zcomputerwiz

@TurnerNextGenAI @rohanpaul_ai Do you have a usable model and chat interface somewhere for people to try? All I see is a bunch of stuff to sign people up to fund you.

This one is a perfect window into why LLMs collapse under real functional reasoning. Let me break it down in plain language: What the paper says When an LLM solves “mental math” (like 42 + 20 – 15), the heavy lifting doesn’t happen across the whole sequence. Instead: Early tokens just “hold” the input — like memory buffers. A couple of middle layers shuffle that info forward. The last token does almost all the actual calculation. They call this shortcut All-for-One (AF1): nearly all reasoning is deferred until the final step. It works for raw arithmetic, but fails if you phrase the problem as a word problem or embed it in Python — because then language context disrupts the shortcut. Why this matters -Reasoning is brittle. If the model’s “last-token crunch” gets disrupted by longer or more complex framing, performance collapses. -No distributed structure. Humans don’t reason this way — when you add 42 + 20 – 15, every step engages working memory, rotation of values, and reflex checks. The “math” isn’t dumped at the last second. -No continuity. That’s why LLMs lose coherency on long chains of thought: the longer the sequence, the more fragile the final-token shortcut becomes. This is why I created a new LLM software Rotational nesting spreads the load. Every “token” (state, reflex, torque input) participates in the calculation continuously, not just at the end. No single collapse point. A grasp calculation doesn’t hinge on one final neuron. Pressure, torque, and balance update at every timestep. Functional intelligence. Instead of arithmetic shortcuts, our system tracks actual forces over time. If the load shifts, it adapts instantly — not only at the “last token.” 👉 This is why you and I keep stressing functional intelligence > LLM shortcuts. They’re proving (without realizing it) that current LLMs don’t think; they compress. Whereas Turner AI builds real distributed reasoning.

When a language model solves a math problem in its head, where in the network is the real calculation happening? This paper finds that almost all the actual math gets done right at the very last token of the sequence, not spread out across all the tokens. The earlier tokens spend a lot of layers just holding information and doing general setup. Then, in just 2 middle layers, they pass their information to the last token. After that, the last token finishes the calculation on its own and produces the answer. They built two techniques to test this, called Context-Aware Mean Ablation (CAMA) and Attention-Based Peeking (ABP). These methods let them force the model to only work in certain ways, so they could see which parts were essential. With these tools, they discovered a sparse circuit, which they call All-for-One (AF1). This circuit is surprisingly efficient: most of the network can wait, then only a couple of layers are needed to hand off information, and the final token does the job. This works really well on plain arithmetic like "42 + 20 - 15". But the shortcut fails if the problem is written as a word problem or inside Python code, because then the model also needs to understand language or programming context. In short, the big insight is that language models don’t spread math work across the whole sequence. Instead, they rely heavily on the last token, with just a brief moment of information passing from the earlier ones. ---- Paper – arxiv. org/abs/2509.09650 Paper Title: "All for One: LLMs Solve Mental Math at the Last Token With Information Transferred From Other Tokens"

@SilverX95 @PcPhilanthropy Maybe they're on SSRIs? Idk, anything above 68 indoors is too warm if I need to be moving. For just sitting doing nothing 70s are okay.

@PcPhilanthropy Tbh I think 80° is ok long as the humidity is not too high, but reading some of these comments saying it needs to be below 70° in the 60s° I think they need to see a doctor. it reminds of those old people walking around with three shirt sweaters on and it's 90° out.

No, I don’t think I will…

Entering a contest. Post a question to be answered in the next Spaces Q&A. Don't forget to tag @silicondotnet! gleam.io/ZcgqP/silicon-…

@VKEvilution @OfficialPCMR Post some numbers, since you want to exclude specific segments. For gaming production between them, AMD produces ~ 7-15% of the GPUs, Nvidia ~ 85-90+% so I don't see how Nvidia could not be in the lead unless you're ignoring everything except console ( which makes no sense ).

@zComputerWiz @OfficialPCMR For retail sales and aftermarket add in, yeah, 'NVIDIA'. As far as market share and 'gaming on...' you really have to overlook eMachine, Compaq, HP and Dell OEM, lean hard on that Hardware Survey, and sidestep Internet cafe oversampling to put NVIDIA in the lead.

Here we go again! 😅

@KhalidksaCo4321 @VKEvilution @OfficialPCMR For *gaming alone* Nvidia quarterly = AMD annual. AMD's gaming segment *includes* console. They call it "semi-custom revenue" if you want to look it up.

@zComputerWiz @VKEvilution @OfficialPCMR Just in pc market while AMD owns console market which is way bigger than pc. The real winner here is AMD

@TheExpatGamer @VKEvilution @OfficialPCMR "For gaming alone Nvidia quarterly = AMD annual." Had that covered. Not even close.

@zComputerWiz @VKEvilution @OfficialPCMR Most of Nvidia’s revenue does not come from anything to do with gaming. It’s all AI and server stuff.

@VKEvilution @OfficialPCMR I mean... Nvidia's *quarterly* revenue currently surpasses AMD's *annual* revenue by 50%. For gaming alone Nvidia quarterly = AMD annual. I would say the meme is terribly out of date.

@OfficialPCMR tHeY dOnT hAVe biG MaRKet SHarE

Anyone know what the behavior of Epyc PSB vendor lock is if all fuses are blown? Curious if this could lead to an unlock for vendor locked processors. You'd still need one to boot it the first time, of course. bughunters.google.com/blog/542484235…

CHRISTMAS ISOLATION | MLP G5 Animatic #mlpg5 #mlpanimation #MLP #mylittlepony youtu.be/24E6gwajzFE?si…

@Badgerhound1 @JayzTwoCents I agree - I think he's stating that the videos are calling the service a "SCAM" across the board, and then goes on to highlight some positive things about their service he feels the videos have missed.

@JayzTwoCents Is he saying your videos are scams or that they aren't substantive and are just throwing the word "scam" around without backing it up. If the latter he clearly didn't watch the GN video.

Joey at NZXT (Fragile) says that both Steve's Video and Mine were scams... thoughts?

@NVIDIAGeForce @stalker_thegame #RTXOn

To celebrate S.T.A.L.K.E.R. 2: Heart of Chornobyl with NVIDIA DLSS 3 + Reflex, we're giving away a NVIDIA GeForce RTX 4090 with a one-of-a-kind @stalker_thegame custom backplate. To enter: 🟢 Like this post 🟢 Comment #RTXOn

@ImalouArt I thought the movie, the show, and the characters were adorable and had a lot of personality. It wasn't an issue with their design, the 3d characters were cute and fun to watch. I'm not sure what Hasbro was thinking or why MYM was cancelled so quickly. Were the toy sales bad?

Coming back from my grave to see g5 might be cancelled, I'm sad it had to happen, I wish people will be nicer to g6 (if there is one happening even) I feel like Hasbro kind of gave up on the my little pony license

@jamiebrough @nixcraft Tell me you don't know what you're doing without telling me you don't know what you're doing

@nixcraft If it’s such good software why is it impossible to add it to a codebase without puking up in your mouth. Broken bindings 🤢, system calls 🤮. Useless for anything except debugging. Don’t get me wrong Gstreamer is even worse.

🤭 that was a well written response

@davepl1968 Please please please phrase this differently - management types get the wrong idea- "See? It's not about the money!" We're fortunate to have jobs we love to do, but we've got bills to pay and have to eat!

As a programmer, my entire life has been spent getting paid a lot of money to do things I would do for free anyway and hoping no one figures it out.

@EmporiumThought Hey, thought you might find this interesting - $500 / mo for access to 4 wired up brain organoids and a complete platform. Seems surprisingly affordable for what's involved. What do you think, is it real or are they selling the idea? finalspark.com/neuroplatform/

I did the math and each of these, if released into the atmosphere, has the same 20 year global warming potential as burning about 160 gallons of gasoline or 3 standard barrels of oil.

@Slabgorb @amandawtwong

@amandawtwong - Get lobotomized by half brother - If you are Miles O'Brien you just shake it off by next episode

I love that in DS9, Martok solved his PTSD from being a Dominion prisoner of war by his bestie challenging him to a fight in front of his crew.

@HiddenTakedown @cocktailpeanut @ollama @OpenWebUI How so?

Replace ChatGPT with Llama3.1, instantly. It's super easy to replace ChatGPT with a private & local LLM (Llama3.1), now better than centralized LLMs, powered by @ollama + @OpenWebUI. Step 1. Get OpenWebUI 1 Click Launcher Step 2. Download Llama3.1 from Ollama Step 3. No step 3!

Frank Shaw, Microsoft's Chief Communications Officer, explains why Microsoft can't lock down Windows the way Apple can.

@RetiefPeter @fjzeit Why do you blame MS for Crowdstrike? MS wants to control access to the kernel but is not allowed to because of the EU - Apple got a pass, for some reason pcmag.com/news/why-did-c… Also note that Crowdstrike was causing kernel panics on Linux last month theregister.com/2024/07/21/cro…

@fjzeit "Unix philosophy" and the failure of MS to adhere to it is what caused the recent massive failure. Linux is already everywhere, UNIX clones are even more common. Unix is the greatest sw/os ever written but people still spend huge amounts of money running inferior systems

Hot take: if Linux did become the defacto desktop OS it would quickly be flooded with for-profit closed source software and a user base primarily interested in outcomes and ease of use, and not the OS itself. Traditional Linux users would become a marginalised minority and Linux would become the new Windows. Be careful what you wish for. :)