richardw

Almost perfect. Along with mail that follows you, I’ve always wanted a system that lets you grant the right to send you mail, that you can hard revoke. Physical and digital. We could call it “slightly more complex mail protocol”. No more layers of filters, just grant and revoke.

I’m assuming exploratory work from the exec, not something they make decisions with or put into production. If you need something you can trust, you typically need a lot of checks, including multiple humans.

I play a weird part at work near AI. I use it all the time but I’m the first person to warn everyone that it’s absolutely not trustworthy. No matter your prompt, the data, the guidelines built into it, the output is fundamentally flaky. But I use it while knowing that and working around it. Making the process reliable is a big part of my focus, and usually that means minimising the part the LLM plays. Checks and balances live where things are predictable.

You do what you want to do. You might find a niche you can double down on. $5 is a terrible price but a great way to see if what you build solves anything. In that way it’s a lot better than $0. Then you double down on the winners.

If it doesn’t work, you’ve tried and now you know, and you probably built a bunch of skills you can use.

I had quite a few people who said “I’ve never bought anything ever but I love your app so much I’ve paid for it.” Incredibly gratifying.

Although check out Patio11’s advice on charging more. Customers who spend more give you less shit.

Look, 6 months is better than 18 but it drives the type of startup you’re likely to get. Anyone trying to move fast just can’t participate. They’re forced to go elsewhere. You can only deal with organisations that can survive and plan for 6 months. What do they do in the meantime? Build? Wait? Especially in AI, the world is moving so fast that you need to make decisions within days, not half a year. How many frontier models and innovations arrive in half a year, while this lot think about the now-outdated proposal?

All you’ll get are sluggish companies designed to suck up the funds. I used to work for such a company. The business model seemed to be whatever would fund hours of researchers time, and travel money for those smart enough to write papers for whatever conference was in a country they hadn’t visited yet. Targeting EU funds was top priority and they could do it by creating some arbitrary “consortium” that generated enough credibility with adjudicators. I hated the waste and the lack of actual goals.

Don’t need air supremacy if the boots are attached to ground drones. We’re not far off from a starship equivalent deploying a cloud of ground and air drones, some of which could help effect the air supremacy required. Like rapid dragon but more variations of deployed materiel and…a lot more rapid.

Only if you want a production-ready output. To get execs able to self-feed enough, this works fine. Look, you don’t see value until it’s perfect. Good, other people do. I see your fallacy and raise you a false dichotomy.

The processes the people want the sql for are likely filled with algo’s. An exec wants info in a known domain, set up a text to sql system with lots of context and testing to generate queries. If they think they have something good, get an expert to test and productionise it.

“Thank you for your request. Can you walk me through the steps you’d use to do this manually? What things would you watch out for? What kind of number ranges are reasonable? I can propose an algorithm and you tell me if that’s correct. The admins have set up guidelines on how to reason about customer and purchase data. Is the following consistent with your expectations?”

Your kid with a set of AI’s is going to blow the greybeards out of the water in a few years. They learn and iterate a lot faster. They just accept the latest tech as a given.

- greybeard who is trying his hardest to keep up

Any algo that a human would follow can be built and tested. If you have 10 analysts you have 10 different skill levels, with differing understanding of the database and business context. So automation gives you a platform to achieve a floor of skill and knowledge. The humans can now be “at least this good or better”. A new analyst instantly gets better, faster.

I assume a useful goal would be to guide development of the system in coordination with experts, test it, have the AI explain all trade offs, potential bugs, sense check it against expected results etc.

Taste is hard to automate. Real insight is hard to automate. But a domain expert who isn’t an “analyst” can go extremely far with well designed automation and a sense of what rational results should look like. Obviously the state of the art isn’t perfect but you asked about goals, so those would be my goals.

We only used MQ. Against my recommendation, but it was fine if imperfect, and a known entity to all concerned. Did not use that file system.

Yup. And OpenShift. And Red Hat for Linux. And SAP. And IBM. But you know what? Tiny group of people relative to the impact, revenue and competitors. If we needed skills we clicked “buy”, 100 consultants would arrive who are experts, sort it and we’d move on. Not scratching around looking for people who know what we use and needing to learn 50 different open source tools. Coming from a much looser universe I learned to appreciate the principles for that context.

Not commoditising for enterprise. My last gig wouldn’t allow open source software or any company that might not be there in a decade, or which kept data anywhere but our own tenant. We’d look for the “call us” pricing rather than hate it, which I normally do. We added databricks and it was considered one of my top three achievements, because they don’t have to think about data platforms again, just focus on using it. It’s SO expensive for an enterprise to rejig for a new platform that you can’t rely on (insert open source project here).

I managed to add one startup and so far it’s done very well, but it was an exceptional case and the global CEO wanted the functionality. But it used MongoDB and ops team didn’t have any skills, so rather than learn one tiny thing for an irrelevant data store they added cash to use Atlas with all the support and RBAC etc etc. They couldn’t use the default Azure firewall because they only know one firewall, so added one of those too. Also loaded with contracts. Kept hiring load down, one number to call, job done. Startups cost is $5-10k per year. Support BS about $40k. (I forget the exact numbers but it dwarfed the startup costs.)

Startups are from Venus, enterprise are from Jupiter.

Whoever is paying for your time should calculate how much time you’d save between the different products. The actual product price comparison isn’t as important as the impact on output quality and time taken. Could be $1000 a month and still pay for itself in a day, if it generated >$1000 extra value.

This might mean the $10/month is the best. Depends entirely on how it works for you.

(Caps obviously impact the total benefit so I agree there.)

Do it on April 1. With a “we told you so” tagline. Have a few 2025/6 templates, like writing a presidential executive order because there are so many of them.

Capture enterprise AI enthusiasm by providing a 1-stop shop for data and AI, optionally hosted on your own cloud tenant. Keep deploying functionality so clients never need another supplier. Partner with SAP, OpenAI, anyone who holds market share. Buy anyone that either helps growth or might help a competitor grow.

Enterprise view: delegate AI environment to Databricks unless you’re a real player. Market is too chaotic, so rely on them to keep your innovation pipeline fed. Focus on building your own core data and AI within their environment. Nobody got fired for choosing Databricks.

Ok cool so what should he do today? Close up shop? Resign? Or try?

Or, alternatively, it’s much harder to fight with one hand behind your back. They need to be able to compete for resources and talent given the market structure, or they fail on the mission.

This is already impossibly hard. Approximately zero people commenting would be able to win this battle in Sam’s shoes. What would they need to do to begin to have a chance? Rather than make all the obvious comments “bad evil man wants to get rich”, think what it would take to achieve the mission. What would you need to do in his shoes, aside from just give up and close up shop? Probably this, at the very least.

Edit: I don’t know the guy and many near YC do. So I accept there may be a lens I don’t have. But I’d rather discuss the problem, not the person.

I thought I might get pushback on the example. I honestly just googled and found a random person being Harris, but just add anyone who accepts spacetime/relativity and how it represents not what we experience. Or anyone who accepts that the mind predicts rather than observes reality, because our responses need to be faster than what we can process what we perceive. So even our direct perceptions are not quite exact.

The point, please focus on this point, is that it’s not a done deal that all accept free will. There exists a set of people who are smart and believe the opposite. I’m arguing against “everyone should think what I think”.

Edit: although to be honest if you convinced me otherwise I’d be 100% happy! “Most philosophers” (or my own “lots of smart people”) isn’t an argument I am moved by but I’ll go read up. My personal view is that the disappearance of time at the speed of light is incredibly powerful. I don’t feel we’ve explored the implications of that nearly enough. It either means the future impacts the past or super determinism is correct. Is there an alternative?

We “experience” consciousness in the same way we experience hard objects that are actually made of mostly space. It feels right but it’s impossible to say that it’s the truth.

There are a lot of smart people who are less sure that our perceptions reflect reality. Free will, time, consciousness, space all feel very real. But when you interrogate them enough, those fall away. Go fast enough and your time slows down. Go as fast as a photon and time and distance disappear entirely. Yet we’re sure they’re real. At one point we absolutely knew the sun revolved around us. We could see it doing that daily.

One smart person on free will below. There are so many others. There’s certainly no consensus that “our experience = how things are”.

https://www.samharris.org/blog/the-illusion-of-free-will

The point is that our perception doesn’t provably equal “truth”. To make absolute statements you need to be a lot more sure and be more correct than all the smart people who say “not sure yet”.

How do you know you’re not an automaton that thinks it’s conscious?

Not feel. Not think. Not have a strong vibe. Proof.

I don’t have to prove anything here, because I’m not the one making an assertion. My thought experiment demonstrates a situation where an entity thinks it’s conscious but isn’t. Humans think we’re conscious but can’t prove it.

People run on autopilot unless we put in effort to focus on the current moment. Almost everyone demonstrates bad habits that show how little conscious awareness and control we have. LLM’s can replicate a lot of our activities, so the realm of “consciousness” is in the things that the AI can’t do by simply repackaging and regurgitating content back at us. What we can claim as conscious special sauce has grown smaller over the last 15 years.

Argument 2:

Could we design a machine that is absolutely sure it’s conscious, but is not? If so, what differentiates us from it? Open question, I’m not sure. But I’d be surprised if we can be permanently sure.

My personal mantra for these things is that life is simple [1] when you have only one principle. It gets much harder when you have to balance two or more principles.

It’s still better to have the principles than not have them, just keep balance in mind.

[1] for you, that is. People who don’t share your principles might not like your choices and you might be making their life difficult with your lack of balance. See religion, politics, etc.

Yup, it feel similar to mining. The models are a transformation of the underlying content. The content is owned by “us”. A mining company can come in and transform rock into iron but it owes something to the grantors of the mining rights.

There is certainly a contribution in improving how the body of work is represented but we treat the “AI” as the smarts, when really it’s a lens on the collective knowledge we have built. You can make the lens better, bit not claim ownership of the body of work. Right now that’s what’s happening, with a few edge cases for artists and publishers.

My opinion isn’t fully formed but I currently think either all content producers have a claim (potentially workable as eg a discount), or only those who contribute should get access to AI’s.

And by all I mean the AI companies owe a huge debt to all humans who wrote or designed or drew anything. The vast majority of the benefit of this technology relies on volume: the billions of pages and lines of code we wrote for other humans, but have now been repurposed. This technology relies on bulk, which was mainly unprofessional or freely given content, by those who intended it for other humans. It was not 100% built only on the output of the few who charge for their exquisite words or designs, even if their output is higher quality.

Alternatively, let the AI companies go for it but everyone who uses any kind of AI should understand that they’re standing on the shoulders of the millions of developers and nonprofessional writers whose work has now been repurposed. Not the few artists and journalists. So those artists and journalists should both refuse to contribute to, and use, AI.

* I’ve written very little of this useful content, but would be happy to pay my share to those that have built what we have. I also turn off training on my content, but I pay a lot for models. Feel free to help me think through this with comments of your own.

Yup 30+ year dev here. Tried Cursor for a bit and honestly most of the tokens burned were fighting with stupid code generation, and having it confidently push on. Just cancelled my sub today. Been working with Windsurf the last few weeks and it feels a little more controllable. However, what I tend to do is work on conceptual design, brainstorming etc with ChatGPT (o1/o3-mini-high) or Claude, then when I converge I bring the task to Windsurf for its slightly better view of the code and ability to execute across many files.

Our brains killed us if we figured things out wrong. We’d get eaten. We learned to get things right enough, and how to be pretty right, fast, even when we didn’t know the new context (plants, animals, snow, lava).

LLM’s are just so happy to generate enough tokens that look right ish. They need so many examples driven into them during training.

The map is not the territory, and we’re training them on the map of our codified outputs. They don’t actually have to survive. They’re pretty amazing but of course they’re absolutely not doing what we do, because success for us and them look so different. We need to survive.

(Please can we not have one that really wants to survive.)

You keep the receipts for about 4 years and you speak up one minute after the government changes. You get it done long before the following election.

One thing I think I’ve found is: reasoning models get more confident and that makes it harder to dislodge a wrong idea.

It feels like I only have 5% of the control, and then it goes into a self-chat where it thinks it’s right and builds on it’s misunderstanding. So 95% of the outcome is driven by rambling, not my input.

Windsurf seems to do a good job of regularly injecting guidance so it sticks to what I’ve said. But I’ve had some extremely annoying interactions with confident-but-wrong “reasoning” models.

They don’t have to own stocks. If they like eating and buying household goods then this will affect them negatively. It will help very few people for a long time, even if it goes perfectly to plan.

I’d be extremely surprised if other countries meekly do what Trump wants. There are many options on the table, and change is more likely in a crisis.

Now the task of an adversary is to simply enable the backdoor rather than create it from scratch. The people using Signal for this are doing it on their own devices, so now you have multiple problems.

Eg how to get non technical people to know when they’re using the civilian version.

Alternative crazy universe: Just use the tech that was created for the government and does all the right things.