Xiao Lin

@boowiebear @VideoCardz Does it run on Xe integrated graphics? That'd be a good number of devices

@VideoCardz 0.16% of the market is almost zero. I wish ARC was more successful but it is not taking off due to poor performance, price and availability. This is on Intel.

Intel says it offered years of help for Crimson Desert, Pearl Abyss still shipped without Arc support videocardz.com/newz/intel-say…

@LingelKun @absentprototype @wccftech If the artist will still express intent by build ing a full-blown character model, and DLSS5 is just render ing it efficiently, they should also post the full rendering and say DLSS5 is close. Like DLSS 4k and native 4k.

@absentprototype @wccftech So if the artists build the game with dlss5 in mind it’s okay then?

"If DLSS 5 was shown as a next-gen hardware reveal and not AI, you guys would be going nuts." Veteran developer JP Kellams and analyst Ryan Shrout push back against DLSS 5's social media backlash for being a mere "AI filter". 🔗 wccf.tech/1k1ea

@FelixCLC_ Will you be doing pointwise ops near memory?

A truly good tensor ISA has never been tried

@hyhieu226 Papers and peer reviews so successful start ups using them as a central github.

AI paper authors should move from "we propose X" to "we implement X." If a reviewer complains "lack of novelty" or asks "did you compare to Y et al" (woes to them), the best response is "run our code. it gives better results." But maybe just ditch papers...

@YoungEngnr @dcominottim Probably too hard to program, so they had to focus on accessible performance first.

@dcominottim Come on that is a bit much. He has had a huge role in Apple and AMD architecture success. Maybe Tesla as well but hard to tell without independent testing. Unclear why they are doing this though at Tenstorrent - bug ? Early failures ?

In retrospect, Uncle Keller definitely feels way overrated.

@simonw What about data cost

"This is a 600X cost reduction over 7 years, i.e. the cost to train GPT-2 is falling approximately 2.5X every year."

@rajammanabrolu Transformer architecture also "evolved" out of a ton of GPU hours, benchmarking data and many years of distributed research, if we count evolution in. Also: our ancestor won't be able to explore as well as we do. Genes are the same but context is different.

Human learning is def more sample efficient than machines rn (tho not as much as what some claim given evolutionary + societal knowledge accumulation counts as compute). Again, the "pre training" for humans vs models is v different. Models currently encode a large amount of factual information verbatim whereas what it seems like what humans have done is learn how to learn better at exploration (~RL) time. Anyways, moral of the story is that trajectories are different so far and the path to AGI might have nothing in common with the human path to intelligence. We shouldn't over-index on it

@karlbykarlsmith @rajammanabrolu And there's only a few billion bits that came out from this evolution process. We also know quite accurately what those bits are.

@rajammanabrolu Have you crunched the numbers on evolution. Gut says evolutionary information accumulation blows current pre-training out of the water.

@bubbleboi I sure can still afford a 6502 processor right?

I’ve been thinking a lot about Clawd Bot & the race for Mac mini’s a bit over the past few days and I think I’ve come to a very scary realization that explains this crazy phenomenon. Put simply, building a gaming PC will be nearly impossible in the next 5 years… in fact it already is for the vast majority of consumers. But I will go one step farther—in the next 10 years having any type of personal computing device will be unattainable. Fab capacity will be allocated to its most productive and profitable use which is Cloud & AI data centers. Even today, most of the software you run already won’t work without an internet connection. But now with the opportunity cost being so high consumers will be shafted and will only have one option which is moving to the cloud. It’s looking increasingly likely the only hardware you will have is some terminal that connects to the cloud with no workloads running directly on your own hardware. Your device will just have the most basic single core processor and 4 GB of RAM at most.. This is what most people are missing with the fire escape race to acquire Mac Mini’s. The cost for these AI services aren’t just going up they will scale and capture the profitability of the services they provide the same way consulting and financial services extract rent from larger more productive corporations. The only way to protect yourself from the inevitable is to acquire as much hardware that can run inference as fast as fucking possible… Welcome to the computless class.

@FelixCLC_ Train an RL model as the compiler

Lesson of the last 5 years: If you can't describe your programing model to a person, how in the world do you expect to explain it to a compiler?

@deredleritt3r What if the algorithm discovered something fake? Will openai cover any part of the losses?

Let's cut through the untrue sensationalist reporting. - OpenAI is NOT going to come after random people and demand that they share profits from discoveries made with the aid of ChatGPT. - OpenAI is NOT going to "force" anyone to hand over their profits. Here's what's actually going on: OpenAI will approach enterprise customers and offer them to sign a deal that will include a revenue share with OpenAI from products made with the aid of OpenAI's models. It's up to the enterprise customer - a sophisticated party, represented by competent outside counsel - to agree or disagree to pursue this arrangement. We have a recent quote from Sarah Friar on this: "I like [the idea of] licensing models to really align [the customer's and OpenAI's interests]. Let's say in drug discovery, if we licensed our technology, you have a breakthrough. That drug takes off and we get a licensed portion of all its sales." See also the screenshot below from OpenAI's recent blog, which talks about "licensing, IP-based agreements and outcome-based pricing" that would let OpenAI share in the value created by its models.

@SquashBionic Was literally waiting for ROG phone 10 to replace my Razer phone 2

RIP Zenfone and ROG phone line-up, huge wasted potential

@jimkxa Hodl on, memory stonk going up

Quad TT-Galaxies. 4TB of memory. Bigger soon

@pmddomingos What if physics change over time in an unpredictable manner

The universe is the maximum entropy distribution given its symmetries.

@TrueAIHound The next physics break through might not be as impactful as the previous one. There's diminishing returns. But new species often do make some others go extinct. And revolutions can go right or wrong.

In my opinion, regardless of the nonstop hype and lies, the AI community has never done any AI. We are still in the computer automation era. It really began in the 1970s and accelerated in the 1990s with the advent of fast computers and the internet. It has changed the world as we knew it. I believe that, when the true AI revolution comes, it will be Biblical, as they say in the movies. 😬 Note: I also have reasons to believe that the coming revolution will not be about AI alone. I anticipate the arrival of huge breakthroughs in physics as well.

@real_deep_ml Why? Aren't they the state of the art on common sense benchmarks?

LLMs are amazing, they just lack common sense

@pmddomingos Don't need to learn if you can hardcode them. Turned out that machine learning should be called neural hardcoding

The #1 problem in AI is poor generalization, and the #1 solution is exploiting symmetries.

@EERandomness @burkov At least they got their OSes working. Our GUIs have stayed 2D for too long.

@burkov Also, nobody is buying them. I am stunned that even after Apple failed to sell VR, companies are still thinking they are the future. They will have niche uses, but VR just isn't exciting at all to 80% of people.

When Google presented Glass 10 years ago, everyone freaked out that now someone could secretly film you, despite the fact that Glass had an easily recognizable design. Now, just 10 years later, everyone makes XR glasses that look like regular glasses, and no one cares. This is the Overton window effect: if you are rich enough or you are a government, you can gradually change the population's opinion from strictly negative to acceptable in a matter of a decade. This includes the most "difficult" topics like lowering the legal age of consent or assisted suicide.

@realmemes6 Finding the next curve of the tech stack

I don't understand why every car company thinks they can solve one of the most difficult hardware & software problems ever attempted. If you don't like Nvidia, there's Qualcomm and Mobileye solutions. Why are you building your own engineering teams, chip and software stacks??

@Leofunfacts @Rainmaker1973 It takes a lot of PhDs to get the projector to work. You just need a lot of them.

@Rainmaker1973 Since I was a little kid, I have always read stories about child geniuses who finished college as teenagers. Where are they now, 30 years later, and what have they achieved?

A 15-year-old has just earned a PhD in quantum physics. Laurent Simons, a Belgian child prodigy, has blazed an academic trail unmatched by almost anyone else on the planet, accelerating through education at a velocity that defies norms. He began primary school at age four and wrapped it up by six. At twelve, he already held a master's degree in quantum physics, delving into bosons, black holes, and the intricate mathematics unraveling the universe's deepest enigmas. This week, the boy hailed as Belgium's "Little Einstein" defended his doctoral thesis at the University of Antwerp, cementing his status as one of the youngest physics PhDs in recorded history. His research tackled advanced concepts—like Bose polarons in superfluids and supersolids—that most scholars wouldn't touch until their twenties or thirties. Yet for Laurent, this path has been profoundly intimate: the loss of his grandparents at eleven ignited a fire in him to unravel the secrets of longevity—not for personal gain, but to grant others extended, vibrant years. Experts marvel at his prodigious memory and IQ of 145, a rarity shared by just 0.1 percent of the population. Tech giants from the U.S. and China have dangled lucrative offers to his family, but his parents have rebuffed them, championing his right to evolve on his own terms. Laurent doesn't claim the absolute youngest PhD title—that belongs to Karl Witte, who graduated at thirteen in 1814—but in contemporary physics, his feat stands virtually unparalleled. Now fifteen, he's poised to pivot from quantum realms to medical frontiers, eyeing a second doctorate in medical AI to pioneer breakthroughs in aging. His audacious vision? Crafting "superhumans" through innovations that conquer mortality's puzzles—a domain exploding with promise yet riddled with enigmas. Quantum trailblazer or medical revolutionary, Laurent Simons is poised on the cusp: his extraordinary odyssey has only just ignited.