>It was always explained to me as a mix between, 'are you going to fuck things up by being in an altered state' and 'is someone going to blackmail you to make you into a double agent?'

You are missing the foremost consideration - how critical/specialised/irreplacable is this person in their role and can we just ignore the positive test instead.

If you are good enough at what you do and management like you positive tests dont seem to matter if you make the right noises about it being a one off, retesting clean etc.

Counterpoint, you'v just changed publishing incentives. If I write 10 public interest stories, but notice a particular topic is making me more $$, I'll focus in on that.

You could end up with bad actors/PR management types promoting particular stories constantly or to detract from investigation they want less public.

That incentive already exists completely in ad-driven media.

At least this way, readers have a way to vote beyond just what happens to consume their attention and get their eyeballs on ads.

Thanks. I always like putting my ideas out there as I often get feedback on points I hadn't considered.

On this leading to reporting being pushed more towards what people consider in the public interest, I would argue that's a good thing. It's also maybe not that different to the current status quo if you look at how different TV or news networks cater to one target market or another.

With regards to your second point, sounds like you're essentially describing a Sybil attack. I agree with you that that's a serious concern. I'm just going to think aloud here so not sure where I'll end on this but let's see:

1. Having to make a micropayment for each upvote will at least put some cost to the bad actors, cf spam email vs spam SMS. That still makes it more amenable to abuse by the rich but I guess in the end it would come down to the balance of economic power and interest/participation between the masses and the elites?

2. How does it compare to the current model? You pay for a monthly subscription which in this context is an undifferentiated portfolio bundle of the news stories of a particular month. I guess that does make the cost of a Sybil attack higher because you can't just spend on the particular other stories trending at the same time as the one that you are currently trying to bury. For the good actor consumer this means that they generally have to pay for/support a news organisation that is broadly aligned with their interests or ideology. It does create this cliff effect though in that if I don't think I'll read enough stories in a particular month to justify my subscription then I won't pay for it at all.

3. I guess this mirrors the kind of debate and thought that's been going on in decentralized blockchain and web3 circles over the past decade or more about how to structure incentives that create broad distributions of power without too much concentration. I'm not up to speed on that so would love it if someone more knowledgeable would weigh in.

4. I had a thought about tying subscriptions to real world identities of natural persons but I don't like that from a privacy and censorship perspective.

5. I was reminded of Quadratic Voting (https://en.wikipedia.org/wiki/Quadratic_voting) which I believe would work well if there is a limited set of participants but I think won't protect you from a Sybil attack when it is cheap to create identities.

6. Building on points 5 and 4, what about this? You need a decentralized digital identity in the form of a DID which would have to be signed by some trusted party, either a government or some other institution. This could still be anonymous or pseudononymous in that when you go to the authority to have it signed, the signing authority checks that you currently don't have another active identity or any previous ones have been revoked (potentially problematic as I'm guessing the authority would have to maintain a link and I don't know if that's possible in a zero knowledge way). Alternatively identities are unlimited but expensive to create, say something like $1000 a piece so that creating 1M fake ones at least costs you $1bn. It's yours for life though so you can amortize the cost over time. In order to not cut out people at the bottom of the income distribution, they can use the government signed one but they lose the benefit of anonimity.

7. Once you have something like 6 then maybe QV is a good mechanism for apportioning micropayments on stories in the public interest?

Anyway, would love to know what current SOTA on this is in web3 circles rather than my speculations over my morning coffee.

I thought the same until my latest pixel refused to use the headphone jack to the car because it detected the hands free communications in the steering wheel as a microphone and decided to block audio out with notifications telling me to set up Google Voice Assistant first (get fucked).

GrapheneOS.org

Collision detection sensors do the job just fine without a screen though.

I have a 2016 vehicle with no console screen and they have saved me from hitting all sorts on things, and are sensitive enough to detect minor obstacles like long grass.

Moving data for 'heavy' workflows into the cloud is the most common performance bottleneck I see.

Is there a big enough dataset of 'good' code to train from though?

I (and lots of people) used to think the models would run out of training data and it would halt progress.

They did run out of human-authored training data (depending on who you ask), in 2024/2025. And they still improve.

> They did run out of human-authored training data (depending on who you ask), in 2024/2025. And they still improve.

It seemed to me that improvements due to training (i.e. the model) in 2025 were marginal. The biggest gains were in structuring how the conversation with the LLM goes.

> And they still improve.

But what asymptote are they approaching? Average code? Good code? Great code?

I'd argue that "good", or at least "good enough", is when they reach a point where it becomes preferable to spend your time prompting rather than reading and writing code. That the final output meets the feature specifications is more or less the goal.

A lot of developers are having a difficult time accepting that the code doesn't matter nearly as much anymore, myself included. The feedback cycles that made hot fixing, bug fixing, customer support, etc. so expensive, have shrunk by orders of magnitude. A codebase that can be maintained by humans is perhaps not a goal worth pursuing anymore.

To really see this and feel this, I think it's worthwhile to spend at least a weekend or two seeing what you can build without writing or reviewing any of the code. Use a frontier model. Opus 4.6 or Codex 5.3. Probably doesn't matter which one you choose.

If you give it an honest try, you'll see that a lot of the limitations are self-imposed. Said another way: the root problem is some flavor of the user under specifying a prompt, having inconsistent design docs, and not implementing guard rails to prevent the AI from reintroducing bugs you previously squashed.

It's a very new way of working and it feels foreign. But there are a lot of very smart, very successful people doing this. People who have written millions of lines of code over their lifetime, and who enjoyed doing it, are now fully delegating the task.

They ran out of passively collected data. RLHF allows them to gather deeper more targeted data.

There is a lot of RLHF effort around this.

Yeah this is just trading largely known & controllable labour management risks for some fun new unknown software ones.

You can negotiate with your human engineers for comp, you may not be able to negotaiate with as much power against Anthropic etc (or stop them if they start to change their services for the worse).

I always assumed mobile webpage misbehavior was to force you to use the app.

It could very much be confirmation bias, but I do feel like most "please use our app" popups appear after a mobile site breaks or refuses to load something

This is very prevalent in South Africa, to the point there is a legal cottage industry around verifying original documents vs counterfits (down to fingerprint testing, chemical analysis of inks).

Harder to direct waste heat in space if you dont have gravity for convection.