-mid tier was good enough. Yes you would have to turn settings down eventually but you could still play at medium to high settings for years to come (unless there was a sudden tech breakthrough).

Now you can spend over a thousand on a new top-end graphics card only for some games a year later recommending you only play on medium settings at the highest.

Don't need to optimize textures, SSDs will compensate for it (and fuck anyone who uses an HDD).

Don't need to worry about memory leaks 'cause RAM will cover that up (and fuck anyone who has 16GB or less).

Don't need to worry about smooth FPS at 1440p/4k, let AI handle the upscaling (and fuck anyone who wants a steady 60 FPS without system stutters or a high power draw).

Don't worry about low VRAM on GPUs, future software will fix it* (and fuck anyone who doesn't use an Nvidia card).

*at the expense of FPS

There's not massive graphical leaps these days, not like there was between console generations in the 90s, 00s, and even into the 10s. Games these days look similar to ones that came out five years ago but the minimum specs has ballooned.

There's no reason why Indiana Jones and the Great Circle needs those specs to run while it looks on par (maybe worse) than games like Assassin's Creed: Valhalla or Ghost of Tsushima. Even Cyberpunk 2077 has a generous minimum requirement compared to what AAA publishers have been putting out these days. It's noticeably less than the recommended specs but it's still specs that are more than a couple years old.

To be clear, I'm not just blaming developers for this. Publishers restart projects (sometimes multiple times) and that can mean losing years of optimization. Developers who are used to working with confines of tech specs and/or know how to account for backwards compatibility get fired so shareholders can get more money. MBAs call the shots even if they have no knowledge or experience with programming. MBAs/tech bros/CEOs are used to always having access to high-end hardware/fast internet.

The list goes on. Losing the confines of physical media has allowed for bad practices to flourish and consumers pay the price.

game developers you are banished to wii hardware limits until you learn your lesson

[bonediggercharleston]

I am very wary of people going "China does it better than America" because most of it is just reactionary rejection of your overlord in favor of his rival, but this story is 1. absolutely legit and 2. way too funny.

US wants to build an AI advantage over China, uses their part in the chip supply chain to cut off China from the high-end chip market.

China's chip manufacturing is famously a decade behind, so they can't advance, right?

They did see it as a problem, but what they then did is get a bunch of Computer Scientists and Junior Programmers fresh out of college and funded their research in DeepSeek. Instead of trying to improve output by buying thousands of Nvidia graphics cards, they tried to build a different kind of model, that allowed them to do what OpenAI does at a tenth of the cost.

Them being young and at a Hedgefund AI research branch and not at established Chinese techgiants seems to be important because chinese corporate culture is apparently full of internal sabotage, so newbies fresh from college being told they have to solve the hardest problems in computing was way more efficient than what usually is done. The result:

American AIs are shook. Nvidia, the only company who actually is making profit cause they are supplying hardware, took a hit. This is just the market being stupid, Nvidia also sells to China. And the worst part for OpenAI. DeepSeek is Open Source.

Anybody can implement deepseek's model, provided they have the hardware. They are totally independent from DeepSeek, as you can run it from your own network. I think you will soon have many more AI companies sprouting out of the ground using this as its base.

What does this mean? AI still costs too much energy to be worth using. The head of the project says so much himself: "there is no commercial use, this is research."

What this does mean is that OpenAI's position is severely challenged: there will soon be a lot more competitors using the DeepSeek model, more people can improve the code, OpenAI will have to ask for much lower prices if it eventually does want to make a profit because a 10 times more efficient opensource rival of equal capability is there.

And with OpenAI or anybody else having lost the ability to get the monopoly on the "market" (if you didn't know, no AI company has ever made a single cent in profit, they all are begging for investment), they probably won't be so attractive for investors anymore. There is a cheaper and equally good alternative now.

AI is still bad for the environment. Dumb companies will still want to push AI on everything. Lazy hacks trying to push AI art and writing to replace real artists will still be around and AI slop will not go away. But one of the main drivers of the AI boom is going to be severely compromised because there is a competitor who isn't in it for immediate commercialization. Instead you will have a more decentralized open source AI field.

Or in short:

[bixels]

Portal 2 is still the perfect game to me. I hyperfixated on it like crazy in middle school. Would sing Want You Gone out loud cuz I had ADHD and no social awareness. Would make fan animations and pixel art. Would explain the ending spoilers and fan theories to anyone who'd listen. Would keep up with DeviantArt posts of the cores as humans. Would find and play community-made maps (Gelocity is insanely fun).

I still can't believe this game came out 12 years ago and it looks like THIS.

Like Mirror's Edge, the timeless art style and economic yet atmospheric lighting means this game will never age. The decision not to include any visible humans (ideas of Doug Rattmann showing up or a human co-op partner were cut) is doing so much legroom too. And the idea to use geometric tileset-like level designs is so smart! I sincerely believe that, by design, no game with a "realistic art style" has looked better than Portal 2.

Do you guys remember when Nvidia released Portal with RTX at it looked like dogshit? Just the most airbrushed crap I've ever seen; completely erased the cold, dry, clinical feel of Aperture.

So many breathtakingly pit-in-your-stomach moments I still think about too. And it's such a unique feeling; I'd describe at as... architectural existentialism? Experiencing the sublime under the shadow of manmade structures (Look up Giovanni Battista Piranesi's art if you're curious)? That scene where you're running from GLaDOS with Wheatley on a catwalk over a bottomless pit and––out of rage and desperation––GLaDOS silently begins tearing her facility apart and Wheatley cries 'She's bringing the whole place down!' and ENORMOUS apartment building-sized blocks begin groaning towards you on suspended rails and cement pillars crumble and sparks fly and the metal catwalk strains and bends and snaps under your feet. And when you finally make it to the safety of a work lift, you look back and watch the facility close its jaws behind you as it screams.

Or the horror of knowing you're already miles underground, and then Wheatley smashes you down an elevator shaft and you realize it goes deeper. That there's a hell under hell, and it's much, much older.

Or how about the moment when you finally claw your way out of Old Aperture, reaching the peak of this underground mountain, only to look up and discover an endless stone ceiling built above you. There's a service door connected to some stairs ahead, but surrounding you is this array of giant, building-sized springs that hold the entire facility up. They stretch on into the fog. You keep climbing.

I love that the facility itself is treated like an android zooid too, a colony of nano-machines and service cores and sentient panel arms and security cameras and more. And now, after thousands of years of neglect, the facility is festering with decomposition and microbes; deer, raccoons, birds. There are ghosts too. You're never alone, even when it's quiet. I wonder what you'd hear if you put your ear up against a test chamber's walls and listened. (I say that all contemplatively, but that's literally an easter egg in the game. You hear a voice.)

Also, a reminder that GLaDOS and Chell are not related and their relationship is meant to be psychosexual. There was a cut bit where GLaDOS would role-play as Chell's jealous housewife and accuse her of seeing other cores in between chambers. And their shared struggle for freedom and control? GLaDOS realizing, after remembering her past life, that she's become the abuser and deciding that she has the power to stop? That even if she can't be free, she can let Chell go because she hates her. And she loves her. Most people interpret GLaDOS "deleting Caroline in her brain" as an ominous sign, that she's forgetting her human roots and becoming "fully robot." But to me, it's a sign of hope for GLaDOS. She's relieving herself of the baggage that has defined her very existence, she's letting Caroline finally rest, and she's allowing herself to grow beyond what Cave and Aperture and the scientists defined her to be. The fact that GLaDOS still lets you go after deleting Caroline proves this. She doesn't double-back or change her mind like Wheatley did, she sticks to her word because she knows who she is. No one and nothing can influence her because she's in control. GLaDOS proves she's capable of empathy and mercy and change, human or not.

That's my retrospective, I love this game to bits. I wish I could experience it for the first time again.

[mostlysignssomeportents]

“Humans in the loop” must detect the hardest-to-spot errors, at superhuman speed

I'm touring my new, nationally bestselling novel The Bezzle! Catch me SATURDAY (Apr 27) in MARIN COUNTY, then Winnipeg (May 2), Calgary (May 3), Vancouver (May 4), and beyond!

If AI has a future (a big if), it will have to be economically viable. An industry can't spend 1,700% more on Nvidia chips than it earns indefinitely – not even with Nvidia being a principle investor in its largest customers:

https://news.ycombinator.com/item?id=39883571

A company that pays 0.36-1 cents/query for electricity and (scarce, fresh) water can't indefinitely give those queries away by the millions to people who are expected to revise those queries dozens of times before eliciting the perfect botshit rendition of "instructions for removing a grilled cheese sandwich from a VCR in the style of the King James Bible":

https://www.semianalysis.com/p/the-inference-cost-of-search-disruption

Eventually, the industry will have to uncover some mix of applications that will cover its operating costs, if only to keep the lights on in the face of investor disillusionment (this isn't optional – investor disillusionment is an inevitable part of every bubble).

Now, there are lots of low-stakes applications for AI that can run just fine on the current AI technology, despite its many – and seemingly inescapable - errors ("hallucinations"). People who use AI to generate illustrations of their D&D characters engaged in epic adventures from their previous gaming session don't care about the odd extra finger. If the chatbot powering a tourist's automatic text-to-translation-to-speech phone tool gets a few words wrong, it's still much better than the alternative of speaking slowly and loudly in your own language while making emphatic hand-gestures.

There are lots of these applications, and many of the people who benefit from them would doubtless pay something for them. The problem – from an AI company's perspective – is that these aren't just low-stakes, they're also low-value. Their users would pay something for them, but not very much.

For AI to keep its servers on through the coming trough of disillusionment, it will have to locate high-value applications, too. Economically speaking, the function of low-value applications is to soak up excess capacity and produce value at the margins after the high-value applications pay the bills. Low-value applications are a side-dish, like the coach seats on an airplane whose total operating expenses are paid by the business class passengers up front. Without the principle income from high-value applications, the servers shut down, and the low-value applications disappear:

https://locusmag.com/2023/12/commentary-cory-doctorow-what-kind-of-bubble-is-ai/

Now, there are lots of high-value applications the AI industry has identified for its products. Broadly speaking, these high-value applications share the same problem: they are all high-stakes, which means they are very sensitive to errors. Mistakes made by apps that produce code, drive cars, or identify cancerous masses on chest X-rays are extremely consequential.

Some businesses may be insensitive to those consequences. Air Canada replaced its human customer service staff with chatbots that just lied to passengers, stealing hundreds of dollars from them in the process. But the process for getting your money back after you are defrauded by Air Canada's chatbot is so onerous that only one passenger has bothered to go through it, spending ten weeks exhausting all of Air Canada's internal review mechanisms before fighting his case for weeks more at the regulator:

https://bc.ctvnews.ca/air-canada-s-chatbot-gave-a-b-c-man-the-wrong-information-now-the-airline-has-to-pay-for-the-mistake-1.6769454

There's never just one ant. If this guy was defrauded by an AC chatbot, so were hundreds or thousands of other fliers. Air Canada doesn't have to pay them back. Air Canada is tacitly asserting that, as the country's flagship carrier and near-monopolist, it is too big to fail and too big to jail, which means it's too big to care.

Air Canada shows that for some business customers, AI doesn't need to be able to do a worker's job in order to be a smart purchase: a chatbot can replace a worker, fail to their worker's job, and still save the company money on balance.

I can't predict whether the world's sociopathic monopolists are numerous and powerful enough to keep the lights on for AI companies through leases for automation systems that let them commit consequence-free free fraud by replacing workers with chatbots that serve as moral crumple-zones for furious customers:

https://www.sciencedirect.com/science/article/abs/pii/S0747563219304029

But even stipulating that this is sufficient, it's intrinsically unstable. Anything that can't go on forever eventually stops, and the mass replacement of humans with high-speed fraud software seems likely to stoke the already blazing furnace of modern antitrust:

https://www.eff.org/de/deeplinks/2021/08/party-its-1979-og-antitrust-back-baby

Of course, the AI companies have their own answer to this conundrum. A high-stakes/high-value customer can still fire workers and replace them with AI – they just need to hire fewer, cheaper workers to supervise the AI and monitor it for "hallucinations." This is called the "human in the loop" solution.

The human in the loop story has some glaring holes. From a worker's perspective, serving as the human in the loop in a scheme that cuts wage bills through AI is a nightmare – the worst possible kind of automation.

Let's pause for a little detour through automation theory here. Automation can augment a worker. We can call this a "centaur" – the worker offloads a repetitive task, or one that requires a high degree of vigilance, or (worst of all) both. They're a human head on a robot body (hence "centaur"). Think of the sensor/vision system in your car that beeps if you activate your turn-signal while a car is in your blind spot. You're in charge, but you're getting a second opinion from the robot.

Likewise, consider an AI tool that double-checks a radiologist's diagnosis of your chest X-ray and suggests a second look when its assessment doesn't match the radiologist's. Again, the human is in charge, but the robot is serving as a backstop and helpmeet, using its inexhaustible robotic vigilance to augment human skill.

That's centaurs. They're the good automation. Then there's the bad automation: the reverse-centaur, when the human is used to augment the robot.

Amazon warehouse pickers stand in one place while robotic shelving units trundle up to them at speed; then, the haptic bracelets shackled around their wrists buzz at them, directing them pick up specific items and move them to a basket, while a third automation system penalizes them for taking toilet breaks or even just walking around and shaking out their limbs to avoid a repetitive strain injury. This is a robotic head using a human body – and destroying it in the process.

An AI-assisted radiologist processes fewer chest X-rays every day, costing their employer more, on top of the cost of the AI. That's not what AI companies are selling. They're offering hospitals the power to create reverse centaurs: radiologist-assisted AIs. That's what "human in the loop" means.

This is a problem for workers, but it's also a problem for their bosses (assuming those bosses actually care about correcting AI hallucinations, rather than providing a figleaf that lets them commit fraud or kill people and shift the blame to an unpunishable AI).

Humans are good at a lot of things, but they're not good at eternal, perfect vigilance. Writing code is hard, but performing code-review (where you check someone else's code for errors) is much harder – and it gets even harder if the code you're reviewing is usually fine, because this requires that you maintain your vigilance for something that only occurs at rare and unpredictable intervals:

https://twitter.com/qntm/status/1773779967521780169

But for a coding shop to make the cost of an AI pencil out, the human in the loop needs to be able to process a lot of AI-generated code. Replacing a human with an AI doesn't produce any savings if you need to hire two more humans to take turns doing close reads of the AI's code.

This is the fatal flaw in robo-taxi schemes. The "human in the loop" who is supposed to keep the murderbot from smashing into other cars, steering into oncoming traffic, or running down pedestrians isn't a driver, they're a driving instructor. This is a much harder job than being a driver, even when the student driver you're monitoring is a human, making human mistakes at human speed. It's even harder when the student driver is a robot, making errors at computer speed:

https://pluralistic.net/2024/04/01/human-in-the-loop/#monkey-in-the-middle

This is why the doomed robo-taxi company Cruise had to deploy 1.5 skilled, high-paid human monitors to oversee each of its murderbots, while traditional taxis operate at a fraction of the cost with a single, precaratized, low-paid human driver:

https://pluralistic.net/2024/01/11/robots-stole-my-jerb/#computer-says-no

The vigilance problem is pretty fatal for the human-in-the-loop gambit, but there's another problem that is, if anything, even more fatal: the kinds of errors that AIs make.

Foundationally, AI is applied statistics. An AI company trains its AI by feeding it a lot of data about the real world. The program processes this data, looking for statistical correlations in that data, and makes a model of the world based on those correlations. A chatbot is a next-word-guessing program, and an AI "art" generator is a next-pixel-guessing program. They're drawing on billions of documents to find the most statistically likely way of finishing a sentence or a line of pixels in a bitmap:

https://dl.acm.org/doi/10.1145/3442188.3445922

This means that AI doesn't just make errors – it makes subtle errors, the kinds of errors that are the hardest for a human in the loop to spot, because they are the most statistically probable ways of being wrong. Sure, we notice the gross errors in AI output, like confidently claiming that a living human is dead:

https://www.tomsguide.com/opinion/according-to-chatgpt-im-dead

But the most common errors that AIs make are the ones we don't notice, because they're perfectly camouflaged as the truth. Think of the recurring AI programming error that inserts a call to a nonexistent library called "huggingface-cli," which is what the library would be called if developers reliably followed naming conventions. But due to a human inconsistency, the real library has a slightly different name. The fact that AIs repeatedly inserted references to the nonexistent library opened up a vulnerability – a security researcher created a (inert) malicious library with that name and tricked numerous companies into compiling it into their code because their human reviewers missed the chatbot's (statistically indistinguishable from the the truth) lie:

https://www.theregister.com/2024/03/28/ai_bots_hallucinate_software_packages/

For a driving instructor or a code reviewer overseeing a human subject, the majority of errors are comparatively easy to spot, because they're the kinds of errors that lead to inconsistent library naming – places where a human behaved erratically or irregularly. But when reality is irregular or erratic, the AI will make errors by presuming that things are statistically normal.

These are the hardest kinds of errors to spot. They couldn't be harder for a human to detect if they were specifically designed to go undetected. The human in the loop isn't just being asked to spot mistakes – they're being actively deceived. The AI isn't merely wrong, it's constructing a subtle "what's wrong with this picture"-style puzzle. Not just one such puzzle, either: millions of them, at speed, which must be solved by the human in the loop, who must remain perfectly vigilant for things that are, by definition, almost totally unnoticeable.

This is a special new torment for reverse centaurs – and a significant problem for AI companies hoping to accumulate and keep enough high-value, high-stakes customers on their books to weather the coming trough of disillusionment.

This is pretty grim, but it gets grimmer. AI companies have argued that they have a third line of business, a way to make money for their customers beyond automation's gifts to their payrolls: they claim that they can perform difficult scientific tasks at superhuman speed, producing billion-dollar insights (new materials, new drugs, new proteins) at unimaginable speed.

However, these claims – credulously amplified by the non-technical press – keep on shattering when they are tested by experts who understand the esoteric domains in which AI is said to have an unbeatable advantage. For example, Google claimed that its Deepmind AI had discovered "millions of new materials," "equivalent to nearly 800 years’ worth of knowledge," constituting "an order-of-magnitude expansion in stable materials known to humanity":

https://deepmind.google/discover/blog/millions-of-new-materials-discovered-with-deep-learning/

It was a hoax. When independent material scientists reviewed representative samples of these "new materials," they concluded that "no new materials have been discovered" and that not one of these materials was "credible, useful and novel":

https://www.404media.co/google-says-it-discovered-millions-of-new-materials-with-ai-human-researchers/

As Brian Merchant writes, AI claims are eerily similar to "smoke and mirrors" – the dazzling reality-distortion field thrown up by 17th century magic lantern technology, which millions of people ascribed wild capabilities to, thanks to the outlandish claims of the technology's promoters:

https://www.bloodinthemachine.com/p/ai-really-is-smoke-and-mirrors

The fact that we have a four-hundred-year-old name for this phenomenon, and yet we're still falling prey to it is frankly a little depressing. And, unlucky for us, it turns out that AI therapybots can't help us with this – rather, they're apt to literally convince us to kill ourselves:

https://www.vice.com/en/article/pkadgm/man-dies-by-suicide-after-talking-with-ai-chatbot-widow-says

If you'd like an essay-formatted version of this post to read or share, here's a link to it on pluralistic.net, my surveillance-free, ad-free, tracker-free blog:

https://pluralistic.net/2024/04/23/maximal-plausibility/#reverse-centaurs

Image: Cryteria (modified) https://commons.wikimedia.org/wiki/File:HAL9000.svg

CC BY 3.0 https://creativecommons.org/licenses/by/3.0/deed.en

[ayeforscotland]

Sora looks awesome from OpenAI and then also Chat with RTX (Nvidia) will have a personal local LLM on your own machine but new windows updates will have co-pilot too. The future of AI is going to be awesome. As someone in the data field, you have to keep moving with it or be left without. IT is definitely an exciting time.

As someone else in the data field, my full background is in data and data flow, AI is the latest buzzword that a small group of people in Silicon Valley have pushed to work people up into a frenzy.

The people cheering on AI are the same people who said NFTs were going to radically change the world of work.

I think there’s positive uses for AI, particularly in pattern recognition, like detecting cancer.

However, Sora looks like shit. It’s producing videos of three-legged cats, and it’s using stolen work to do it. And sure, it’ll get better, but without regulation all it will do is poison the well of human knowledge as certain groups begin to create things that aren’t real. We move into a world where evidence can be fabricated.

Why are generative AI fans targeting artists who voice their concerns? Every day I see some AI techbro tweeting an artist and saying they’ve just scrolled through their art and fed it to an algorithm. It is scummy behaviour.

As a fellow ‘data field’ person, you’ll know that AI is also only as useful as what we feed it. Most organisations don’t know where their data actually is, they’re desperately trying to backpedal their huge push to the cloud and host things on premise. The majority of digital transformation projects fail, more fines are being handed out for failing compliance than ever, and companies can’t possibly claim to be cyber secure when they don’t know where they’re holding their data.

AI won’t fix any of this. It needs human engineering and standardisation to fix, non-technical and technical teams need to understand the connectivity of every process and piece of technology and maybe then some form of AI can be used to optimise processes.

But you can’t just introduce AI and think it fixes large-scale issues. It will amplify them if you continue to feed it garbage.

[autumnslance]

Benchmark Tech Notes

Running the Benchmark

If your Benchmark isn't opening, it's an issue with the executable file, and something not completing properly on either download, or extracting the Zip file. The Benchmark is designed to run and give you scores for your potato computer, I promise.

I actually saved my Benchmark to my external drive, and it still pulls and saves data and runs as it should. Make sure you allowed the download to complete before extracting the zip.

Resolution

Check your Settings; in Display, it may be defaulting your monitor Resolution to something than you might otherwise use if you aren't on standard 1920x1080.

To check your monitor Resolution, minimize everything on your screen and right click anywhere on your Desktop. Go to Display Settings and scroll down to find Resolution and what it's set at.

You can set the Graphic Settings 1 tab to Maximum, or to Import your game settings. Display Settings tab is where you set it to be Windowed, Bordered, or Full Screen, as well as select Resolution to match your monitor in the dropdown (or customize it if needed). I speak on Resolution as some folks in my FC noted it changed how their characters looked.

The Other tab in Settings is where you can change the text output, or even check a box to disable the logo and score; I do this on subsequent plays, once I have my scores at various settings, to get the clean screenshots.

@calico-heart has a post about fixing graphics settings, with screenshots of the settings tab. Basically, change graphics upscaling from AMD to NVIDIA, and/or uncheck Enable Dynamic Resolution. Also check the Framerate Threshold dropdown.

Screenshots

The benchmark auto-saves 5 screens each playthrough. In the Benchmark folder there is a Screenshots folder to find the auto-images taken of your characters.

Character Appearance

If you want to get your current in game appearance, including non-standard hairstyles, make sure to load up the live game, right click and "Save Character Settings."

Then go to Documents/My Games/Final Fantasy XIV: A Realm Reborn (this is the default in Windows 10 so mileage varies). The file will have the date you last updated their settings and be named FFXIV_CHARA_01.dat (or however many saves you have/made).

Grab those newly updated DAT files for your character(s) and copy them, then in the same base folder, go to Final Fantasy XIV: A Realm Reborn (Benchmark).

Paste the copied DAT files in there, and rename to FFXIV_CHARA_BENCH01.dat (the number doesn't matter, and you may have more).

When running Benchmark Character Creation, use the dropdown menu.

If you do Create a Custom Character and Load Appearance Data, it will give you default hairstyles again. Meteor's Dawntrail hairstyle is a new default.

In Char Gen I am finding that a very pale hrothgal reflects the green scenery around her, giving her white skin/fur a green tinge. The other zones do not have this problem, or at least not to the same degree.

They added a Midday vs Evening setting in outdoor areas as well to test lighting. The lighting in the Gridanian innroom is better; not as bright as outdoors, to be expected, but not completely useless.

New voice type icons to clarifying the sounds you make.

Remember we're getting a free fantasia with the expansion, so some tweaking may be needed; Iyna I felt like I needed to adjust her jaw. Other colors--skin, hair, eyes, tattoos, etc--are showing differently in the various kinds of lighting.

Uncertain if the limit on hairstyles for the Hrothgals so far is just a Benchmark thing; they do have set styles for different head options. Everyone gets Meteor's hair though, so it may be a temporary/Benchmark limit. But which clan and face you choose drastically alters what hair and facial feature options you have access to.

Check your settings, tweak them a bit, play around with chargen, and remember this is still a Benchmark; they always strike me as a little less polished than the finished game, but so far I'm actually pretty pleased with having defined fingers and toes, the irises in the eyes, scars looking cut into the skin, and other improvements.

[canmom]

using LLMs to control a game character's dialogue seems an obvious use for the technology. and indeed people have tried, for example nVidia made a demo where the player interacts with AI-voiced NPCs:

this looks bad, right? like idk about you but I am not raring to play a game with LLM bots instead of human-scripted characters. they don't seem to have anything interesting to say that a normal NPC wouldn't, and the acting is super wooden.

so, the attempts to do this so far that I've seen have some pretty obvious faults:

relying on external API calls to process the data (expensive!)

presumably relying on generic 'you are xyz' prompt engineering to try to get a model to respond 'in character', resulting in bland, flavourless output

limited connection between game state and model state (you would need to translate the relevant game state into a text prompt)

responding to freeform input, models may not be very good at staying 'in character', with the default 'chatbot' persona emerging unexpectedly. or they might just make uncreative choices in general.

AI voice generation, while it's moved very fast in the last couple years, is still very poor at 'acting', producing very flat, emotionless performances, or uncanny mismatches of tone, inflection, etc.

although the model may generate contextually appropriate dialogue, it is difficult to link that back to the behaviour of characters in game

so how could we do better?

the first one could be solved by running LLMs locally on the user's hardware. that has some obvious drawbacks: running on the user's GPU means the LLM is competing with the game's graphics, meaning both must be more limited. ideally you would spread the LLM processing over multiple frames, but you still are limited by available VRAM, which is contested by the game's texture data and so on, and LLMs are very thirsty for VRAM. still, imo this is way more promising than having to talk to the internet and pay for compute time to get your NPC's dialogue lmao

second one might be improved by using a tool like control vectors to more granularly and consistently shape the tone of the output. I heard about this technique today (thanks @cherrvak)

third one is an interesting challenge - but perhaps a control-vector approach could also be relevant here? if you could figure out how a description of some relevant piece of game state affects the processing of the model, you could then apply that as a control vector when generating output. so the bridge between the game state and the LLM would be a set of weights for control vectors that are applied during generation.

this one is probably something where finetuning the model, and using control vectors to maintain a consistent 'pressure' to act a certain way even as the context window gets longer, could help a lot.

probably the vocal performance problem will improve in the next generation of voice generators, I'm certainly not solving it. a purely text-based game would avoid the problem entirely of course.

this one is tricky. perhaps the model could be taught to generate a description of a plan or intention, but linking that back to commands to perform by traditional agentic game 'AI' is not trivial. ideally, if there are various high-level commands that a game character might want to perform (like 'navigate to a specific location' or 'target an enemy') that are usually selected using some other kind of algorithm like weighted utilities, you could train the model to generate tokens that correspond to those actions and then feed them back in to the 'bot' side? I'm sure people have tried this kind of thing in robotics. you could just have the LLM stuff go 'one way', and rely on traditional game AI for everything besides dialogue, but it would be interesting to complete that feedback loop.

I doubt I'll be using this anytime soon (models are just too demanding to run on anything but a high-end PC, which is too niche, and I'll need to spend time playing with these models to determine if these ideas are even feasible), but maybe something to come back to in the future. first step is to figure out how to drive the control-vector thing locally.

[argumate]

nostalgebraist said: yeah i think this drop is just more evidence that the concept of “ML scaling” is _still_ not priced into nvidia and co. despite being well-known in the industry for some time. there are papers on more efficient training methods being published all the time (deepseek’s are especially good ones, but they are part of an existing literature), and no one at the big labs would ever read this stuff and think “oh i guess i don’t need to buy more gpus now that i have this.” you do the efficiency improvements *and* buy more gpus, and so does the other guy, and that’s why you do it, *because* you’re racing the other guy and you’re going to do the absolute best you can. deepseek themselves are building on an immense amount of efficient training lore invented at google/meta/etc (transformer, MoE, etc). it’s only because of the anomalous situation created by export controls that we see someone land on “fewer/worse gpus + more efficiency -> similar results” rather than “more/better gpus + more efficiency -> better results”; the latter is the default and that’s why the former feels so surprising. the market is just not very smart about this topic – recall that nvidia’s price is still 2x above where it was for all of 2023, despite chatgpt and everything! (…possible gell-mann amnesia case?)

gell-man markets hypothesis, where markets are efficient in every area except those where we have personal knowledge 😆

[itstimewehavesomesoliddick]

Style was forged by the constraints of graphics limitations, it was a way to make your game look good and set it apart with limited means, no limitations = no need to bother since the game will look "good" (read: bland) anyway with fancy premade shaders and physX particles and shit, everything can look bland and the same and uninspired BUT THOSE 8K TEXTURES WITH RAYTRACING THO

Zoomers don't even know things used to be better and we're getting old, you better hold on to your emulators and dgvoodoo

I really hope there’s major pushback against photorealism in gaming but I wouldn’t hold my breath

[mr-entj]

Hello Mr. ENTJ. I'm an ENTJ sp/so 3 woman in her early twenties with a similar story to yours (Asian immigrant with a chip on her shoulder, used going to university as a way to break generational cycles). I graduated last month and have managed to break into strategy consulting with a firm that specialises in AI. Given your insider view into AI and your experience also starting out as a consultant, I would love to hear about any insights you might have or advice you may have for someone in my position. I would also be happy to take this discussion to somewhere like Discord if you'd prefer not to share in public/would like more context on my situation. Thank you!

Insights for your career or insights on AI in general?

On management consulting as a career, check the #management consulting tag.

On being a consultant working in AI:

Develop a solid understanding of the technical foundation behind LLMs. You don’t need a computer science degree, but you should know how they’re built and what they can do. Without this knowledge, you won’t be able to apply them effectively to solve any real-world problems. A great starting point is deeplearning.ai by Andrew Ng: Fundamentals, Prompt Engineering, Fine Tuning

Know all the terminology and definitions. What's fine tuning? What's prompt engineering? What's a hallucination? Why do they happen? Here's a good starter guide.

Understand the difference between various models, not just in capabilities but also training, pricing, and usage trends. Great sources include Artificial Analysis and Hugging Face.

Keep up to date on the newest and hottest AI startups. Some are hype trash milking the AI gravy train but others have actual use cases. This will reveal unique and interesting use cases in addition to emerging capabilities. Example: Forbes List.

On the industry of AI:

It's here to stay. You can't put the genie back in the bottle (for anyone reading this who's still a skeptic).

AI will eliminate certain jobs that are easily automated (ex: quality assurance engineers) but also create new ones or make existing ones more important and in-demand (ex: prompt engineers, machine learning engineers, etc.)

The most valuable career paths will be the ones that deal with human interaction, connection, and communication. Soft skills are more important than ever because technical tasks can be offloaded to AI. As Sam Altman once told me in a meeting: "English is the new coding language."

Open source models will win (Llama, Mistral, Deep Seek) because closed source models don't have a moat. Pick the cheapest model because they're all similarly capable.

The money is in the compute, not the models -- AI chips, AI infrastructure, etc. are a scarce resource and the new oil. This is why OpenAI ($150 billion valuation) is only 5% the value of NVIDIA (a $3 trillion dollar behemoth). Follow the compute because this is where the growth will happen.

America and China will lead in the rapid development and deployment of AI technology; the EU will lead in regulation. Keep your eye on these 3 regions depending on what you're looking to better understand.

[mckitterick]

Techbros are sad

from The Beaverton, Canada's The Onion, on the latest AI hubub:

America’s magic bean market is in turmoil after a Chinese company unveiled a competing bean that’s less than one tenth the price but still just as useless.

'DeepBean’s new beans only require a third of the water that American beans need to not grow into anything of value,' an analyst said. 'Any serious investor or cow owner would do well to give DeepBean a look.'

American magic bean companies like Beanco, The Boston Bean Company, and Nvidia have already shed hundreds of billions of dollars in stock value as investors contemplate getting their magic beans from overseas.

'Society can’t advance without magic beans to grow into bizarre and ultimately worthless simulacrums of real beanstalks,' one investor said. 'If we’re only wasting moderate amounts of water on them instead of massive amounts, that’s a plus.'

awww, poor techbros sad that their trillion-dollar boondoggle is no better than the million-dollar Chinese boondoggle?

full Beaverton story here: X