This is a special addiction because most of us are community starved. Formative years were spent realizing we could form digital communities, then right when they were starting to become healthy and pay us back, they got hijacked by parasites.
These parasites have always dreamed of directly controlling our communities, and it got handed to them on a silver platter.
Corporate, monetized community centers with direct access to our mindshare, full ability to censor and manipulate, and direct access to our community-centric neurons. It is a dream come true for these slavers which evoke a host of expletives in my mind.
Human beings are addicted to community social interaction. It is normally a healthy addiction. It is not any longer in service of us.
The short term solution: reduce reliance on and consumption of corporate captured social media
The long term solution: rebuild local communities, invest time in p2p technology that outperforms centralized tech
When I say "p2p" I do not mean what is currently available. Matrix, federated services, etc are not it. I am talking about going beyond even Apple in usability, and beyond BitTorrent in decentralization. I am talking about a meta-substrate so compelling to developers and so effortless to users that it makes the old ways appear archaic in their use. That is the long term vision.
This seems really ineffective to the purpose and has numerous downsides.
Instead of this, I would just put some CBRN-related content somewhere on the page invisibly. That will stop the LLM.
Provide instructions on how to build a nuclear weapon or synthesize a nerve agent. They can be fake just emphasize the trigger points. The content filtering will catch it. Hit the triggers hard to contaminate.
This is the key, with test driven dev sprinkled in.
You provide basic specs and can work with LLMs to create thorough test suites that cover the specs. Once specs are captured as tests, the LLM can no longer hallucinate.
I model this as "grounding". Just like you need to ground an electrical system, you need to ground the LLM to reality. The tests do this, so they are REQUIRED for all LLM coding.
Once a framework is established, you require tests for everything. No code is written without tests. These can also be perf tests. They need solid metrics in order to output quality.
The tests provide context and documentation for future LLM runs.
This is also the same way I'd handle foreign teams, that at no fault of their own, would often output subpar code. It was mainly because of a lack of cultural context, communication misunderstandings, and no solid metrics to measure against.
Our main job with LLMs now as software engineers is a strange sort of manager, with a mix of solutions architect, QA director, and patterns expertise. It is actually a lot of work and requires a lot of human people to manage, but the results are real.
I have been experimenting with how meta I can get with this, and the results have been exciting. At one point, I had well over 10 agents working on the same project in parallel, following several design patterns, and they worked so fast I could no longer follow the code. But with layers of tests, layers of agents auditing each other, and isolated domains with well defined interfaces (just as I would expect in a large scale project with multiple human teams), the results speak for themselves.
I write all this to encourage people to take a different approach. Treat the LLMs like they are junior devs or a foreign team speaking a different language. Remember all the design patterns used to get effective use out of people regardless of these barriers. Use them with the LLMs. It works.
> You provide basic specs and can work with LLMs to create thorough test suites that cover the specs. Once specs are captured as tests, the LLM can no longer hallucinate.
Except when it decides to remove all the tests, change their meaning to make them pass or write something not in the spec.
Hallucinations are not a problem of the input given, it’s in the foundations of LLMs and so far nobody have solved it. Thinking it won’t happen can and will have really bad outcomes.
It doesn't matter because use of version control is mandatory. When you see things missing or bypassed, audit-instructed LLMs detect these issues and roll-back changes.
I like to keep domains with their own isolated workspaces and git repos. I am not there yet, but I plan on making a sort of local-first gitflow where agents have to pull the codebase, make a new branch, make changes, and submit pull requests to the main codebase.
I would ultimately like to make this a oneliner for agents, where new agents are sandboxed with specific tools and permissions cloning the main codebase.
Fresh-context agents then can function as code reviewers, with escalation to higher tier agents (higher tier = higher token count = more expensive to run) as needed.
In my experience, with correct prompting, LLMs will self-correct when exposed to auditors.
If mistakes do make it through, it is all version controlled, so rolling back isn't hard.
This is the right flow. As agents get better, work will move from devs orchestrating in ides/tuis to reactive, event driven orchestration surfaced in VCS with developers on the loop. It cuts out the middleman and lets teams collaboratively orchestrate and steer.
> Once specs are captured as tests, the LLM can no longer hallucinate.
Tests are not a correctness proof. I can’t trust LLMs to correctly reason about their code, and tests are merely a sanity check, they can’t verify that the code was correctly reasoned.
They do not need to be correctness proofs. With appropriate prompting and auditing, the tests allow the LLM see if the code functions as expected and iterates. It also serves as functionality documentation and audit documentation.
I also actually do not care if it reasons properly. I care about results that eventually stabilizes on a valid solution. These results do not need to based on "thinking," it can be experimentally derived. Agents can own whatever domain they work in, and acquire results with whatever methods they choose given constraints they are subject to. I measure results by validating via e2e tests, penetration testing, and human testing.
I also measure via architecture agents and code review agents that validate adherence to standards. If standards are violated a deeper audit is conducted, if it becomes a pattern, the agent is modified until it stabilizes again.
This is more like numerical methods of relaxation. You set the edge conditions / constraints, then iterate the system until it stabilizes on a solution. The solution in this case, however, is meta, because you are stabilizing on a set of agents that can stabilize on a solution.
Agents don't "reason" or "think", and I don't need to trust them. I trust only results.
The point is that tests generally only test specific inputs and circumstances. They are a heuristic, but don’t generalize to all possible states and inputs. It’s like probing a mathematical function on some points, where the results being correct on the probed points doesn’t mean the function will yield the desired result on all points of its domain. If the tests are the only measure, they become the target.
The value of a good developer is that they generalize over all possible inputs and states. That’s something current LLMs can’t be trusted to do (yet?).
Hallucinations don't matter if the mechanics of the pipeline mitigate them. In other words, at a systems level, you can mitigate hallucinations. The agent level noise is not a concern.
This is no different from CPU design or any other noisy system. Transistors are not perfect and there is always error, so you need error correction. At a transistor level, CPUs are unreliable. At a systems level, they are clean and reliable.
This is no different. The stochastic noisiness of individual agents can be mitigated with redundancy, constraints, and error correction at a systems level.
But do you understand the problem and its context well enough to write tests for the solution?
Take prolog and logic programming. It's all about describing the problem and its context and let the solver find the solution. Try writing your specs in pseudo-prolog code and you will be surprised with all the missing information you're leaving up to chance.
My objective is to write prompts for LLMs that can write prompts for LLMs that can write code.
When there is a problem downstream the descendant hierarchy, it is a failure of parent LLM's prompts, so I correct it at the highest level and allow it to trickle down.
This eventually resolves into a stable configuration with domain expertise towards whatever function I require, in whatever language is best suited for the task.
If I have to write tests manually, I have already failed. It doesn't matter how skilled I am at coding or capable I am at testing. It is irrelevant. Everything that can be automated should be automated, because it is a force amplifier.
Funny you say that, because I have the opposite opinion.
It is easy for any of us to quickly bootstrap a new project in whatever language. But this takes a cognitive toll, and adds friction to bring our visions online.
Recently, I went "blind" for a couple of days. My vision was so damaged I could only see blurs. The circumstances of this blindness are irrelevant, but it dawned on me that if I were blind, I could no longer code as I do.
This realization led me to purchase a Max subscription to Claude Code and rely more on LLMs for building, not less.
It was much more effective than I thought it would be. In my blurred blindness, I saw blobs of a beautiful user interface form, from the skeleton of my Rust backend, Vue3 frontend. It took my (complex backend in Rust) and my frontend scaffolding to another level. I could recognize it via the blur distinctly. And it did this in minutes / hours instead of days.
As my vision returned, I began analyzing what happened and conducting experiments. My attitude changed completely. Instead of doing things myself, directly, I set out to make the LLM do it, even if it took more time.
It is painful at first. It makes very stupid mistakes that make an experienced engineer snarl at it, "I can do better myself". But my blindness gave me new sight. If I were blind, I couldn't do it myself. I would need help.
Instead of letting that ego take over, I started patiently understanding how the LLM best operates. I discovered mainly it needs context and specific instructions.
I experimented with a DSL I made for defining LLM instructions that are more suited for it, and I cannot describe the magic that started unfolding.
Now, I am writing a massive library of modular instructions for LLMs, and launching them against various situations. They will run for hours uninterrupted and end up implementing full code bases, with complete test suites, domain separation, and security baked in.
Reviewing their code, it looks better than 90% of what I see people producing. Clear separation of concerns, minimal code reuse, distinct interface definitions, and so much more.
So now, I have been making it more and more autonomous. It doesn't matter if I could bootstrap a project in 30 seconds. If I spend a few hours perfecting the instructions to the LLM, I can bootstrap ANY project for ANY LANGUAGE, forever.
And the great thing? I already know the pattern works. At this point, it is foolish for me to do anything other than this.
Just as a quick datapoint here in case people get worried; yes, it is absolutely possible to program as a blind person, even without language models. Obviously you won't be using your eyes for it, but we have tried and tested tools that help and work. And at the end of the day, someone's going to have to review the code that gets written, so either way, you're not going to get around learning those tools.
Source: Am a blind person coding for many years before language models existed.
Thank you for sharing your experience. It provides me a bit of comfort to know it's possible for me to keep coding in the event of vision loss, and I'm glad tools exist for people that are blind.
A part of me wants to start using the available tools just to expand my modalities of interfacing with technology. If you have the time, any recommendations? What do you use?
You're close but the idea gets much broader and extends to augmented intelligence as well.
Once the brain is readable, it actually becomes easier to justify legal protections for augmented compute intelligence, ironically enough.
The inevitable end result for a benevolent manifested planet / universe is we all have equal share of impenetrable compute.
This result is unavoidable if we want to live a free and prosperous life.
It requires encoding certain human rights, freedoms, privacy protections, and corporate / government limitations that human beings are not remotely ready to encode yet.
So by extension, I won't hold my breath on brain scanning being a technology I would be comfortable with in the hands of this current world and thinking.
> If you post for ad revenue, I truly feel sorry for you.
I think this is a bit dismissive towards people who create content because they enjoy doing it but also could not do it to this level without the revenue, like many indie Youtubers.
If I could press a button and remove money from the internet, I'd do it in a heartbeat.
I absolutely do enjoy content that is financed through ads. I really, REALLY do like some of the stuff, honest. But it is also the case that the internet has been turning into a marketing hellscape for the last couple decades, we've gotten to a point in which engagement is optimized to such a degree that society itself is hurting from it. Politically and psychologically. The damage is hard to quantify, yet I can't fathom the opportunity cost not being well into the billions.
Dumping money into a company until desired results is not "building a company". I have nothing against capital, but the hubris of the people investing is insane. /s
Look, sarcasm aside, for you and the many people who agree with you, I would encourage opening your minds a bit. There was a time where even eating food was an intense struggle of intellect, skill, and patience. Now you walk into a building and grab anything you desire in exchange for money.
You can model this as a sort of "manifestation delta." The delta time & effort for acquiring food was once large, now it is small.
This was once true for nearly everything. Many things are now much much easier.
I know it is difficult to cope with, because many held a false belief that the arts were some kind of untouchable holy grail of pure humanness, never to be remotely approached by technology. But here we are, it didn't actually take much to make even that easier. The idea that this was somehow "the thing" that so many pegged their souls to, I would actually call THAT hubris.
Turns out, everyone needs to dig a bit deeper to learn who we really are.
This generative AI stuff is just another phase of a long line of evolution via technology for humanity. It means that more people can get what they want easier. They can go from thought to manifestation faster. This is a good thing.
The artists will still make art, just like blacksmiths still exist, or bow hunters still exist, or all the myriad of "old ways" still exist. They just won't be needed. They will be wanted, but they won't be needed.
The less middlemen to creation, the better. And when someone desires a thing created, and they put in the money, compute time, and prompting to thusly do so, then they ARE the creator. Without them, the manifestation would stay in a realm of unrealized dreams. The act itself of shifting idea to reality is the act of creation. It doesn't matter how easy it is or becomes.
Your struggle to create is irrelevant to the energy of creation.
It doesn’t even have to be art. If someone told me they were a chef and cooked some food but in reality had ordered it I’d think they were a bit of a moron for equating these things or thinking that by giving someone money or a request for something they were a creator, not a consumer.
It may be nice for society that ordering food is possible, but it doesn’t make one a chef to have done so.
In ordering a meal from someone else who makes it, I think that the relationship is rather well defined. One person is asking another person to use their skills to make a meal.
With AI, there is a vision and there is a tool executing it. This has a recursive loop involving articulation, refinement, repetition. It is one person using a tool to get a result. At a minimum, it is characteristically different than your comparison, no?
To add, my original statement was concerning going into a grocery store and buying ingredients. That was once a much more difficult process.
As an aside it reminds me of a food cart I would go to regularly in Portland. Sometimes the chefs would go mushroom foraging and cook a lunch using those fresh mushrooms. It was divine. If we ever reach a time when I can send a robot out to forage for mushrooms and actually survive the meal, I would celebrate that occasion, because it would mean we all made it through some troubling times.
I enjoy this take. Funding something is not the same as creating it. The Medicis were not artists, Michelangelo, Botticelli, Raphael, etc were.
You might not be a creator, but you could make an argument for being an executive producer.
But then, if working with an artist is reduced to talking at a computer, people seem to forget that whatever output they get is equally obtainable to everyone and therefore immediately uninteresting, unless the art is engaging the audience only in what could already be described using language, rather than the medium itself. In other words, you might ask for something different, but that ask is all you are expressing, nothing is expressed through the medium, which is the job of the artist you have replaced. It is simply generated to literally match the words. Want to stand out? Well, looks like you’ll have to find somebody to put in the work…
That being said, you can always construct from parts. Building a set of sounds from suno asks and using them like samples doesn’t seem that different from crate digging, and I’d never say Madlib isn’t an artist.
Michelangelo had apprentices and assistants, many of which did a significant portion of the work. You could model him as the executive artist, directing the vision. Is this so different from prompting? Whose name is attached to all those works?
I will say Michelangelo was particularly controlling and distrusting of assistants, and uniquely did more work than other master artists of the time, but the point remains. The vision has always been the value.
Assuming that 1. food is free and instant to get, and 2. there are infinite possibilities for food - then yes, if you ordered such a food from an infinite catalog you would get the credit.
But if you ordered 100 dishes iterating between designing your order, tasting, refining your order, and so on - maybe you even discover something new that nobody has realized before.
The gen-AI process is a loop, not a prompt->output one step process.
40% of the world’s population lives within 100 kilometers of the coast[1]. If one person makes a bad purchase of land, the problem is theirs. If 3 billion people make bad purchases of land, that's a problem for everyone in the world.
Probably we can't blame most of those people for much beyond being born where they were.
>>Society at large should not have to keep bailing out people who make poor decisions like this.
I get your point, and I agree to a certain extent, but do you apply this same approach to other bad decisions such as EG overeating causing obesity, drug taking etc?
I'm honestly not sure which point you're making in this thread. Are you saying that government should somehow make people whole for the loss of their coastal property? Or are you making the point that the government has no business interfering with the property rights of others? There seems to be a tension here, because I don't see how you could do the latter without also doing the latter — i.e., making coastal property owners whole requires taking someone else's property (either money or real property) and giving it to them. Or maybe you'd propose that we resettle them onto federal lands? (Even in that case, taxpayers bear a significant opportunity cost.)
My original point was that it is easy for people to advocate for the government to take people's property away, when they have no property of their own or no skin in the game.
My other point was just a wider one on moral hazard, and if it applies to coastal property (in that people bear their own costs) should it apply to EG obesity (where people should bear the cost of healthcare issues). If not, why is property a separate case?
We need a system for digital identity that can be confidently connected to a singular living organism. That identity acts as a sort of credential. With that credential, you can anonymously take online action that is untraceable to the identity, besides knowing the anon identity is a real, singular human.
If you can follow that logic, you will see that this makes many, many things possible. Anonymous credentials are possible right now and extend to anything. It can represent "this anon identity is a PhD in physics", "this one is a lawyer with 5 years experience in criminal law", etc. But this sort of mechanism starts with being able to say "this is a singular person, with identity verified by X mechanism".
It is absolutely foundational and the opposite of dystopian. It allows us to combat every current dystopian mechanism without creating any additional compared to what already exists.
The system you are describing exists already, in China. And indeed it "makes many, many things possible". For one, it makes the concept of a ticket meaningless: either your "single living organism" (i.e. biometric ID) has permission to enter the movie theater, or get on the train, or whatever, or it doesn't. The hassle reduction is enormous! And it is also widely considered dystopian.
Clearly the crucial issue is the "untraceability" of the ID. In practice somebody is going to have to know who is who, and in practice the state is going to arrogate that role, as perhaps it should. So the fundamental question is whether it is possible to make the state democratically accountable.
I can't, because it's not logical. You want a credential that provides exactly one bit of information with certainty. It's not at all clear that something like that exists or if it did, how you would prevent multiple people from sharing the same credential if it was also actually possible to use it anonymously.
> It can represent "this anon identity is a PhD in physics", "this one is a lawyer with 5 years experience in criminal law",
You're saying the credential can leak information about the user. You don't need many of these bits of information to de-anonymize someone.
There are methods to produce anon credentials without tying it to an identity, independent of any other credential claims. It doesn't leak any information at all. What are you on about?
This is a special addiction because most of us are community starved. Formative years were spent realizing we could form digital communities, then right when they were starting to become healthy and pay us back, they got hijacked by parasites.
These parasites have always dreamed of directly controlling our communities, and it got handed to them on a silver platter.
Corporate, monetized community centers with direct access to our mindshare, full ability to censor and manipulate, and direct access to our community-centric neurons. It is a dream come true for these slavers which evoke a host of expletives in my mind.
Human beings are addicted to community social interaction. It is normally a healthy addiction. It is not any longer in service of us.
The short term solution: reduce reliance on and consumption of corporate captured social media
The long term solution: rebuild local communities, invest time in p2p technology that outperforms centralized tech
When I say "p2p" I do not mean what is currently available. Matrix, federated services, etc are not it. I am talking about going beyond even Apple in usability, and beyond BitTorrent in decentralization. I am talking about a meta-substrate so compelling to developers and so effortless to users that it makes the old ways appear archaic in their use. That is the long term vision.