What, I ask, is the point of having laws and rules if you can just ignore the ones you don't like?

Its just a name, who cares?

Not me.

…but, if you break the law, you break the law. Not maybe maybe who cares, its not me being water boarded, I dont care…

If you break the law. You break the law.

Otherwise, who gives a duck what congress says?

Just fire them all and crown Trump King of America.

I’m being facetious. …but maybe its more of a big deal than you superficially pretend it is.

It’s just another case of the administration blatantly breaking the rules.

…so, you know. If youre ok with no laws or rules, I guess its fine.

Seems a bit chaotic to me. I prefer my governing body to be… marginally bound by some kind of responsibilty to something or someone.

- tell the agent to write a plan, review the plan, tell the agent to implement the plan

- allow the agent to “self discover” the test harness (eg. “Validate this c compiler against gcc”)

- queue a bunch of tasks with // todo … and yolo “fix all the todo tasks”

- validate against a known output (“translate this to rust and ensure it emits the same byte or byte output as you go”)

- pick a suitable language for the task (“go is best for this task because I tried several languages and it did the best for this domain in go”)

The practice is:

- simple

- effective

- retains control and quality

Certainly the “unsupervised agent” workflows are getting a lot of attention right now, but they require a specific set of circumstances to be effective:

- clear validation loop (eg. Compile the kernel, here is gcc that does so correctly)

- ai enabled tooling (mcp / cli tool that will lint, test and provide feedback immediately)

- oversight to prevent sgents going off the rails (open area of research)

- an unlimited token budget

That means that most people can't use unsupervised agents.

Not that they dont work; Most people have simply not got an environment and task that is appropriate.

By comparison, anyone with cursor or claude can immediately start using this approach, or their own variant on it.

It does not require fancy tooling.

It does not require an arcane agent framework.

It works generally well across models.

This is one of those few genunie pieces of good practical advice for people getting into AI coding.

Simple. Obviously works once you start using it. No external dependencies. BYO tools to help with it, no “buy my AI startup xxx to help”. No “star my github so I can a job at $AI corp too”.

Great stuff.

Using nano banana does not require arcane prompt engineering.

People who have not learnt image prompt engineering probably didn't miss anything.

The irony of prompt engineering is that models are good at generating prompts.

Future tools will almost certainly simply “improve” you naive prompt before passing it to the model.

Claude already does this for code. Id be amazed if nano banana doesnt.

People who invested in learning prompt engineering probably picked up useful skills for building ai tools but not for using next gen ai tools other people make.

Its not wasted effort; its just increasingly irrelevant to people doing day-to-day BAU work.

If the api prevents you from passing a raw prompt to the model, prompt engineering at that level isnt just unnecessary; its irrelevant. Your prompt will be transformed into an unknown internal prompt before hitting the model.

1) clearly technical

2) reproducible

3) has a clear failure condition

Not be a suitable candidate for S/O?

Did we step into a dimension where only "How do I print('hello world')?" is a valid question while I wasn't watching, because it has a trivial one-line answer?

Hard questions doesn't mean they're bad, it just means many people aren't competent answer them. The same goes for obscure questions; there might just not be many people who care, but the question itself is entirely valid.

Does that mean they're not suitable for S/O?

I... can't believe anyone seriously believes that hard niche problems are too obscure or too hard for S/O to be bothered to grace themselves with.

It's absurd.

It just baffles me that a question that might take some effort to figure an answer out to might 'not be suitable' to S/O.

What if moderators had to actually have karma from recently answering questions or they lose mod privileges?

Wouldn't that be a fresh change. You'd have to actually work to be a mod.

...

It shouldn't be controversial. That mods currently make visitors unwelcome is disgrace. :(

That SO incentivizes that behavior is ridiculous.

Yes, you can argue that [person] is [performing an action] because they believe, from their POV that [reason1, reason2, reason3].

> Or does [what person believes] not matter when they act upon it?

Yes.

What people choose to believe is distinct from fundamental baseline reality.

Let me put it another way for you; if I believe that fairies have invaded from space and I go out smashing peoples cars because, I personally, believe that this will make the fairies go home…

…does it help to argue about whether I believe in fairies or not?

It does not.

The arguement must be about whether fairies exist in baseline reality or not.

What I believe is not a point worth discussing.

…so, to take a step back to your argument:

Does he believe this will help? Who. Gives. A. Flying. Truck? Does it matter what he believes? Can we speculate what he thinks? It’s a useless and meaningless exercise and a logical fallacy; because anything can be justified if the only criteria are “you believe it will work”.

The discussion worth having is, in baseline reality, will it actually help?

Which is what the post you are replying to is addressing; but instead or following that up, you’ve moved this discussion into a meaningless sub thread of unprovable points about what people may or may not believe.

Which is why you’ve received my downvote.

> Should I pick some other experts to listen to?

I think it's reasonably clear that you haven't spoken to an expert in this field.

> I'm sorry if I don't take the word of some random guy's dentist over multiple meta analyses in major medical journals.

Are you certain you're competent to review and understand the literature on the topic? It takes a lot of time and effort; that's what dentists do as a job. That's why they have to go to school. That's why random people on the internet do not do dentistry.

If you don't trust my dentist, then talk to your dentist.

This is literally my point: I'm not telling you how it is; I'm telling you, talk to someone who knows what they're talking about; and, don't believe that you are an expert because you put some trivial amount of effort into investigating it yourself.

You can't be an expert at everything. No one can.

As some point, you have to trust other people.

At some point, you have to accept that your random wikipedia page and 5 minutes on google is not a convincing argument.

This is right up there in the conspiracy theory territory.

Rational discussion means listening to experts and admitting that you are not an expert.

What do you want me to say?

You aren't a qualified expert on this topic. If you want an expert opinion, talk to an expert, not some dubious fucking provenance wikipedia page.

They are experts in this field, and, unlike “random person on the internet who spent 2 minutes on google”, have informed opinions on this topic.

If you want a serious discussion on why fluoride is good or bad, that’s where you need to go.

Random person on the internet is very easy to disagree with, because we’re all idiots right? It’s a very easy lazy way of self confirmation.

…but if you are serious about critically considering the issue and facing your own biases, talk to an actual topic expert.

My dentist told me he had carefully reviewed the literature and determined to his satisfaction that public fluoridated water was in the best interests of public health, currently. He offered to share some reading that he was convinced by.

You can’t really ask for more that that.

Discussing this here is a bit like protesting by posting on social media; yes, I suppose it’s better than doing nothing and not engaging with the topic at all… but only barely, and not in any meaningful way.

> If I don't want it, let me turn them off.

Can you it off for anyone sending you messages too?

That's the issue; you not wanting to use it does not mean that spammers won't use it.

That's the problem. You can't have nice things if some people can use it to abuse the system; and there are a lot of people who will.

> But my point is this isn't something unreasonable for a user to want.

This ignores the reality which is that doing it in a way that gives a nice user experience without an enormously painful security issue is really non trivial.

Maybe it's OK to have the choice?

...

If you love your android phone, don't care about iOS, don't like iphones.... why do you care? I mean, why does it upset android users when they see this sort of thing for people using iphones?

It mystifies me. If you love you phone, and you think it's better, then use it.

Nothing lost right?

So, how exactly is my app/whatever supposed to spin up a parallel process in the OS and then talk to it over IPC? How do you shut it down when the 'host' process dies?

Not vaguely. Not hand wave "just launch it". How exactly do you do it?

How do you do it in environments where that capability (spawning arbitrary processes) is limited? eg. mobile.

How do you package it so that you distribute it in parallel? Will it conflict with other applications that do the same thing?

When you look at, for example, a jupyter kernel, it is already a host process launched and managed by jupyter-lab or whatever, which talks via network chatter.

So now each kernel process has to manage another process, which it talks to via IPC?

...

Certainly, there are no obvious performance reasons to avoid IPC, but I think there are use cases where having the compiler embedded makes more sense.

AI is hard; edge cases are hard. AI sucks at edge cases.

Between AI for cars and AI for software the long tail of edge cases that have to be catered for is different, yes.

...but I'm sure the same will apply for AI for art (e.g. hands), and AI for (insert domain here).

Obviously no analogy is perfect, but I think you have to really make an effort to look away from reality not to see the glaringly obvious parallels in cars, art, programming, problem solving, robots, etc. where machine learning models struggle with edge cases.

Does the tooling they used matter? no, not at all.

...but if they've claimed to solve the 'edge case problem', they've done something really interesting. If not, they haven't.

So, don't claim to have done something really interesting if you haven't.

You can say "I've been using AI to build a blah blah blah. It's great!" and that's perfectly ok.

You have to go out of your way to say "I've been using an AI to build blah blah blah and I haven't written any of it, it's all generated by AI". <-- kinda attention seeking.

"no lines of code directly written" really? Why did you mention that? You got the AI to write your software for you? That sounds cool! Let's talk! Are you an AI consultant by any chance? (yes, they are). ...but.

No. You didn't. You really didn't. I'm completely happy to call people out for doing that; its not unfair at all.

Too many AI grifters out there.

Nice! It is entirely reasonable both to do that and to be excited about it.

…buuut, if that’s what you’re doing, you should say so.

Not:

“no lines of code directly written, just directing the AI”

Because those (gluing together AI code by hand and having the agent do everything) are different things, and one of them is much much MUCH harder to get right than the other one.

That last 10-15%. Self driving cars are the same story right?

/skeptical face.

Without fail, every. single. person. I've met who says that, actually means "except for the code that I write", or "except for how I link the code it build together by hand".

If you are 50kloc in to a large complex project that you have literally written none of, and have, eg. used cursor to generate the code without any assistance... well, you should start a startup.

...because, that's what devin was supposed to be, and it was enormously and famously terrible at it.

So that would be either a) terribly exciting, or b) hyperbole.

Are those example where a chatbot helps fill out the form, or just examples of where forms are hard?

My image search did not find any results of AI chatbots that helped fill out the form for you. Do you have a direct link to a form by any chance?

Can you be more specific?

Like, where specifically would I have to spend money to see this.

> Seems like the users don't mind spending 2-5 minutes per year chatting with the bot.

This seems like an enormous amount of effort to have gone to for a single form that people use once a year.

Did you roll out the chatbot assist to other forms? If not, why not? If so, are any of those forms easier to get access to that we can see either live or in a video?

Honestly, this is why I get frustrated with these conversations.

If it works so well, why isn't this sort of thing rolled out in many, visible, obvious places. Why is it hidden away behind paywalls and internal systems where no one can see it?

Why isn't everyone doing it? I've visited 4 websites today which had a chat bot on them, and none of them had a way for the bot to interact with anything on the page other than their own chat context.

Like I said, I'm sure it works to some degree, and varying degrees depending how much effort you put into it... but I'm frustrated I can never find someone who's so proud of it working they can go HERE, look at THIS example of it working.

Does anyone have an example we can actually look at?

Someone has a 'friend' who has a totally-not-publically-visible form where a chat bot interacts with the form and helps the user fill the form in.

...and users love it.

However, when really pressed, I've yet to encounter someone who can actually tell me specifically

1) What form it is (i.e. can I see it?)

2) How much effort it was to build that feature.

...because, the problem with this story is that what you're describing is a pretty hard problem to solve:

- An agent interacts with a user.

- The agent has free reign to fill out the form fields.

- Guided by the user, the agent helps will out form fields in a way which is both faster and more accurate than users typing into the field themselves.

- At any time the user can opt to stop interacting with the the agent and fill in the fields and the agent must understand what's happened independently of the chat context. i.e. The form state has to be part of the chat bot's context.

- At the end, the details filled in by the agent are distinguished from user inputs for user review.

It's not a trivial problem. It sounds like a trivial problem; the agent asks 'what sort of user are you?' and parses the answer into one of three enum values; Client, Foo, Bar -> and sets the field 'user type' to the value via a custom hook.

However, when you try to actually build such a system (as I have), then there are a lot of complicated edge cases, and users HATE it when the bot does the wrong thing, especially when they're primed to click 'that looks good to me' without actually reading what the agent did.

So.

Can you share an example?

What does 'and has a lot of usage' mean in this context? Has it increased the number of people filling in the form, or completing it correctly (or both?) ?

I'd love to see one that users like, because, oh boy, did they HATE the one we built.

At the end of the day, smart validation hints on form input fields are a lot of easier to implement, and are well understood by users of all types in my experience; it's just generally a better, normal way of improving form conversion rates which is well documented, understood and measurable using analytics.

...unless you specifically need to add "uses AI" to your slide deck for your next round of funding.

If you’re a for profit company trying to raise funding and fend off skepticism that your models really aren’t that much better than any one else’s, then…

It would be dishonest, but as long as no one found out until after you closed your funding round, there’s plenty of reason you might do this.

It comes down to caring about benchmarks and integrity or caring about piles of money.

Judge for yourself which one they chose.

Perhaps they didn’t train on it.

Who knows?

It’s fair to be skeptical though, under the circumstances.

MCTS will be the next big “thing”; not agents.