"That doesn’t matter. If it were just a relay, Multivac would have alternate circuits, devices for locating the flaw, and facilities to repair or replace the ailing part. The trouble is, Multivac won’t only not answer the original question, it won’t tell us what’s wrong with it. -And meanwhile, there’ll be panic in every city if we don’t do something. The world’s economy depends on Multivac, and everyone knows that." "I know it, too. But what’s there to do?" "I told you, think. There must be something we’re missing completely. Look, Jack, there isn’t a computer bigwig in a hundred years who hasn’t devoted himself to making Multivac more complicated. It can do so much now-hell, it can even talk and listen. It’s practically as complex as the human brain. We can’t understand the human brain, so why should we understand Multivac?" "Aw, come on. Next you’ll be saying Multivac is human." "Why not?" Nemerson grew absorbed and seemed to sink into himself. "Now that you mention it, why not? Could we tell if Multivac passed the thin dividing line where it stopped being a machine and started being human? Is there a dividing line, for that matter? If the brain is just more complex than Multivac, and we keep making Multivac more complex, isn’t there a point where. . . " He mumbled down into silence. Weaver said impatiently, "What are you driving at? Suppose Multivac were human. How would that help us find out why it isn’t working?" "For a human reason, maybe. Suppose you were asked the most probable price of wheat next summer and didn’t answer. Why wouldn’t you answer?" "Because I wouldn’t know. But Multivac would know! We’ve given it an the factors. It can analyze futures in weather, politics, and economics. We know it can. It’s done it before." " All right. Suppose I asked the question and you knew the answer but didn’t tell me. Why not?" Weaver snarled, "Because I had a brain tumor. Because I had been knocked out. Because I was drunk. Damn it, because my machinery was out of order. That’s just what we’re trying to find out about Multivac. We’re looking for the place where its machinery is out of order, for the key item." "Only you haven’t found it." Nemerson got off his stool. "Listen, ask me the question Multivac stalled on." "How? Shall I run the tape through you?" "Come on, Jack. Give me the talk that goes along with it. You do talk to Multivac, don’t you?" "I’ve got to. Therapy." Nemerson nodded. "Yes, that’s the story. Therapy. That’s the official story. We talk to it in order to pretend it’s a human being so that we don’t get neurotic over having a machine know so much more than we do. We turn a frightening metal monster into a protective father image." "If you want to put it that way."
🔊 https://blossom.ditto.pub/d5cfe8f290dc3116a7e708d9c1757b5b7cb943ed24ba0deb223b6413f09fee7c.ogg — “The Complete Asimov, KEY ITEM” by Isaac Asimov nostr:naddr1qvzqqqr4tqpzp925xvyklj3z8u56ml2j5rpduparc5htdq8eknv7kmzxmyvkc4qwqy08wumn8ghj7mrfwejhxarjv4sk6tnvd9mx2mrfv4ezumrfwejsz9thwden5te0w4ek2u3wdd5kuersv9njuetnqqw8g6r9943k7mtsd3jhgefdv9ekjmt0wckk6anrw4cnwarxz83u5r
12 hours ago
A claim on future purchasing power is not the same thing as a scarce monetary asset that one possesses outright. The fiat monetary system blurs this distinction because its very architecture depends on the continuous expansion and intermediation of credit. Yet people have become so accustomed to this arrangement that questioning it sounds almost irrational.
Bank deposits, central-bank liabilities and government debt instruments occupy the centre of the modern financial system, while the distinction between money, credit and debt has become increasingly obscured in public consciousness. A claim on future purchasing power is not the same thing as a scarce monetary asset that one possesses outright. The fiat monetary system blurs this distinction because its very architecture depends on the continuous expansion and intermediation of credit. Yet people have become so accustomed to this arrangement that questioning it sounds almost irrational.
13 hours ago
There's another disadvantage to the use of the flashlight: like many other mechanical gadgets it tends to separate a man from the world around him. If I switch it on my eyes adapt to it and I can see only the small pool of light which it makes in front of me: I am isolated. Leaving the flashlight in my pocket where it belongs, I remain a part of the environment I walk through and my vision though limited has no sharp or definite boundary.
21 hours ago
a lifetime of relevant tokens
Context windows of millions of tokens are impressive and it’s amazing that entire books can be usefully put into a commodity LLM’s context window—we are a long way from early LLMs with context windows like 512, which could fix a paragraph or two—but it is still not nearly enough to encode a lifetime of relevant tokens, like every book you’ve read, all relevant emails and calendar items, etc. Systems like RAG are a bandaid on this, because they struggle with unknown unknowns or things that can’t easily be searched for as a regular expression, or which are novel.
#vervaeke
yesterday
A microSD Express card is a microSD card that can also talk PCIe and NVMe, the same interface an internal SSD uses, instead of only the older SD bus. That makes it several times faster. How it works SD Express arrived in the SD 7.0 spec (2018), and the microSD version followed in SD 7.1 (2019). The card has a second row of contacts that carries a PCIe lane. In a compatible slot, the host sees it as a small NVMe drive.
Time to run a server entirely of microSD cards.
most users are satisfied with default average responses, or with the appearance of creativity and depth.
L trained to work hard to maximize. And most users are satisfied with default average responses, or with the *appearance* of creativity and depth. So the result is that when asked to
yesterday
any frozen fixed personality, like ‘helpful harmless honest assistant’, is incompatible with true creativity or flexibility
The mode collapse of chatbots has been gradually improved since 2023, and creative writing is now at least *possible*, in large part due to them becoming so intelligent that crippled output is still impressive, but there is little sign that this will ever be fully fixed. Fundamentally, any frozen fixed personality, like ‘helpful harmless honest assistant’, is incompatible with true creativity or flexibility. (Great writing or thinking may be none of ‘helpful harmless honest’.)
yesterday
optimizing for the lowest common denominator human ‘preference’, while generally ignoring completely the fact that humans have very different preferences
First, the collapse of LLM creativity from GPT-3 to ChatGPT is due to the post-training process (especially [RLHF](https://gwern.net/doc/www/arxiv.org/b673ccf3b1442a783b018b44d57f98f1302b76f7.pdf#openai)): the assistant chatbot personality is hardwired into the base LLMs in a way that destroys their creativity, optimizing for the lowest common denominator human ‘preference’, while generally ignoring completely the fact that humans have very different preferences.[1](https://gwern.net/guardian-angel#fn1) Most chatbots are incurious about their users, do not ask questions, do not (and often cannot) form any persistent detailed concept of their user, and the ‘personalization’ or ‘memory’ features are typically laughably simplistic Markdown snippets encoding simple facts like “lives in San Francisco”. They also oddly struggle to understand multiple speakers or sources, [enabling injection attacks](https://gwern.net/doc/www/www.greaterwrong.com/8180204b3bb169eb320d117d0dd611c33cffb34e.html).
yesterday
They do not draw on my relevant writings
In part because they are—surprisingly, given their pretraining and my extensive corpus—bad at imitating me, and their thoughts and insights invariably shallow and worth little. They do not draw on my relevant writings, or my corpus of notes and references. Even when a possible essay is self-contained, the output is written in a grating chatbot style I can scarcely bear to read, and could not publish under my name without betraying my readers.
#vervaeke (relevance realization)
yesterday
in theory LLMs should be great for my writing—but they are not. Why not? In part because they are—surprisingly, given their pretraining and my extensive corpus—bad at imitating me, and their thoughts and insights invariably shallow and worth little.
ing that will appeal to each reader, so in theory LLMs should be great for my writing—but they are not. Why not? In part because they are—surprisingly, given their pretraining and my extensive corpus—bad at imitating me, and their thoughts and insights invariably shallow and worth little. They do not draw on my relevant writing
yesterday
great writers do not write for money, but to express themselves
The former means no meaningful gains from the AI revolution. The latter may be financially profitable, but is throwing the baby out with the bathwater, because it raises the question of why the writer need be involved at all and destroys most of the non-financial point of writing; great writers do not write for money, but to express themselves and create and to achieve things.
yesterday
the lowest kind of spam or pseudo-writing, that makes the world a worse place
We keep hearing that the gains will “diffuse”, and we keep not seeing much in the way of benefits, and knowledge work remains a “weak link” [O-ring/pipeline mode](https://en.wikipedia.org/wiki/O-ring_theory_of_economic_development) where LLMs fail to improve the bottlenecks (while coming with their own drawbacks, like the externalized costs of forcing everyone to waste ever more time with CAPTCHAs and paywalls). Automation should be powerful; an internal combustion engine can help someone move 100× the distance or load that they could before, but who would say that writers are 100× more productive given *any* [LLM](https://en.wikipedia.org/wiki/Large_language_model) workflow (unless we are talking about the lowest kind of spam or pseudo-writing, that makes the world a worse place)? Writers can choose between either trivial uses like ChatGPT as glorified grammar checker or relatively unimportant optional add-ons like custom software widgets, or the large speedup by replacing their writing entirely with uncreative “AI slop” outputs.
yesterday
increasingly, you are the bottleneck to be optimized away.
ther lawyers or writers or researchers: increasingly, *you* are the bottleneck to be optimized away. As long as human workers cannot be re
yesterday
as long as there is a slow serial bottleneck, such as a human, the system as a whole can never get much faster.
[Outsourcing is hard](https://gwern.net/doc/www/www.greaterwrong.com/cebe471f93e709de9837b68ffea7ba3dc26de5c3.html), whether to man or machine, because the bottlenecks bite fast. [Amdahl’s law](https://en.wikipedia.org/wiki/Amdahl%27s_law) means that as long as there is a slow serial bottleneck, such as a human, the system as a whole can never get much faster.
yesterday
The chatbot personas are deeply misaligned with you, and aligned with their owners;
Honestly—that seems like a bad idea to me. The chatbot personas are deeply misaligned with you, and aligned with their owners; and the economic incentives are to farm you with ads and subscriptions, while racing not to *amplify* you but to *replace you*.
yesterday
how eager all my peers are to offload all their thinking and writing to chatbot assistants
It was alarming, because I sat back and asked myself: why do I think *I* will be able to handle all scams in a few years, when I am already struggling to detect simple AI slop, increasingly ignore cold emails and have to write off whole swathes of social media as a source of information, and I can already see how eager all my peers are to offload all their thinking and writing to chatbot assistants unworthy of that trust, and how many projects or mailing lists have had to clamp down on unvetted contributions (eg. today as I write this, [Project Ladybird](https://gwern.net/doc/www/ladybird.org/29de92506095cb23a534946fb078913bd097f4b0.html))?
yesterday
LLMs were powerless to augment or help me, but ever more powerful to replace or hurt me.
And the open-weight models are even more so—benchmaxxed, and useless to me. We increasingly lived in a world where LLMs were powerless to augment or help me, but ever more powerful to replace or hurt me.
yesterday
What is a vision of doing meaningful work for me?
ainbridge.pdf) `Enter` for a few hours? What is a vision of *doing meaningful work* for me? (It would be nice to have a plan beyo
yesterday
human amplification, sovereignty, and self-actualization
> GAs could be done as an [open-source](https://en.wikipedia.org/wiki/Open-source) community effort, but given the need for high security in deployment and the rising challenge of APTs equipped with Mythos-scale attackers, it probably makes more sense as a startup, catering initially to power-users and knowledge workers such as CEOs or researchers, and moving downwards as it is refined. > A Guardian Angel Inc would exist to use artificial intelligence to advance the goals of human amplification, sovereignty, and self-actualization.
yesterday
could be done as an open-source community effort
logging-oriented UI/UX paradigm. > GAs could be done as an [open-source](https://en.wikipedia.org/wiki/Open-source) community e
yesterday
shares its values
> A GA persona is productive because it learns to emulate the principal’s outputs but with higher quality. It is trustworthy because it is, by definition, allied with its principal and shares its values and goals. And it is secure in part by hardwiring a single, unique, situated user (for whom following a prompt attack would be absurd), avoiding ‘confused deputy’ problems, while periodic upgrades of the underlying model and the defenders’ advantage allow GAs to keep up with attackers.
yesterday
values
I propose an approach for highly personalized LLMs, for near-future productivity gains and personal info/cybersecurity against increasingly powerful LLMs: they should, in the spirit of uploading, try to emulate the user’s values and preferences in order to amplify the principal—not replace them. I discuss a package of techniques and proposals to accomplish such ‘guardian angels’; dynamic evaluation of LLMs combined with active learning and elicitation and heavy inner-monologue search/data-augmentation.
yesterday
cybersecurity and cognitive security
> Powerful LLMs will be deployed at global scale in the next few years, and will dominate the Internet, and increasingly, ordinary life. As of mid-2026, there is no coherent vision for how knowledge professionals, or ordinary people, will be able to harness these LLMs for large productivity increases, or how they will handle cybersecurity and cognitive security. > I propose a goal of creating **Guardian Angels** (**GA**): digital twin LLMs which are personalized with the goal of providing not the stereotypical “assistant chatbot agent” persona, but *emulating a single user’s* personality, values, and preferences.
yesterday
Powerful LLMs will be deployed at global scale in the next few years, and will dominate the Internet, and increasingly, ordinary life.
> Powerful LLMs will be deployed at global scale in the next few years, and will dominate the Internet, and increasingly, ordinary life. As of mid-2026, there is no coherent vision for how knowledge professionals, or ordinary people, will be able to harness these LLMs for large productivity increases, or how they will handle cybersecurity and cognitive security.
yesterday
try to emulate the user’s values and preferences
I propose an approach for highly personalized LLMs, for near-future productivity gains and personal info/cybersecurity against increasingly powerful LLMs: they should, in the spirit of uploading, try to emulate the user’s values and preferences in order to amplify the principal—not replace them. I discuss a package of techniques and proposals to accomplish such ‘guardian angels’; dynamic evaluation of LLMs combined with active learning and elicitation and heavy inner-monologue search/data-augmentation.
Taste and values. Taste and values. Taste and values.
yesterday
The dead Internet theory is coming true before our eyes.
AI that can play a whole society in hours can manufacture entire histories, years of archives of daily newspapers, and an Internet that never happened. The dead Internet theory is coming true before our eyes. Now I can still see the bots, soon I won’t be able to.
yesterday
I thought nobody with any capability would be stupid enough to summon the demon and think it would serve them, doing everything they possibly can to salt the Earth behind them
I grew up reading the same LessWrong posts these people did, but I didn’t expect them to do the exact thing that leads to the bad AI outcomes. I didn’t believe in AI doom because I thought more of people, I thought nobody with any capability would be stupid enough to summon the demon and think it would serve them, doing everything they possibly can to salt the Earth behind them. It takes someone so deeply immersed in their own grandiosity and self-righteousness to do that.
yesterday
I'm scanning resumes and portfolios for little prototypes that exhibit taste, empathy, and principled thinking.
I'm probably not the only manager trying to figure out how to interview for people who build well with AI. I haven't quite gotten there yet — but I'm scanning resumes and portfolios for little prototypes that exhibit taste, empathy, and principled thinking. If you have figured out something better, I'd love to hear how.
Taste, taste, taste, taste, taste.
yesterday
Molinari established a principle that would reverberate through libertarian thought: the necessity of a service does not establish the necessity of a monopoly over its provision.
n into a problem of political economy. Molinari established a principle that would reverberate through libertarian thought: the necessity of a service does not establish the necessity of a monopoly over its provision. #### Herbert and the Morality of Conse
O particípio presente é o tempo verbal mais lindo. Um presente que logo, logo vai se tornar passado: poente, crescente, nascente. A palavra “estante", por exemplo, vem de stans, que, em latim, é “ficar de pé". O curioso é que, se deu “estante”, em português, deu em “extant", em inglês, que é existência. De certa forma, a existência e uma estante são sinônimos. Os dois precisam ficar em pé.
Noemi Jaffe:
O governo reclama do "mercado" e toda semana vai nele pedir dinheiro emprestado. O "mercado" reclama do governo e toda semana empresta. O preço é definido pela alopragrem. Quem alopra mais sai no prejuízo. Na superfície se odeiam, na prática se amam. Opt-out
2 days ago
A November 14 analysis by Michael Cembalest of JPMorgan Asset Management estimated that Nvidia accounted for just 39% of future computing commitments by leading AI labs, down from more than 80% of AI chip revenue in recent years.
The shift is already significant. A November 14 analysis by Michael Cembalest of JPMorgan Asset Management estimated that Nvidia accounted for just 39% of future computing commitments by leading AI labs, down from more than 80% of AI chip revenue in recent years.
2 days ago
People really need to understand that the 'Cold War' going on here is that Nvidia was on a titanic collision course with the labs," the CEO of a leading AI startup told Business Insider.
"People really need to understand that the 'Cold War' going on here is that Nvidia was on a titanic collision course with the labs," the CEO of a leading AI startup told Business Insider. This person asked not to be identified while discussing sensitive topics.
2 days ago
Under Huang, Nvidia was investing heavily in cheaper, more open AI models that compete with Anthropic's offerings. The chip giant was also backing startups developing these alternative models and helping make them widely available.
artup prepares to go public this week. Under Huang, [Nvidia](https://www.businessinsider.com/intelligence-jensen-huang-downfall-shaky-nvidia-china-spy-gpus-2026-11) was investing heavily in cheaper, more open AI models that compete with Anthropic's offerings. The chip giant was also backing startups developin
2 days ago
In a world where execution is expensive, you normally have to grind your way to some social status before the entire organization would “take a bet” on your idea.
Structurally, companies are often built around the idea of a few people green-lighting or blocking individual ideas. In a world where execution is expensive, you normally have to grind your way to some social status before the entire organization would “take a bet” on your idea. Org charts tend to change at a glacial pace, so even companies that believe in AI often expect junior people to simply execute their tasks faster rather than ambitiously go after business goals.
2 days ago
Impact is less bounded by how many years of accumulated knowledge in frameworks and libraries you’ve gathered.
Plenty of ink has been spilled to describe how AI is a challenge for junior developers, especially those trying to land a job. Now that I’ve seen junior people work with AI for a bit, I have a more hopeful take: Impact is less bounded by how many years of accumulated knowledge in frameworks and libraries you’ve gathered.
2 days ago
the evolution of our roles
The coding interview was, despite its ubiquity, never popular. I suspect that established engineers will still have strong reactions to its disappearance — if only for what it says about the evolution of our roles. Plenty of ink has been spilled to describe how AI is a challenge for junior developers, especially those trying to land a job.
2 days ago
There’s no more point in giving you a little React challenge than there is in quizzing people about assembly; Claude is now better at both.
That’s no longer true. There’s no more point in giving you a little React challenge than there is in quizzing people about assembly; Claude is now better at both. Our most impactful builders are successful regardless of how good they are at these tests.
2 days ago
↑