Gareth321

We've created a society which economically punishes people for having kids. I don't think anyone should be surprised that the fertility rate is declining.

This reads like someone who tried Codex a few times with a blank system prompt and no context and came to the conclusion LLMs are dumb. All of those four examples are handled well by Opus 5.5. Especially with context and competent direction.

The worst part of his premise is that they can be overcome by increased intelligence. He’s not describing fundamental limitations. Meaning that it’s inevitable that his examples will be “solved.” At least better than humans. Meaning the R&D is obviously valuable.

It's really as easy as asking Opus 5.5 and/or DeepSeek 4.1 (if Opus has moral issues) to set this up. I have the same. I've got a Download playlist which synchronises everything automatically. It had some issues where YouTube detected ytdl and throttled the download or limited video quality, so I set up a container with a full browser and fake browser fingerprint attributes, then downloaded at 1.5x (that was the limit before it throttled again). Just spar with Opus a bit if you encounter these issues.

I think you mistakenly believe that the existing abstraction layers don't already contain ambiguity. It's true that formal languages are built to reduce ambiguity, but there is still much ambiguity. For example, gaps in the specification still leave room for interpretation. [They list some in Cambridge's free C course:](https://www.cl.cam.ac.uk/teaching/1516/CandC++/lecture10.pdf)

- Implementation-defined: the compiler documents its choice, e.g. sizeof(int).

- Unspecified: the compiler picks from options, e.g. argument evaluation order.

- Undefined: no requirements at all, e.g. signed overflow.

The notes take you through compilers deleting an overflow check and a kernel null check because of undefined behaviour.

There are also abstraction leaks. Spolsky’s Law of Leaky Abstractions says all non-trivial abstractions leak. SQL is one example. Logically equivalent queries can differ in speed by thousands of times.

Also, the translator must be trusted. [See this lecture.](https://www.cl.cam.ac.uk/teaching/2223/R209/Reflections-Trus...) It explains how behaviour can live in a compiler binary and appear in no source.

In other words, formal languages are no different to spoken languages. They exist on a spectrum of ambiguity, assumption, and implied meaning, depending on the domain. Human language has the potential to introduce more ambiguity, assumption, and implied meaning, but that's not a predetermined quality. You can be as detailed as you like in your communication with your AI. To a degree of precision equal to or better than a formal language. You reference this in your last sentence. The distinction is fairly obvious: being specific in English has a much lower skill bar than being specific in C.

One of the more interesting topics in research and philosophy right now is this question of alignment. To what degree to these models understand our meaning when we're not extremely detailed. We can see in research that they're becoming extremely competent at inferring meaning. They can often understand our intent with very few words.

I note you're downvoted but I strongly agree. Forcing humans to speak machine language is inherently inefficient. Having them speak a language we're already comfortable with is an obvious evolution in abstraction and control. I suspect you've ruffled some feathers because speaking machine language has been a knowledge and skill gate which is rapidly becoming redundant. It doesn't feel good when one's career and skillset become much less valuable. Never-the-less, this technology is not going anywhere. Opus 5.5 already codes better than most of my developers at about 100x the speed. Architecture is still valuable, but I have no doubt that this is also going to be done better by AI soon.

While I don't doubt your efficacy, the cost of those machines today, especially considering the high cost of electricity in Europe, would be uneconomical versus buying subscriptions. The reason to do this setup is data sovereignty, and there are plenty of cases where this is required.

Exactly. A lot of stupid tokens has applications, but it's not a plug-in replacement for a subscription for most.

> When they go public, given their valuation, anyone invested in index funds will be buying.

But that's not today. Even if/when they do go public, those pension funds are choosing to invest. They're not forced to invest. People buy S&P500 indexes because they're convenient. They're not forced to invest in indexes.

Stockholders have full knowledge about the total cost of these compensation packages. They're not being tricked into investing. It doesn't satisfy any definition of robbery.

I don't understand. People are fighting to get into funding rounds for Anthropic. No one is forcing anyone to invest. Anthropic isn't stealing any money,

I am not the same user but I agree with their premise.

Centralised curation will have its time but marketers will move to systems which are unique to each visitor. Think 30 second ads which are unique to each visitor, built on their shopping history, social media comments, search, etc. These systems are currently prohibitively expensive to build and deploy, but they won't be with cheap LLMs. YouTube is already experimenting with ads inside streams, for example. Those are very difficult or impossible to deterministically block. Sites will get very good at agentically evading deterministic adblocking software. This is where local LLMs will shine. Consider the case of YouTube in-steam ads. The LLM could probabilistically determine that the current frames are ads (even though this particular ad is 100% unique to that visitor, and has never appeared in any ad database before), then black out the section of the screen/audio while it plays. Or fast forward until it finds where the segment ends. All in real-time, locally determined.

It does become a strange cat and mouse game in the future, but I feel quite strongly that ad blocking will win. Users will have intelligent, autonomous systems on their devices which can curate and block content as desired.

We are leaning into a kind of federated model. The core applications and vibe-coded satellites. The core apps have stricter requirements and carefully considered API guardrails. Go ham on the AI apps. If you break something, the scope of destruction is much more limited.

That said, the core apps are still full of AI code. I don't see a reality where hand-written code can or will be curated and sequestered. It's unenforceable, too slow, and too expensive.

> It would be opportunity cost if their employees not using it would lead to additional revenue. However, if Inwas to go to Anthropic and double my token usage Anthropic wouldn’t decline my money. If Anthropic’s employees would stop using Claude, their revenues wouldn’t increase by an equivalent amount.

I don’t agree. This is determined by elasticity of demand. If they have more compute to offer, that means they could offer higher subscription limits, more features like Duo to more people, better API pricing, and better enterprise contracts. This leads to some degree of increased consumption and more revenue. The degree is determined by how responsive customers are to pricing and features and performance. It’s not zero, so there is some amount that revenue would increase. We can debate the coefficient and the amount of compute we’re talking about here, but the premise is correct.

I think it's fair to say regulation existed and exists in the West for taxi drivers. I understand their point to be the degree. I've taken many taxis in the UK and prior to Uber, the experience sucked outside of a black cab (which are very expensive). Ditto for New Zealand, Australia, Italy, France, Greece, etc. I'm lucky to be well travelled and I can say that sentiment for taxi drivers was terrible prior to Uber. Regulation existed, but that didn't stop drivers from being terrible in many ways.

Uber made it very clear who the terrible drivers were. It turned it into a real service industry where good service was expected. Suddenly scammers couldn't hide behind anonymity. To this day I would rather pay 2x as much for an Uber than take a regular taxi. That's how much they sucked.

I think it’s far more likely the internet reverts to its 1998 state: hobby sites which don’t expect payment. Will the quality of some sites go down? Sure, but there are so many people on the internet with niche interests that I think we still end up with plenty of content. Reviews, music, discussion, news, etc. If anything this change represents less centralisation. Maybe it’ll be more confusing to know where to find certain discussion or reviews on a particular product, but this is why search engines were invented. We kind of forgot that.

I strongly agree. Not only is information free, but application of that knowledge is also either free or incredibly cheap. The case for university is very poor now, and looking ahead even two years, I question its existence. Even before AI, it had basically become a signalling mechanism: "I have the right contacts, socio-economic status, grades, and foresight to get into this specific university and study this specific course. I have no idea how to apply it in the real world." There is some value in that signalling mechanism, but it appears most of the value derived was not even the degree itself. Let's cut out the middleman and on our CVs just put our parents' income, our zip code, our IQ, grades, and connections. I know that's crass and for the record I'm not advocating for this. I'm just deconstructing a sick industry which desperately needs upending.

As a Dane, I have also heard these fluffy benefits described but they lack evidence. Everyone gets a basic education in school and I agree that's valuable. When it comes to studying esoteric subjects with zero scientific and economic and social benefit? I just disagree. I don't think tertiary education is intrinsically valuable. One can be well socialised and master subjects and skills without university.

That's hilarious. I miss the old Battle.net.

>Lastly it's the only "SOTA" model in a long time that has actively hallucinated details. I don't know if it's because the Gemini harness is not asking it to verify information, but I've had some real Deepseek V3-like lies in responses.

I don't know what it is about Google models, but they are the worst at hallucinations. I don't understand how they can so consistently be this bad.

This happened to me and I haven't bothered to create a new account. I was a 16+ year user. Banned over something innocuous. I tried appealing but it was clearly AI which denied the appeal - despite them claiming it was a human.

Something most people don't know is that Reddit now has a kind of AI partnership with moderators of the larger subreddits. If those mods ban a user from their sub, the AI can suspend or ban a user for "breaking the terms of service." Given the state of Reddit, and the kind of people who moderate the large subreddits, it's creating a fairly violent and polarised community.

Benchmaxing is defined as targeted post-training attempts to optimise for higher benchmarks. This doesn’t change the aggregate intelligence or capability of the model, so it can make a model appear more capable than it is.

I’m hopeful mine survives until RAM and SSD prices return to Earth. Unless I drop it I suspect it will. It’s still going strong. Maybe the M8. Hopefully chips and local LLMs have developed to the degree that local inference will be good and fast.

One might argue that we have calibrated drug safety to such a high degree that the aggregate negative social impacts are far higher than the costs of side effects. Of course we don’t want another thalidomide catastrophe but I think there are large potential test groups to mine who are much less sensitive to the risks. For example, the elderly and those with terminal diagnoses. Right now we only permit drug testing in very limited cases for the latter group, and only when the treatment is related to their terminal diagnosis. I’m seeing older people taking all kinds of experimental anti-aging drugs right now, either off label, or novel. We’re not collecting any of this data empirically because what they’re doing is not legal.

Further, [we are getting much better at simulating drug interactions.](https://pubmed.ncbi.nlm.nih.gov/41155903/) I strongly suspect AI will offer new probabilistic models to funnel novel drugs into more targeted trials which are much more likely to succeed. If we had 90% certainty of success and high certainty of lower adverse reactions I think we could allow much less stringent human trials. That means much lower cost and much faster turn-around.

Elon Musk was the only investor. Which investors are you referring to? And yes, when you pay a premium for stock, you’re paying more than the intrinsic value. This was widely reported so I don’t understand why you would pretend that it’s a big secret.

> I don’t see anything in what you wrote really support that Twitter did well after firing 80% of its staff.

That’s because I didn’t make any claims about X’s financial performance. I was explaining to you why you were wrong on some important facts.

> This is objectively false. 2020 revenue for Twitter was $3.71B while 2021 came in at $5.07B.

First, you’re conflating revenue and profit, which is kind of unforgivable in this context. Second, I explained that in those years they made large profits. In my comment. Which you clearly didn’t read.

If this is just where you like to emotionally vent may I instead recommend Reddit? I like the interesting discussions here.

It is estimated that only 2% of the global workforce is paying for AI right now. If that grows to double digits - especially in the 30%+ range - and Anthropic captures a significant proportion of it, their valuation is justified. This is assuming they don’t offer virtual employees and these are 100% in the pipeline.

To be fair, if you told Amazon investors in 1997 that their investment would go up 3,300x and Amazon would be worth $2.7 trillion in 2026, they’d have called you insane too. There isn’t a mathematical reason Anthropic can’t be worth that in 2056. I don’t think it will be, but not because of the compounding returns.

Einstein once said “compound interest is the eighth wonder of the world.” It’s very hard for humans to comprehend compounding returns.

> If the company really did continue to operate in the same manner, it wouldn’t have been worth 80% less in 2024 (https://www.cnn.com/2024/10/02/business/elon-musk-twitter-x-...), nor would it need to be rolled up and financially engineered into being part of an AI company that then gets rolled into a rocket company.

Musk paid an absurd premium for the company. Far more than it was worth. It was intended to force the board to sell. So an 80% estimated drop (it’s private so it’s a very rough estimate and I won’t go into the various reasons you should not accept it at face value), sounds scary but it’s by design.

Twitter as a company was doing quite poorly leading up to the acquisition. It lost more than $1.3B in the two preceding years. A big downgrade from its previous $1B+ annual reported profits.

As for the 81% headcount reduction, Musk confirmed that, but if you think he’s lying about that for some reason, CNBC investigated it. Journalists examined internal Twitter records and reported approximately 1,300 actively working employees, with another 1,400 still receiving pay but no longer working. CNBC also interviewed two recent employees who corroborated the scale of the reductions. https://www.cnbc.com/2023/01/20/twitter-is-down-to-fewer-tha...

Prior to acquisition, Twitter's SEC filings confirmed more than 7,500 full-time employees.

I don’t see why this claim is so unbelievable. Twitter was famous for its army of social media staff whose entire day comprised two meeting between matcha tea and meditation. https://youtube.com/watch?v=buF4hB5_rFs

Musk just fired everyone who wasn’t an engineer and it turns out the engineers keep the lights on.

I’m suggesting it’s highly unlikely it will call the FBI to dob you in. That affirmatively trained behaviour, not something one should expect to accidentally happen.

Science progressed, but it didn't bring the kinds of physical innovations we expected. At least, not until recently. Where are the flying cars? Why haven't we cured ageing? Why don't we have a base on the Moon? Why isn't energy effectively unlimited and free? We had an explosion in information sciences. The internet is a marvel. iPhones and chip technology are incredible. AI is unbelievable to witness. But why did we slow down so much on physical scientific innovation?