ulrikrasmussen

I also don't think it will kill creativity in the space, although my phrasing might have suggested that, just that it might diminish it and make implementation styles more conformative.

On the other hand, we might trade that for creativity in other more high level designs where things like coding style is an unimportant detail for the experiment. For example, I have started playing around with microcontrollers because I can have the LLM generate the software while I focus on learning how to solder and think about how to modularize my hobby automation system. That's creativity and taste at a higher level which I didn't have access to before because I didn't want to invest my precious free time in learning the software ecosystem.

An important feature of taste is also that it is highly variable between people, and because there is no system or proof of good taste, we cannot really say that one person's taste is objectively better than another's. This is what causes one person to build a large software system in a classical OOP language while another builds almost the same system in a typed functional language, each following their own internal logic of what is good taste. In the end we learn something from it, and maybe someone who worked on both projects refine their taste and starts preferring something that lies in between the two paradigms and comes up with a new interesting programming language.

One thing I fear is going to happen when more and more people use LLM's for coding is that they will default to the "average taste" of the LLM, and we will stop getting new interesting ideas about how to build things.

I lost a family member to anorexia, and the only thing I really regret is that I didn't suggest that we try psychedelic therapy now that everything else had failed and things looked pretty grim. I mostly blame myself for not pushing the idea. But I am also angry at the war on drugs which has caused all use of psychoactive drugs to be associated with such taboo.

There's Curve Pay which works on GrapheneOS

I ended up brewing just one cup of filter in the morning, then drinking one more cup when I arrive at the office. I have a rule to stop before 9am which I found is the latest I can do caffeine without disrupting my sleep. I think I am one of the slow metabolizers.

I quit cold turkey and consumed around 1200mg/day, not by drinking energy drinks but coffee, 6 strong cups of filter or more. I made sure to also quit caffeinated sort drinks and switched to herbal tea.

The immediate withdrawal symptoms started after around 24h and consisted of headache, cold sweats and brain fog that lasted around three days and seriously impeded my ability to work. However, the psychological craving just... never really got better once it plateaued. I always felt a bit more understimulated than before I stopped, and a bit less motivated.

I kept it up for about 1.5 years or so before I decided it wasn't worth it. The first cup of coffee seriously gave me euphoria comparable to a low dose of amphetamine. That was so pleasurable that I tried to keep my consumption to one cup per week, but even with that spacing I felt the tolerance build up again.

A password manager has an almost as efficient UX

I don't know if I have been damaged by reading so much AI, but I swear the narrator of that video had the script for it written by Claude without any of those rules applied. There are so many platitudes and AI-isms in the way he speaks:

    "now the honesty - those translation gains were real but small"
    "and here is the honest part"
    "same disease but with different symptom"
    "so that brutal 3% number was a Claude quirk, not a law of nature"

And the goddamn emojis. I get it, people use them when texting to make the writing a bit more chummy, but I can really live without them everywhere else.

Diagnosis of rare diseases is a subset of "true" statements, no?

Unless you have the technology from The Matrix which can inject knowledge into your brain, then no, you definitely cannot delegate understanding.

You can delegate the responsibility of understanding to an LLM, but that is not the same.

No, it doesn't, because I gave up using them.

Sure, but at that point it hardly makes sense to discuss any of the details of the implementation since we are just treating the LLM as a black box. For all we know, it could be maintaining its own specialized programming language made just for the application.

I don't really understand why we need an LLM-specific programming language - all of the properties you list are properties that modern programming language designers already try to achieve.

> - Inspect-able, testable, and reviewable in small pieces

This is more a property of the architecture than a property of the language, although some programming languages support it better than others.

There's very little evidence, if any, to suggest that slop is actually "good enough". I have interacted with multiple apps that clearly were written using AI by people who did not actually spend the time understanding what was made, and it most certainly was not an experience that I would classify as "good enough".

I know this may be some kind of reverse survivorship bias and that I simply do not notice all of the slop around me that somehow is truly "good enough", but I think I need to see some concrete examples to convince me.

Can't remember who said it, but I think the most poignant commentary about the prospects of LLMs completely replacing humans is the simple observation that yes, you can delegate many if not all of your tasks requiring thinking to LLMs, but it is inherently impossible to delegate the task of understanding. If you hope to write an efficient prompt which gives you what you think you want, you need to actually understand what you want, which also requires you to understand what you currently have. Programming languages are a medium for transferring and - not the least - maintaining shared understanding of a software system in terms of the basic abstractions that the programming language provides. The need for good programming languages does not go away just because we use LLMs to write the code, on the contrary: would you rather review LLM-written assembly or LLM-written Haskell?

I use git every day and can do quite advanced stuff with it, but I do all of my work using magit in emacs. I don't even use emacs for writing code anymore, I only use it for git. I just can't be bothered with using the CLI, it is too painfully inconsistent.

The only downside to this is that it is hard for me to help people with git problems since I can only tell them what conceptually has to be done, not how to accomplish it using the CLI.

I can have an internal monologue about something completely unrelated to the story while reading it out aloud, but I won't be able to follow the story. I don't think I make mistakes, but I probably fail to put the correct intonation on sentences which require higher order thinking to infer the emotions of the characters in the story.

If you are the kind of person who unlocks the bootloader and installs GrapheneOS, then definitely not.

Instead of trying to whack a mole on all addictive mechanisms, just ban the business model driving all of them: targeted advertising.

Exactly the same setup for me. It works great and the whole setup consists of a single Vagrantfile which I maintain via Claude itself. All relevant dotfiles are synced to a folder on the host so e.g. conversations are preserved across reprovisioning of the VM. It's a simple solution, but very flexible.

I had a similar experience as you. We normally use Kotlin for everything, but last year we had to do a small project in Python. Setting up the tooling and choosing the tools is quite overwhelming, and the inconsistency between what the tools actually consider type errors is incredibly frustrating. I am actually happy that the project failed and that we don't have to work in that environment anymore.

I think Python is probably good for many things, including scripts and as a starter language, but I don't understand how anyone can stand writing large software systems in it.

It writes code as if the audience is you, the user of Claude, and not other developers reading the code in the future. I found that it helped to instruct it to keep in mind who the audience is and only write comments that describe the current state of the code and never describe anything that can just be inferred from the git history. I found that that helped, and I almost never see these nonsense comments anymore.

What legal requirements are you referring to?

> the usual flow is rather like Google's OAuth - the site needing you to prove your age rediects to the provider (Google, or whoever), who asks questions to verify your identity, and then replies with "over 18" or "not over 18".

This is false. There are many problems with age verification, but the EU approach does not involve the id provider in the verification flow. The site requiring verification presents a QR code which encodes a presentation request and the provider controlled URL which is to receive a presentation of the age credential, and then the smartphone generates a unique presentation signed by a device bound key and sends it to that endpoint.

It is however true that in addition to the one bit of information saying age>18, what is also revealed is the public key of the identity provider. This will at least reveal the nationality of the credential holder and - in the case there are multiple issuers within a nation - may reveal even more information about their demography.

> Someone over 18 person could indeed decide to help others prove age, but they would either have to do it in person or be willing to loan their unlocked security element to those others.

You can still automate age-verification-as-a-service using a physical autoclicker on a smartphone with the camera oriented at a screen showing QR codes. I expect this to happen, and for that reason I also expect true anonymity to be something that will last a year or so until it will be politically decided to fix the issue using some central party that verifiers have to contact to impose rate limiting.

The article claims this is what the ZKP scheme reveals:

> Here is cryptographic proof that I hold a valid credential proving I am over 18. You can verify the proof, but you learn nothing about who I am.

Is this true though? I am genuinely interested as I haven't looked into the details, but must the user not at least also disclose who the issuer of the credential is so the verifier can verify it against a public key? This also reveals at least the nationality of the user and could be misused to block access to foreigners using VPN.

You can't really prevent that unless you design a system which is inherently designed to track people, e.g. by phoning home to the issuer on each credential verification. The system being deployed right now is based on the issuer issuing batches of single-use credential tokens to device-bound single-use keys, which on the plus side means that colluding verifiers cannot use age credentials as cookies to track people. It is still vulnerable to colluding verifiers and issuers though, because the issuer can de-anonymize the tokens (it knows them and their linking to the identity of the user). This scheme also means that if the keys that the tokens are issued to are not device bound, then it is trivial to copy the age credentials to someone else.

To my knowledge, even more sophisticated ZKP schemes still rely on device bound keys to protect against duplication.

The software layer in age verification is not necessary to trust though. The worst that could happen is that a compromised software layer steals your age credential, but it is by design anonymous so you don't risk getting your money or account stolen or anything. This makes it a different threat model from the banking case.

I think the reason these systems require device bound keys is because the government is concerned with easily mass-produced forged age certificates. With software keys you can get an age certificate which can be copied instantly to a large number of devices, with hardware keys the government knows that the certificate is tied to a single physical unit.