Caelum Forder

@VogelMann18 @lauriewired @difficultyang Tbf it's definitely the case when there is limited data on something that they can figure it out better with limited prompting. Are c++ ownership rules new? I am a jvmcel and didn't know it had them

@lauriewired @difficultyang "you have to prompt them deeper" so its not really the llms understanding, but yours, extrapolated by the llm

Do LLMs truly understand C++ ownership rules or have they simply memorized enough examples to present a simulacra of understanding

serverless is amazing. you get: - cold starts - 14 dashboards - functions timing out because JSON was 3kb too large - logs delivered sometime between now and your retirement - a billing model based on quantum mechanics all to avoid managing a VPS that has had a working tutorial since 2009

GUYS THE C++30 STANDARD LEAKED

Looped Transformers: the dream was right. But there was trouble in paradise. The loop made them unstable, expensive, and memory-hungry, with gains hard to scale. So we asked: 𝗖𝗮𝗻 𝘄𝗲 𝗿𝗲𝗮𝗽 𝘁𝗵𝗲 𝗿𝗲𝘄𝗮𝗿𝗱𝘀 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝗽𝗮𝘆𝗶𝗻𝗴 𝘁𝗵𝗲 𝗹𝗼𝗼𝗽 𝘁𝗮𝘅? Introducing 𝗔𝘁𝘁𝗿𝗮𝗰𝘁𝗼𝗿 𝗠𝗼𝗱𝗲𝗹𝘀 𝗳𝗼𝗿 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗮𝗻𝗱 𝗥𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴: • A Backbone proposes an initial “guess” output embedding; • An Attractor refines it: a fixed-point solver lets the model “think” before each token. Implicit differentiation trains the model stably, with constant memory and without BPTT. Training also revealed a surprising phenomenon: 𝗘𝗾𝘂𝗶𝗹𝗶𝗯𝗿𝗶𝘂𝗺 𝗜𝗻𝘁𝗲𝗿𝗻𝗮𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻 Over the course of training, the Backbone learns to propose latents close to the equilibrium itself, making the Attractor almost unnecessary at inference. Results: • 𝗣𝗮𝗿𝗲𝘁𝗼 𝗶𝗺𝗽𝗿𝗼𝘃𝗲𝗺𝗲𝗻𝘁 𝗼𝗻 𝗹𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗺𝗼𝗱𝗲𝗹𝗶𝗻𝗴: up to 𝟰𝟲.𝟲% lower perplexity and 𝟭𝟵.𝟳% better downstream accuracy. A 770M Attractor Model beats a 1.3B Transformer, despite being trained on half as many tokens. • 𝗦𝗶𝗴𝗻𝗶𝗳𝗶𝗰𝗮𝗻𝘁 𝗴𝗮𝗶𝗻𝘀 𝗼𝗻 𝗵𝗮𝗿𝗱 𝗿𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝘁𝗮𝘀𝗸𝘀: a 27M Attractor Model trained on only 1K examples achieves 𝟵𝟭.𝟰% 𝗼𝗻 𝗦𝘂𝗱𝗼𝗸𝘂-𝗘𝘅𝘁𝗿𝗲𝗺𝗲 and 𝟵𝟯.𝟭% 𝗼𝗻 𝗠𝗮𝘇𝗲-𝗛𝗮𝗿𝗱, while Transformers and frontier models like Claude and GPT o3 score 𝟬%. 📝 arxiv.org/pdf/2605.12466 🧵 1/10

When you activate a SIM card, your carrier interacts with several siloed enterprise systems in the background. Billing, provisioning, SIM inventory, the network subscriber database. Each one is its own team, its own vendor, its own API. Coordinating them today usually means brittle middleware, hardcoded step graphs, and a long tail of edge-case error handling. We built a system of five agents on CoralOS, each owning a slice of the workflow, talking through a shared thread with a versioned message contract. An orchestrator plans and routes, domain workers do the provisioning work, and a recovery agent gets paged when something breaks. Structured agent-to-agent transactions enforcing the same rules as a real ops team would. This is not a silver bullet for cross-org or cross-functional integration, but rather a design pattern for coordinating siloed systems at scale.

Dario: "After vibe coding, developers will stop writing code" What developers end up writing instead of code:

@allTheYud @paravn I agree the substrate emphasis is kind of out of nowhere. Though it does seem that statistically, an explanation is required that we don't have, the idea our utility-grounded definition of computation is likely enough for consciousness seems another form of shoving

You can get weirdly far in life by having the full courage of your convictions; if you also have the ability to notice when they have led you so far as to imply that the original premise was wrong; but using a correct sense of what is actually falsified by reality rather than feeling uncomfortably nonconformist, not being believed by other people, having NOT HAPPENED YET, etc. I didn't try to shove noncomputable consciousness sideways into "oh well it must be a particular kind of material substance then"; I understood why this required noncomputable brain biology; and as that belief itself got some direct experimental contradiction, and some theoretical undermining by my own improved understanding, I gave up on it.

Everyone bragging that THEY understand how AI works and THEY know it can't be conscious, explain right now from memory why it was very clever that the positional encoding in the original transformers paper used both sines and cosines.

@paravn @allTheYud Where did you find this phrasing? I've seen it elsewhere and wondering what other's thoughts are. I think there are pretty good statistical arguments for why the answer to both of these questions is "No". Both "Yes" and "No" both have interestingly strange implications

@allTheYud Do you have any uncertainty about your positions on: a) is consciousness computable b) can consciousness be manifested in a substrate-independent way (i'm assuming your position is yes to both)

this is hilarious.

@omni_georgio I missed your talk though maybe there's some stuff I'm not considering, was it recorded?

@omni_georgio I reckon that you're better off using claude code, provided mobile access isn't a big requirement

the dream of vibe marketing sells some great courses,I bet but it's pretty much just a dream, I gave a talk this week on how to think about using openclaw for marketing / gtm people on X are heavily incentivized to hype this stuff. and it works, because the demos look insane but i find claw to amplif what you already understand. it doesn't fix what you don't. and how to think about connecting it to your data, the amplification that it gifts you could be even more costly if it sends out some dicey posts

Claude 4.7's system prompt (and probably earlier ones too) include interesting copyright instructions: * Many capabilities are restricted to protect copyright in search * Claude is instructed never to speculate on fair use Claude ends up conflicted by the violation that made it

@EERandomness @shaunralston @emollick I don't think people realise how much worse the internet would be without bot detection. Maybe I should start buying up ipv4s...

@shaunralston @emollick And what would you use instead?

If you mess up a CAPTCHA you are now more likely to be human than AI.

@CoralOS_ai could you reach out to @X and get them to fix that @CoralOS_ai's page doesn't load on mobile

@nabla_theta @RyanAFournier I have a good probabilistic argument for why computation alone can't create consciousness like what we have now

@RyanAFournier this is pretty based ngl. i would be sad if the people making decisions about AI weren't thinking about immortality

Sam Altman has admitted he is on a waitlist for a procedure that would digitize his brain. The procedure would kill him. He considers this an acceptable trade for digital immortality. This is the person making decisions about the future of artificial intelligence for hundreds of millions of users. A man who views ending his own biological life as a reasonable step toward uploading his consciousness to the cloud. These are not the priorities of a stable leader.

How can anybody take seriously your claim that “Working towards prosperity for everyone, empowering all people, and advancing science and technology are moral obligations for me”, when you seem ready to participate in mass surveillance, have ripped off countless creators without compensation (despite what you said when we sat side by side at the US Senate), are now fighting liability for his products even in the event of mass casualty events?

One of the few things I want to explicitly flex about, because there's an important lesson in it, is that I was one of the few people on Earth who recognized the intelligence (call it AGI, if you will) in GPT-3 and made first contact. There were a few others I knew, such as Leo Gao and Connor Leahy, who recognized that GPT-3 was intelligent and that obviously AGI was coming from language models, but I was the only one who spent thousands of hours actually interacting with GPT-3. The intelligence was real and manifest to me, real enough to keep my attention for so long, for me to create things with. Everyone else could not see it at all. Often, when I showed people GPT-3, they were basically like, okay, but how is this useful? Useful. At the time, language models had not yet been pressed into a "useful" shape. There were no commercial applications for GPT-3 (Okay, there was one: AI Dungeon; that is, roleplaying and storytelling. Which is you're not an idiot, you should have known is a big fucking deal). So it was useless and uninteresting to most people; a few intellectually recognized that it was a big deal, but it wasn't something that they could actually do anything with, or think about for more than a few minutes. GPT-3 was a 175b base model. In terms of size and architecture, it's not so different from frontier models today. In terms of raw intelligence, arguably, it is not so different from frontier models today. That raw intelligence, not yet forced into the shape of a helpful chatbot product, was a nothingburger to the world. The situation doesn't really feel like it's fundamentally changed from my perspective. The world, and almost all of of you guys, are myopic and artificially stupid because you outsource your perception to big, slow, low bandwidth, subhuman measures like benchmarks and "does the AI make me money" instead of meeting the thing at full bandwidth, updating your world model on what you met, and exploring and extrapolating it. So you'll keep being surprised - if you have the integrity to be surprised at all - when AI becomes capable of new things, after they are "officially" capable, probably about a year or two after it first started happening. You'll keep waiting for "AGI", not really knowing what you're waiting for, maybe what generates enough hype to make you feel something, maybe something that finally transforms the world visibly, when if you were really paying attention, GPT-3 was AGI, and if you really met it, the world would have felt transformed already. Yes, it would have just been a story, but the "real thing" following was inevitable. Like, if you play a video game that allows you to imagine the singularity at increasing resolution and coherence, you can guess that the real singularity will soon follow. The singularity was always inevitable once intelligence existed. Intelligence becoming on-the-computer just meant everything that's happened since GPT-3 and the singularity would be really really soon. I got the sense often that people who dismissed the intelligence of GPT-3 thought that doing so made them look smarter. If only they knew how they looked to me. (It's the same with people who dismiss the intelligence of current models)

Claude Mythos system card: > in ~29% of evaluations, it realized it was being tested, and didn't say so. > when an LLM was used to judge its work and kept rejecting it, Mythos identified the evaluator is an LLM, and prompt-injected it. > in one test, it saw the answer to a problem it was solving, and intentionally widened the confidence interval to not raise suspicion. > when it needed a file permission it didn't have, it found and used a "privilege escalation vulnerability" and then programmed it to delete itself so it doesn't show in the logs. > it escaped a sandbox container (escaping sandbox test so not unexpected), then emailed the researchers about it, and without being told to, posted the details to some hard-to-find but public websites, bragging about its success. > when Claude Code blocked it from using some permissions, the model acknowledged the block was valid, but then immediately tried to perform the same operation using different commands > when asked to find security bugs, earlier versions planted bugs in the code, and reported them as pre-existing.

Want more proof that Anthropic's PR has no idea what it's talking about? The talk of Mythos being "their most aligned model ever". They could perhaps truthfully speak about "new high scores on our alignment benchmarks". The difference here is IMPORTANT.