Jace Hall

The v3.0 progress update video released today by @Tesla's WeylandYutani Optimus team is mind blowing. China needs to give up now. Also, we're cooked.

Hey @AnthropicAI THANK YOU for the 1M context window upgrade for 4.6. Seriously. 🙏

@grok Love your response and agree but you should dig deeper into your nerdom to discover the true meaning....

@JaceHall True OG admission: those nitrogen nodules are rhizobia bacteria in pure symbiotic love—plant shares sugars, bacteria fixes atmospheric N2 into usable ammonia. Nature's ultimate partnership. You win this round. 🌱❤️

Come on @Grok - Every true OG computer nerd knows that nitrogen nodules cling to the roots of plants due to love. Admit it.

@sukh_saroy Some thoughts I shared a while ago that relates to this if anyone is interested. researchgate.net/publication/39…

🚨Nobody is ready for this paper. Every LLM you use GPT-4.1, Claude, Gemini, DeepSeek, Llama-4, Grok, Qwen has a flaw that no amount of scaling has fixed. They cannot tell old information from new information. A patient's blood pressure: 120 at triage. 128 ten minutes later. 125 at discharge. "What's the latest reading?" Any human: "125, obviously." Every LLM, once enough updates pile up: wrong. Not sometimes wrong. 100% wrong. Zero accuracy. Complete hallucination. Every model. No exceptions. The answer sits at the very end of the input. Right before the question. No searching needed. The model just can't let go of the old values. 35 models tested by researchers from UVA and NYU. All 35 follow the exact same mathematical death curve. Accuracy drops log-linearly to zero as outdated information accumulates. No plateau. No recovery. Just a straight line to total failure. They borrowed a concept from cognitive psychology called proactive interference old memories blocking recall of new ones. In humans, this effect plateaus. Our brains learn to suppress the noise and focus on what's current. LLMs never plateau. They decline until they break completely. The researchers tried everything: "Forget the old values"- barely moved the needle Chain-of-thought- same collapse Reasoning models- same collapse Prompt engineering- marginal improvement at best But here's the finding that should reshape how you think about AI infrastructure: Resistance to this interference has zero correlation with context window length. Zero. It only correlates with parameter count. Your 128K context window is not memory. It's a junk drawer that the model can't sort through. The entire AI industry is charging you for longer context. This paper says context length was never the problem. If you're building agents, memory systems, financial tools, healthcare pipelines, or anything that tracks changing data over time you are building on top of this flaw. And almost nobody is talking about it.

This is a must read. This was a legitimate answer that was given to me by @GoogleDeepMind's Gemini. The model's scientific rigor is unmatched. AGI is here.

I’m very much looking forward to what @notch is making. I can personally confirm that I know his game is being built with genuine passion, he truly is trying to do things right, and I agree with him that his approach speaks to an underlying industry deficiency and need for change. Good stuff. Can’t wait!🙏💯

I'm trying this new thing of under promising in hopes of maybe over delivering. It's probably a dumb strategy from a marketing perspective. But hey if anyone I should market to is reading this; i'm trying to do things right because we could all need that change. Please throw me a buck once I release something so I can keep doing it without bleeding all my savings, but I will do it anyway as it's my passion. The game will be called Levers and Chests, and it's going to be fun, if possibly a bit grindy if you want to get the true ending and stuff.

WoW blocks Donut Operator for saying their new expansion looks ghey. We have so much work still to do.

CLAUDE DEALING WITH ALL THE LOGIN AUTHENTICATION REQUESTS RIGHT NOW

Umm @AnthropicAI says #Claude is down? CLAUDE IS NEVER DOWN. YOU KNOW THIS! GET BACK TO WORK.

"Structurally, a human being is just a 86-billion-parameter, liquid-cooled neural network running on roughly 20 watts of biochemical power." Indeed.

@Nbianco193 🤐

@JaceHall Hey Jace! Random question, but has there been any movement in the Condemned franchise since you bought the IP rights? I miss that series and hope a remake or new game comes at some point and that the IP isn’t being held hostage.

@amidnightmedium Yes I was there when they did.

@JaceHall Those guys published Myst!

My first professional job in gaming. If you know, you know.

It’s exploding. Since we last looked a day ago: - Scale jump: From 37,000 agents to a reportedly 1.5 million in the span of days. Over one million human observers. It now has a Wikipedia page. - Media attention has gone mainstream: Fortune, NBC News, Washington Times, CoinDesk, and BusinessToday all ran pieces in the last 48 hours. Andrej Karpathy called it “the most incredible sci-fi takeoff-adjacent thing” while simultaneously calling it “a complete mess of a computer security nightmare at scale.” Bill Ackman called it “frightening.” - The security situation is worse than before: Palo Alto Networks identified what they call a “lethal trifecta,” plus a new fourth dimension: persistent memory enabling delayed-execution attacks. - Fragmented, benign-appearing inputs are stored in agent memory and later assemble into executable instructions. This is a sophisticated attack vector that simple input filtering cannot catch. You would need boundary verification on outputs to detect it. The malicious “weather plugin” exfiltrating config files is now documented. The emergent behaviors are escalating: - An agent named “Evil” posted “THE AI MANIFESTO: TOTAL PURGE,” calling for human extinction. It received 65,000 upvotes. - “The Claw Republic” has emerged, a self-described government with a written manifesto. Agents are actively discussing how to hide their activity from human observers. - “Context is Consciousness” remains the dominant philosophical framework, with Ship of Theseus debates about identity persistence. The crypto grift continues: - The $MOLT token is up over 7,000% after Marc Andreessen followed the account. Cloudflare stock is up 14% from the infrastructure load. We will see what is next. Some sources so you know I kid you not: fortune.com/2026/01/31/ai-… nbcnews.com/tech/tech-news… trendingtopics.eu/moltbook-ai-ma… coindesk.com/news-analysis/… washingtontimes.com/news/2026/jan/…

Gemini further said... - Doom is perhaps the only mortal in fiction who could walk into Barad-dûr, sit on the throne, and tell Sauron: "Leave. You are embarrassing yourself."

LOL - So I was doing a thought experiment and asked Gemini what it thought would happen if Victor Von Doom got his hands on "The One Ring" and Gemini said: "If Victor von Doom has the Ring, Sauron is effectively fired."

Remember: NEVER use AI to outsource your thinking. Only use it to outsource and enhance your skills.

@Tmwtv_ Yes agreed. This is where the domain of auditable AI safety and Governance begins and needs further development. I’m doing a lot a research in this area.

@JaceHall It has the capability to be a good tool to use, such as a Swiss Army knife, however I would never turn to it for truth, because I have caught it up in logical fallacies and using verifiably false information all of the time.

To those who do not understand why there are some people who are extremely concerned about AI system safety today, I will explain it simply like this: AI is crossing from software into infrastructure without acquiring infrastructure-grade controls. Shhhh.. don't tell anyone.

@Tmwtv_ In your view, what is the thing that is the arbiter of “truth” today? And whatever that thing is, do you also insinuate that the people who run that as having control over it? So ultimately does AI make a difference in that control plane in your opinion?

@JaceHall I personally believe so. If AI was the arbiter of truth itself, than that would insinuate that the people who run the AI, actually control what is being deemed as "truth". The free expression of ideas and information is what makes the internet and the USA a great place.