I'm sure we'll all get replaced eventually. But we won't be alone, the entire society is going to be upended. So we will be part of a vast social and economic experiment in which nobody will be in a position much different from ours. Hopefully, solutions will be found to avoid complete social breakdown.
So I feel like I'm on the Titanic- the ship is sinking and we're going all to hit the water eventually, the trick is to try to keep dry as long as possible. If you've been in an organisation for long, and you know the business, the people, the organisation, have domain knowledge and can contribute beyond translating to code someone else's requirements... These are all valuable assets that will keep you relevant and useful for some time.
To me, it just sounds as he didn't understand where the message was really coming from:
> Fuck you people. Raping the planet, spending trillions on toxic, unrecyclable equipment while blowing up society, yet taking the time to have your vile machines thank me
Yes, the sender organisation is not the one doing all this, but merely a small user running a funny experiment; it would have indeed been stupid if Anthropic had sent him a thank you email signed by "Opus 4.5 model".
This is just a funny experiment, sending 300 emails from in weeks is nothing compared to the amount of crap that is sent by the millions and billions every day, or the stuff that social media companies do.
> When you slow down your eating speed, say to half or a third your default speed, you get much more enjoyment out of a smaller amount of food. The extra attention given to each bite allows more of the “good stuff,” whatever that is exactly, to reach you.
I wonder if this contributes to a good chunk of the experience of fine dining. When you get served expensive food in micro-portions that are accompanied by long explanations, you don't gobble it down but take as much time as possible to savour it.
For a while I've thought so many of us are afraid to slow down because we might feel sad, as sadness is often one of the slowest emotions.
Yet this quote has me thinking that maybe we fear slowing down because of the consciousness overload. That we get so overwhelmed by our senses. Maybe even that can lead to the tears of being so alive.
> The only plausible explanation is that it was tall in order to ensure access to light.
Thought the same, but that implies both that it grew in very dense "forests" (mono-species because there were no competitors) and probably that it had leaves (because otherwise trunks don't occlude much light).
Although, counterexample: why do (some) cactuses grow tall? Claude provides these explanations that might apply:
Water collection and storage — Height means more volume for water storage. A large saguaro can hold thousands of liters of water in its stem, which is crucial for surviving long droughts.
Temperature regulation — Being taller gets the growing tip and flowers farther from the scorching ground surface, where temperatures can be extreme. The ground in deserts can reach 70°C (160°F), while air temperature a few meters up is significantly cooler.
The 2 explanations given for cactuses seem non-applicable to Prototaxites, as the fossils appear to have formed in some swamps with abundant water.
According to the linked paper, the structure of the stem of Prototaxites contained several kinds of tubes, which might have formed some kind of simple vascular system, able to extract the water from the soil and circulate it through the body.
You are right however that the plants among which Prototaxites was growing had a much smaller height so the competition with them would not have been a strong reason for its height and for the competition between Prototaxites individuals there is no evidence that they would occlude much light.
Still, I am not aware of any better explanation for its height. At that time there were no flying insects. The terrestrial vertebrates and bigger arthropods were predatory. There were a few groups of non-predatory arthropods, i.e. millipedes, mites and springtails, some of which might have been able to feed on Prototaxites tissues, but such small arthropods are likely to have been able to easily climb its stem, so it seems unlikely that its height could have provided any protection for its reproductive parts.
Besides avoiding shadows, there is another explanation for the great height, but that is also applicable only to organisms able to capture solar light. As there is evidence in its isotopic composition that Prototaxites was not phototrophic, any explanation based on capturing light must involve a symbiotic alga. A great height could have helped with the ascent of water through the stem of Prototaxites, due to capillarity and evaporation at its top. However this explanation requires for the pumped water to be useful somewhere high in the stem, which would be the case if the water were given to a symbiotic alga, which would provide food in return.
Confirmed.
I find article 10 of the European Convention on Human Rights particularly enlightening:
"1. Everyone has the right to freedom of expression. This right
shall include freedom to hold opinions and to receive and impart
information and ideas without interference by public authority
and regardless of frontiers. [...]
2. The exercise of these freedoms, since it carries with it
duties and responsibilities, may be subject to such formalities,
conditions, restrictions or penalties as are prescribed by law and
are necessary in a democratic society, in the interests of national
security, territorial integrity or public safety, for the prevention
of disorder or crime, for the protection of health or morals, for
the protection of the reputation or rights of others, for preventing
the disclosure of information received in confidence, or for
maintaining the authority and impartiality of the judiciary."
To me it reads as "you have the right to free speech without interference by public authorities, except in all cases where public authorities want to interfere in whatever form and for whatever reason".
Sorry, but if you don't trust your coworker to review and shape the AI's output, why would you trust them with actually writing the whole document?
And if you think that at this point you could have done it yourself, then why don't you? The only important thing is that the document is fine, if it takes you too much to verify it then you need to trust your colleague, that was their job.
Signals of competence and diligence help build and reinforce trust.
Crafting a message for your known friend/ coworker almost always comes through in how it is written and structured, because it always weighs the arguments against the context of the business needs, communication norms, shared understanding of what's important, all implicit contacts about how we work together, the long term vision we shared over beers, the teams messages that the CEO sent three days ago, etc.
In a pure design doc - like a wiring diagram + 3 code snippets, this is a non-issue, so just ignore what I said (but consider it possibly).
In a doc for communication, especially of ideas, this is paramount.
The issue isn't using AI tools to write these "RFC" style docs. The issue is that in the very likely event that the output does not contain any of those very important bits (because how could it??), then we are in a situation where 1) this person I trust has lost some of that trust by not acknowledging any of the above or addressing it or structuring it in a useful way or 2) that person didn't try.
This is why communication is a valuable skill. It's always been implicit that effort slowly adds those and many more features of a good doc. Now it's explicitly not doing that, but still feigning effort with lots of formatting, etc. It moves the "add the important intangibles" from the writer to that reader, and like code review, that's laziness. We explicitly did not hire an AI, we explicitly hired a person, and that person should be filtering the world's noise through their valuable experience, and at least telling us that they did that. "I reviewed this and stand by it" is a very low bar to achieve, so I dont understand why there can be any pushback.
Is there a 3rd option here I'm missing?
EDIT: I can temper this a little bit. This is how I like to work. There might be a cadre of devs who are comfortable slinging 10 page unreviwed documents at each other. I'm fine with their existence I just think it's better to carefully review text from a close coworker because they deserve that time, and so I expect the writer would do at least one review themselves out of courtesy. I don't think any of this is arduous. If my boss told me to spend more time reviewing than the author was willing to spend writing, then I would either get comfortable with reduced dev output from this new DDOS, or find a new job.
While I understand perfectly the feeling of uneasiness that comes with reading something that was at least contributed by an LLM, my point is that any lack of trust is just on your side until you can prove that the document contains mistakes or omits important information. The person that produced the document is the one responsible for it- they still need to know and review accurately its content.
However there's another aspect that irks me, and it's the idea that the prompt was much shorter than the document itself. Well if this is the case, then the problem isn't much the use of LLMs, but rather that you consider obvious that your documents are mostly fluff that can be compressed to a bullet list. If the final document is much harder to verify than the information it contains then it means that you're wasting time and resources to ofbuscate rather than clarify.
> The person that produced the document is the one responsible for it- they still need to know and review accurately its content.
That's exactly what I'm saying, so we agree perfectly there.
> However there's another aspect that irks me, and it's the idea that the prompt was much shorter than the document itself.
That's not part of my argument, it's an assumption on your part.
> If the final document is much harder to verify than the information it contains ...
This is precisely the problem. LLMs generate far too much text for their information content, and often contain subtle errors etc etc (I dont need to rehash two whole HN threads here, the point is made). If a coworker sent me the bullets, I'd be happier. Because the cost of generating arguments is now effectively zero, it's imperative to use restraint to get back to the concise, high SNR message. Precisely because otherwise it becomes a DDOS on the reader.
I might be completely off road, but I can't help thinking of convolutions as my mental model for the K Q V mechanism. Attention has the same property of a convolution kernel of being trained independently of position; it learns how to translate a large, rolling portion of an input to a new "digested" value; and you can train multiple ones in parallel so that they learn to focus on different aspects of the input ("kernels" in the case of convolution, "heads" in the case of attention).
I think there are two key differences though: 1) Attention doesn't doesn't use fixed distance-dependent weight for the aggregation but instead the weight becomes "semantically-dependent", based on association between q/k. 2) A single convolution step is a local operation (only pulling from nearby pixels), whereas attention is a "global" operation, pulling from the hidden states of all previous tokens. (Maybe sliding window attention schemes muddy this distinction, but in general the degree of connectivity seems far higher).
There might be some unifying way to look at things though, maybe GNNs. I found this talk [1] and at 4:17 it shows how convolution and attention would be modeled in a GNN formalism
>A single convolution step is a local operation (only pulling from nearby pixels), whereas attention is a "global" operation.
In the same way where the learned weights to generate K,Q,V matricies may have zeros (or small values) for referencing certain tokens, convolution kernels just have defined zeros.
The whole reason for the first "AI Winter" was because people were trying to solve problems with smaller neural nets, and of course you run into problems during training, where you can't get things to converge.
Once compute became more available, and you had more neural nets, and thus more dimensionality (in the sense of layer sizes), during training, you had more directions for gradient descent, so things started happening with ML.
And all the architectures that you see today are basically simplifications of the fully connected layers with max dimensionality. Any operation like attention, self attention, or convolution can be unrolled into matrix multiples.
I wouldn't be surprised if Google TPUs basically do this. It seems to reason that they are the most efficient because they don't move memory around, which means that the matrix multiply circuitry is hard wired, which means that the compiler basically has to lay out the locations of the data in the spaces that are meant to be matrix multiplied together, so the compiler probably does that unrolling under the hood.
Have you tried asking e.g. Claude to explain it to you? None of the usual resources worked for me, until I had a discussion with Claude where I could ask questions about everything that I didn't get.
In some respects, yes. There is no single human being with a general knowledge as vast as that of a SOTA LLM, or able to speak as many languages. Claude knows about transformers more than enough to explain them to a layperson, elucidating specific points and resolving doubts. As someone who learns more easily by prodding other people's knowledge rather than from static explanations, I find LLMs extremely useful.
So you have one of the scientists at the forefront of quantum computing theory telling you that he has no idea if quantum computing is already in a much more advanced state that he himself knows about?
If results in quantum computing would start to "go dark", unpublished in scientific literature and only communicated to the government/ military, shouldn't he be one of the first to know or at least notice?
So I feel like I'm on the Titanic- the ship is sinking and we're going all to hit the water eventually, the trick is to try to keep dry as long as possible. If you've been in an organisation for long, and you know the business, the people, the organisation, have domain knowledge and can contribute beyond translating to code someone else's requirements... These are all valuable assets that will keep you relevant and useful for some time.
reply