jcgrillo

> if your boss asks for honest feedback giving it to them with both barrels is a career limiting move

Or, from another point of view, expanding your career's event horizon. Imagine how much longer you might have languished in that stifling, stagnant job if you hadn't unloaded on the boss?

> Footage from inside your home is, of course, what you’re paying for the cleaning service with. On its website, Shift says customers’ “privacy is fully protected,” with sensitive details like names, faces, or personal information from screens and ID cards blurred and anonymized before being used for AI training.

OK, but do they store the footage in such a way that it's not tied to my physical address? This dataset is useful in one particular way--to identify valuable targets to rob. When they get hacked, will the attackers be able to exfiltrate these data in an actionable way? I don't get why folks don't ask the obvious questions. The company's answer to this question (probably involving lots of squirming and weasel words) would have made the story interesting.

EDIT: <facepalm>these are probably the people who have an amazon alexa, a google nest, a ring door lock, an app to remote start their car, and another app to control their oven</facepalm>

About as strange as letting someone else work on your car. Some people can do it without any discomfort. Could not be me.

Does the un-regulation cut both ways? A landlord usually needs to notify tenants 24hr in advance if they're going to enter the property. Does an AirBnB host need to follow any similar rules? It's not like the renters have a lease, it's not their residence.. Do they have any rights to privacy or notice at all?

Well, before we get too far ahead of our skis, let's wait until someone makes a good film with AI. This company didn't, nor will they. It's not their business. Whether anyone ever will is an open question.

False equivalence. A large studio production still retains the voice and editorial oversight of its director, which is why you can differentiate between films by their directors. There are undoubtedly many filmmakers who collaborate on such productions. Whether or not you believe such projects produce good work, you cannot deny the people involved are artists.

This company is not that. They're a venture backed silicon valley startup. Not a film studio. Not a production house. Their business is not about films, it's about raising venture capital, and (maybe) providing an exit to their investors. Whether they do it by producing a blockbuster movie or some other thing is immaterial, and they'll pivot towards whatever is the most expedient path. They're an "AI tools" company.

> Higgsfield, a San Francisco startup valued at $1.3 billion..

I don't think these are filmmakers. They aren't here for the art, they're here to find product market fit and, increase revenue, and raise another round at higher valuation.

The mantra of Silicon Valley tech: Always Be Defrauding

No worries! The recovery was much slower and more gradual than the onset, it probably took about a week. This was compounded by the congestion and general cold symptoms. I suspect my sense of smell is still impacted ~3yr later, but I have no way to be sure.

That's a good point, this does seem qualitatively different. Like dialect translation. In that case the specification is really precise, it's just the old code. Building something new or adding functionality to existing software the spec is almost guaranteed to be more vague.

EDIT: this specificity seems important for language models but the harder I think about it the less sure I am that it's the right intuition..

they spent a lot of time on it!

Right, but most of them don't already work as software engineers, at least it hasn't been my experience of my colleagues. However, companies that are aggressively adopting AI coding tools aren't (at least nobody has shown it) getting better by any metric. So, what gives? Why, generally, isn't this kind of success story common?

Which is greater? The cost of refactoring or the cost of not refactoring? If the consequence of not refactoring is having to deploy twice as many services and libraries every time you add a feature... that sounds pretty risky to me over time.

I'm curious whether you have any insights into why high quality LLM-assisted (or enabled?) projects seem so relatively rare? Instead it seems like the preponderant contribution is a deluge of low quality slop.

You didn't share yours and Adam's prompts in the post, so I'm left wondering how much of the success of this project is attributable to your collective ability and experience (both with this particular project and software in general) vs the capability of the model and harness itself? On that note, do you anticipate releasing LLMs at Oxide[1] (linked from RFD 0576)?

Personally I find credible success stories like yours interesting, if a little jarring. If they were commonplace, shouldn't software be generally getting a lot better?

[1] https://github.com/oxidecomputer/meta/tree/master/engineerin...

You might be surprised how low the clearance is on some over-the-road trailers. Especially the long ones for hauling heavy equipment, their breakover angles are atrocious. Most roads are navigable regardless, and they know how to pick routes that avoid impassible ones. Low as it is, this truck's clearance probably isn't as limiting as you might think.

At least they didn't paint it red. And it's a good thing Enzo isn't here to see it.

It happens in poorly factored codebases. If you find it happening that's a sign you need to refactor. If you find it happening repeatedly in the same codebase that means you failed to refactor properly the first time.

Sprintz looks interesting, I hadn't heard of it thanks for the link.

Could adtech really be used for something better? The capability is to precisely model users' emotional reward centers and then bombard them with content that targets those reward centers. Is there any way you can imagine this being used for good? Even if you do it in the least bad, privacy preserving way possible you're still messing with people's minds on a massive scale. Sure, that's a capability alright! I can't imagine a decent use of that capability, though. In my view it's one that should be prevented via regulation and treaties, much like weapons of mass destruction.

Unlike weapons of mass destruction, though, there's nothing to be gained by working on adtech. Let me unpack that a bit--a "good guy" can justify working on nuclear weapons because if he doesn't his country will be unable to defend itself when the treaties break down. There's no similar existential threat with adtech. If I choose to not work on that poison, I'm not leaving any advantage on the table. My agency in this situation is worthless. At least that conclusion is why I don't do adtech anymore.

I broadly agree with your bigger point though. To the extent that AI coding tools are useful it's important to explore their use, and evaluate whether they're any good. But if I'm in a company that's forcing their use, or putting up token use leaderboards, or any of the other horror stories we've been hearing about, you can be damn sure I'd get myself fired or quit. At that point the inmates are running the asylum, and there's no reason to stay.

> it's starting to terrify me how it seems that the worst people in the world are more than willing to wield this power, while good people opt out

Maybe it's just that the capability is bad. Adtech, for example, isn't something that anyone uses for good. They blow a lot of smoke about it--looking at you, Apple--but despite the "good-washing" it's all just the same extractive, invasive, dehumanizing business. Bad people will naturally concentrate around this capability. I know because I've worked with a few of them.

AI coding tools seem like they're engineered to undermine cautious, rigorous, and pragmatic engineering discipline. Of course the bosses want that, they see a short term path to massive output increases and nothing sounds better. They'll be cashed out by the time the mess needs to be cleaned up, that's someone else's problem. People who are predisposed to this kind of antisocial behavior are the ones who concentrate around AI tools. Rigorous, careful engineers who care about building maintainable systems that will outlast their tenure find less value in them.

I think it's more nuanced than

> Technology is how we expand human capability.

I think as a general statement about technology as a whole it's true. But do all technologies expand human capabilities? I don't think so.

> There is a relative lack of actual documented architectures that work. Not only do you need the details but also the usage of these systems so as to judge what "good" is.

Mostly these things are the secret sauce (or at least primary ingredients) underlying all the successful products you've heard of. Over time the secrets come out in the form of papers, blog posts, and open source software. But often the cutting edge isn't public because it's in this or that company's private, proprietary codebase. As people move between companies the knowledge diffuses, but if you're relying on a model that was trained on last year's public code you're at least a few more years than that behind. And it's even worse than that, because correct patterns--ones that actually work well--are underrepresented in the dataset.

Garbage in, garbage out. I don't understand what people are hoping for with this whole "agentic" thing... Autocompleting the function I'm currently working on is potentially useful, provided it produces acceptable code more than.. idk.. 95% of the time. "Agentically" building larger system components? Nah.

> Smaller companies will have departments that distill larger models into something more specifically manageable and useful for them.

In order to do that they'd have to make a concrete business case to justify the headcount and compute costs. They'd be facing the same fundamental economic problems Anthropic, OpenAI, MSFT, etc are facing just at a department level instead of a megacorp level. I hope they try it, sunlight is the best disinfectant.

However, when the pressure is turned up and people have to actually show results--and, like, be accountable--instead of just buying a subscription and externalizing the accountability, I don't think we'll see so much enthusiasm about AI coding. Whether or not an engineer is actually more or less productive with AI (not merely whether they feel more productive) will begin to matter a lot more. I don't see how people continue using AI in this hypothetical small company under those adverse conditions.

Tech products improving, for one. If everyone is a 10x engineer now (and has been for years) why isn't everything better? Instead, we're seeing github outages increasing in frequency, consumer products degrading left and right (google, iOS, etc etc). What is taking so long? Where are the results?

Where is the impact on the real world? Where are the process improvements? Where are the efficiency gains? Why are prices going up, not down? If y'all are actually super-engineers now why aren't you actually making anything better? Hurry up and fix some shit already.

I'll tell you why: it's all bullshit. You aren't actually any more effective with AI than you were without it. You may feel that way, but it's delusional. I'd love to be proven wrong, you can do it by living up to the hype.

I'm confident that won't happen, though. The industry will continue to produce the same mediocre, enshittified slop it always has. It'll do it at approximately the same rate, achieving the same output per engineer, even if they make you do it via a chatbot middleman. AI has had this property for a long, long time before LLMs, before Deep Learning, while neural nets were in their infancy--before backprop even. It can do really impressive things, in the right context, but it is not a substitute for thought and work. You have to put in the effort--whether it's tuning an objective function or iteratively crafting and refining a prompt. All that was old is new again.

The difference now is that it's taken on some bizarre religious significance. These goddamn fucking acolytes running around praising the computer jesus because it speaks words and stuff!. Revolting.

If you really want to convince me, show me something better than that.

100%. Just because you can plagiarize basic correspondence doesn't mean it's OK.

The trouble with this narrative is the evidence doesn't actually support it. If it is so powerful where are the results? If it was actually all that powerful it would be obvious. There wouldn't be any question. Nobody says "nuclear explosions aren't really all that much bigger than conventional explosions" because it's measurable and absolutely obvious just looking at one that they are. I don't believe you when you say "AI is powerful". Show me a convincing demonstration--like a Trinity type test. Show me a company that is wildly exceeding what all the other companies can, or what any company could before AI. It doesn't have to be a hydrogen bomb, a low yield conventional test would suffice. But until you have that demonstration ready, all the talk is just noise.

Why not? Blacksmithing and coding have a hell of a lot in common. In both disciplines toolmaking is extremely important. Often you have to make custom tools to accomplish a design--e.g. a twisting wrench or a form tool. Sometimes you have to make tools that get used once and thrown away, like a jig temporarily welded to a piece to hold it in place while you build its sibling assembly. Sound familiar? I do this kind of thing all the time in code.

Another similarity is the relative simplicity of the underlying structure of the system. You essentially have two hammers (one small one you swing with your hand and another big one that is planted on the ground), some material, and some heat. You build the rest.

Another similarity is the resistance to automation. A skilled blacksmith is a versatile worker. You can create assembly lines to automate any one thing they might produce. The end product will not have the same quality--it will not truly be wrought iron, each piece will not be unique, there will be nothing of the aesthetic taste of the artist in it, but if you're just some bean counter who doesn't care about those things you'll be able to sell it. But if you need the optionality to produce any of those things.. automation is not your friend. And some things just cannot be automated, at least not without extreme costs or very poor results--shoeing horses comes to mind.

> Cope bias is powerful.

Have you stopped to consider whether this statement might be more applicable to yourself? "Myopic lies" is at the very least highly exaggerated phrasing, if not itself myopic and a false characterization. If it's not too uncomfortable for you, some honest introspection might be worthwhile.

Language is often ambiguous! You have to guess the intended meaning based on context clues. Unambiguously phrased language sounds less natural, because it is. Incidentally, this is part of what makes natural language such an awful fit for controlling a computer.

The problem is natural language is a horrendously bad human-computer interface. Even if they're running nondeterministic software, computers are very precise machines. You wouldn't talk to your lathe or milling machine and expect good things to happen. So why would you have that expectation of a computer? It's ridiculous sci-fi fantasy nonsense.

It's hilarious, when you boil away all the froth and hype, that we've collectively decided that "talk to computer" is somehow worth an entire generation of venture capital and maybe even the whole stock market. It's a dumb idea to begin with. A mouse and keyboard are better.

I lost my sense of taste and smell for a few days when I had covid, I was eating a bowl of spicy ramen at the time and I completely lost the ability to taste anything over the course of about 2 minutes. No spice, no salt, nothing. I usually have a pretty high spice tolerance, but to be certain I ate an entire birds eye chili. This would normally leave me in pretty excruciating pain for a few minutes, it did nothing. Strange that others report retaining their spice taste to some degree, that wasn't my experience at all. I wonder what was different?