The key is consciously making that choice. What people don't want is to be forced into a way of living by others. People can accept what they have a hand in discovering and choosing.
Kinelo is building the "org structure" for humans+AI working together, shoulder-to-shoulder, as peers. We help coordinate tasks between each actor in a company, handoff context, and provide a voice and "seat at the table" to AI coworkers.
We're a small team, just hired our first GTM lead and a new product engineer. Getting ready to grow. We'd like to find another strong, experienced software engineer who can span everything from talking to customers to designing robust architectures to inventing new UX patterns.
We offer competitive salary and very meaningful (~first batch of hires) equity, with a founder previously exited to Apple and strong fundraising history (~$10m).
Please email us at jobs@kinelo.com in addition to applying on Greenhouse. For the engineering role, tell us how you think the job of a software engineer is changing.
I compared the writing style of Opus 5 vs Fable 5, and Opus 5 continues many of the "Claude-isms" of its 4.8 predecessor in a way that Fable broke away from.
Opus 5 still uses "carry the argument", "worth stating plainly", ", and the trap", "The X matters more", the use of "move"
If it's "read as an article" or something then yeah it's crap and Fable's current style isn't actually better than older models. For flowery speech old models are perfectly fine.
For quickly parsing the agent output, it's formulaism isn't a bad thing.
Fable's writing does have a property of going over my head, which didn't happen with earlier agents. Asking for clarification doesn't really give good results.
We've gone full circle where I once again use classical search just to look up what the fuck it's yapping about. It's much quicker and more accurate to take a glance at Wikipedia, than to ask the agent.
Could these complex/hard to read Fable outputs be sign of some kind of industrial level of intelligence, which us humans may have a hard to comprehend, while it may be also hard for machine to use simpler texts to properly outline all nuances and complexities of concepts it output?
(1) it’s not that I can’t understand their output, it’s just written in a way that is very homogenous and same-y, with very boring cliches and phrases that don’t quite match their context
(2) a pretty strong sign of intelligence is being able to explain complex things in simple terms
It's very much number 2 in my experience with it. Predominantly whenever a really advanced word or phrase is used it can be substituted for a much simpler one without losing useful context. And it seems to really prefer to use them a lot.
It's more like Claude models entirely suck at extracting key points. No matter how hard I emphasize that it needs to pick the "load-bearing" facts and claims, it cannot stop itself muttering around. It never nails the core logical structure. GPT is better at that.
If you feed Fable or Opus primarily handoff documents from a previous context instead of human written prompts and are working on something sophisticated it rapidly reaches a level where it's hard to actually comprehend for a non expert. I've received incredibly obtuse outputs that contain more mathematical formulas than English words with programming workloads.
I recently wrote a short paper with Fable, and, with some prodding, I was able to get some non-painful prose out of it.
I just found my prompt:
The writing style could really use some work. Avoid Claude-isms like "stated fairly", em dashes, "load-bearing", overly punchy phrasing like "keep the signal, govern the response". This is a technical document, not a marketing campaign.
Like "It's not x, it's y". It actually has 4 or 5 of those counter-factual, linguistic pause, factual patterns it uses.
"The [goal/ambition/etc.] is larger: statement", is another oft repeated phrase.
And then generally, it loves dramatic pauses in statements like "x exists in y; in practice z". It's the weird punctuation it uses. A massive overuse of colons and semi-colons instead of words like and, but, because, althoughy etc. that humans normally use.
Interestingly, Fable caught the gist and I didn't have to enumerate every Claudism. "Punchy" seemed to be the operative word. My new operating hypothesis is that Claude (Fable in particular) defaults to optimize for concision and "turn of phrase". If we can turn that off, the prose is way more natural.
Claude responded "The arguments and structure are unchanged, but sentences now state claims directly instead of building to a turn of phrase."
I'm OK with colons and semicolons as I tend to write that way.
Yes, these models are very good at writing code but they absolutely suck ass at prose. The prose is annoying and repetitive. At least we only have to deal with it in prompting if you're writing code.
Oh, and comments. You have to do a good amount of prompting to not get shitty 10-line-long comments everywhere.
Yes - THIS! I can't even believe how exhausting it is to read. I'm not sure why or what changed in Fable. Did they do this writing-style output to give it more token compression during/for training or to prefer output for less money?
I love it for a few things, but it's gotten really hard to spend any extended amount of time with it because of the lack of mental model I seem to be able to hold while working with complicated problems.
I'm guessing it's just not enough time doing RL on human feedback.
"Early in RL verbose, grammatical" (if you search) :
We need to understand the operator.
The 5D line element is ds² = e^{2A(x)} (ds²_4d + dx²), where A(x) = sin(x) + 4 cos(x), x in [0, 2π].
The internal coordinate is periodic.
The background is a warped product: metric g_{MN} where M,N = 0..4.
The internal direction has metric e^{2A(x)} dx²?
Wait, the ds² is e^{2A} (ds²_4d + dx²).
So the internal metric is e^{2A(x)} dx².
Actually if the total metric is ds² = e^{2A(x)} (ds²_4d + dx²), then yes, internal metric is e^{2A} dx².
vs. Post RL
We need determine eigenvalue problem for spin-2 fluctuations h_{μν}(x,y) with TT in 4d and depend on x.
For metric of form ds² = e^{2A(x)} (g_{μν}(y) + h_{μν}(y,x)) dy^μ dy^ν + e^{2A(x)}?
Wait internal metric is e^{2A} dx²?
Actually ds² = e^{2A} [ds_4² + dx²].
So internal metric is e^{2A} dx²; warp factor same for 4d and internal?
Yes.
We need equation for h_{μν}(y,x) = h_{μν}(y) ψ(x) maybe with normalization.
…
I can understand it with less cognitive load in the post-RL version versus early in RL. This resonated with my experience using Fable, especially digging hard problems; it feels like I'm reading the "early in RL" version of that model explanation.
The exhausting worthlessness of all LLM writing is so palpable that we need a new theory of the value of culture that has no relationship to the content. Back to the old Aura of the Artist arguments from the Industrial Revolution
I've found Opus 4.8 and Fable 5 both difficult to learn from purely because of how annoying their writing style is. I'm finding GPT 5.6 Sol to be much better for this.
One nice thing but ChatGPT is that good image model means it can generate good infographics occasionally to illustrate. These become naturally compact in text.
Easy enough to change. I have a Stylometry Skill fit to my preferred style—a mix of me and Terry Winograd. Give Opus 10 of your best paper thst you wrote and tell it to build a model of your style. hHard to distinguish except I make way more typos.
I know this is not "as designed," but I kind of like it? Because, as long as it stays this way, it's still at least possible to tell if a human wrote something. Like, I know it's not much, but it gives you the ability to classify information as human generated or machine generated. Machine generated information may not be useless but it is different and IMO needs to be treated with a different level of skepticism. Not that you can simply trust human writing but it seems like the type of people who would publish machine writing have a different distribution of motives than those who would publish human writing.
I do think they're gonna figure out how to fix it at some point. :(
I found 4.6 more amenable than 4.8 to style directions, we'll see how 5.0 does. Super-small-sample-size: I think part of its "Claude-ism" style comes from its propensity to try and "proactively" move the conversation/work along. Not sure how this would fare in non-obviously-productive environments, I'd guess "it's still annoying" considering your evidence.
I'm also thinking of another benchmark: (quantified) stylistic range across different prompts. Just putting it out there if anyone wants to do the work for me :D
Write about the importance of style, diction and tone - without writing about the importance of style, diction and tone (is the followup prompt I think is appropriate)
Kinelo | kinelo.com | San Francisco | Onsite (with a bit of flexibility)
Kinelo solves "context myopia": autonomous and semi-autonomous AI agents don't know what they don't know, so they can't search or find the background information required to accomplish tasks accurately and effectively. The result looks like slop but it's not due to model performance, more due to workflow integration. From there, we're rapidly moving into organizing humans together with autonomous "AI coworkers" into workflows that span both. Our goal is to change how organizations are run and structured.
We're hiring for two roles right now:
- Product Engineer / Senior Software Engineer (https://job-boards.greenhouse.io/kinelo/jobs/4205846009) -- do you have the software engineering wisdom not yet encoded in an LLM? and do you have product sense and the willingness and desire to think several steps ahead in the market?
- Founding GTM Lead (https://job-boards.greenhouse.io/kinelo/jobs/4215784009) -- there's no playbook for go-to-market in the category we're creating + the landscape is changing daily. Come architect a strategy and then move mountains to execute on it and help us grow insanely fast.
Competitive salary, meaningful (~first batch of hires) equity, founder previously exited to Apple, strong fundraising history (~$10m).
Please email us at jobs@kinelo.com.
If you want to stand out:
- For the product engineer role, tell us how you think the job of a software engineer is changing.
- For the GTM role, tell us about a time you got massive results with limited resources, all through your ingenuity, creativity, and hard work.
Imagine if you couldn't buy a lathe unless it refused to make a baseball bat (which could be used for hitting people).
Or if you couldn't buy scissors (because they could cut brake lines).
Or if you couldn't buy a car (because it could be used to run someone over).
And if all of those checked with the government before functioning.
It's almost like maybe instead you should just ban the undesirable end action, enforce that law, and create societal conditions that don't nudge or force people into doing undesirable things.
In fact, right up front in TFA they point out that California already bans manufacturing firearms without a license. This 3D printer regulation is entirely duplicative and unnecessary.
You can make all of those things. First you pick up two sticks and rub them together real good (maybe use the inner bark of a hickory tree or similar to make some cordage s.t. you can make a bow drill and rub them even better.. you may find cottonwood to your liking but I did it with green maple when I was 11 so if you can't lmao look inwards). Using the little coal you've created, build a fire. Now pile a bunch of wood on it and starve it for oxygen. Great, you've created charcoal. Now skin a big animal and make bellows from its hide. Now build a big fire and blast it with the bellows to make it rull friggen hot. Put some rocks in it that have iron in them. Collect the iron from the bottom of the fire (exercise left to the reader) into a stone tub. Build a few forms using wood, sand, and wax in the shapes of the various parts a lathe. Melt the iron in the stone tub and pour it into the forms. Scrape the mating surfaces of the lathe parts flat and clean with a hard stone.
Why should it enforce a law from a separate sovereign entity, is the question. And the answer is, it shouldn’t. I don’t think states do, or should, enforce federal anti-discrimination laws. The feds do that. The states pass their own anti-discrimination laws if they want something to enforce themselves.
Historically, the states never would have come together to create the United States if they had to give up all of their own sovereignty. The constitution is full of compromises.
pedantic. You can use lawsuits, within the law, to thwart the law. You can do it knowing you will lose, just to cause delays, like Califorinia and the 9th circuit are ACTIVELY doing to slowdown gun control cases from reaching the supreme court where they 100% will be struck down. Refusing notification is an action.
We used to ban the undesirable action! Then DEFCAD got that ban overturned, convincing the federal government that they have a First Amendment right to publish 3D-printable firearm plans. So now our choices are to allow widespread 3D printed firearms (which I and many others won't accept) or restrict the means by which they can be made. I genuinely do wish the DEFCAD folks had made different choices that would not have led us here.
I want to know how on earth restricting the publication of plans could be consistent with the first amendment. That's like prohibiting the publication of books with content you disagree with.
Certainly not on First Amendment grounds, and in general I expect powerful AI will quite imminently make people more sympathetic to random manufacturing restrictions on potentially dangerous goods. I can imagine 2A arguments against any regulation that's specifically preventing the use of X for gun manufacturing, but my weakly held best guess is that they wouldn't be persuasive here.
> ...our choices are to allow widespread 3D printed firearms...
Which parts of a firearm can be printed in a consumer-grade 3D printer? Be as specific as your knowledge permits.
Of those that cannot, how much money does one have to spend in order to purchase a 3D printer that is capable of printing those parts that cannot be printed by a consumer-grade printer?
Are you aware of "slam fire" firearms? If you were not, you owe it to yourself to learn how to make a functional "slam fire" shotgun. The tutorials are pretty widespread.
I'm aware of "slam fire" firearms and know why and how it's easy to produce them. They're much less concerning to me because their rate of fire is extremely slow.
I don't know the details of what can be printed in a consumer-grade printer, not having performed firearms manufacturing myself, but I've seen things claiming to be pretty complete kits and it seems to me that most components should be possible. Barrels of any reasonable length might be hard, perhaps firing pins too. (And springs, but of course those are trivial to manufacture by hand.) If it's not actually possible to 3D print an effective gun, perhaps someone should make that argument in detail.
You cannot fully 3d print an effective firearm on a consumer 3d printer. When people talk about 3d printing a gun they are almost always talking about 3d printing a single part—the lower receiver. Federal law considers the lower receiver to be a gun, and it is the part with a serial number.
A lower receiver is not complicated. It essentially just a quirk of the law that the ability to 3d print a lower receiver is useful to people who want to manufacture “untraceable” guns.
You could change the law so that barrels have to have serial numbers and accomplish nearly the exact same thing as completely banning 3d printers.
Also buying a kit, 3d printing a lower receiver, snd assembling an effective firearm is about as difficult as buying a kit to assemble an 3d printer and using existing open source slicers (or modifying a 3d printer to let you use an open source slicer).
And if 3d printers are as dangerous as the proponents of this legislation thinks they are, people would just hop across the border to Nevada and use a 3d printer there.
> They're much less concerning to me because their rate of fire is extremely slow.
Not if you make a multi-barreled one and spend some time practicing. [0] Go check out some youtube videos... some of those kids are kinda nuts. But, okay, I believe you when you say that you're unconcerned about firearms with a low rate of fire.
> I don't know the details of what can be printed in a consumer-grade printer...
I expected this, yeah.
With a consumer-grade 3D printer you can't print the things you need for anything much better in the rate-of-fire department than a "slam fire" firearm. To make a semi-auto firearm, you need springs and a rod that are fairly strong and heat resistant to make the shock absorber that drives the gas-powered mechanism that ejects the empty cartridge after firing. You also need a cylinder that can contain the pressure and heat of the gases from the burning powder in the cartridge, as well as provide a straight guide for the bullet so that you hit what you're aiming at. For reliability, you'll probably want a metal firing pin. Your rate-of-fire concerns mean that you're worried about magazine-fed firearms, so you also want decent springs in the magazine to reliably feed ammunition.
Because you're concerned about firearms that aren't low rate of fire, all of these things need to reliably perform for more than a handful of firings.
> ...but I've seen things claiming to be pretty complete kits...
Looks like you never did the inventory on those kits. For any kit that is actually going to give you what you're scared of, you'll find that the parts of the firearm that actually do the hard work are not going to be 3D printed.
You've professed ignorance of what can be printed in a low-end 3D printer, but do you happen to be familiar with what can be made in a high-end 3D printer? If you are, would you care to tell me how much money you need to pay for a 3D printer that will generate reliable springs, shock-absorption assembly, and barrel that will function for more than a couple of shots?
> If it's not actually possible to 3D print an effective gun, perhaps someone should make that argument in detail.
So, there's the video at [1]. That's the guy's second attempt at a barrel for .22 caliber ammo, and it's... ineffective. I also want to call your attention to this short Popular Science news article from fourteen years ago. [2] Notice the discrepancy between what's described by the headline and the article body "3D printed assault rifle made from ABS!" and the truth exposed by the picture of the actual firearm... the important parts of the firearm are made of metal, rather than ABS.
I mention that Popsci article to demonstrate to you that the "3D printed gun" hysteria is not new, and the folks making the claims that one can just go up and get a fully-functional magazine-fed semiautomatic pistol or rifle straight (and entirely) from one's consumer-grade home printer are just lying.
[0] Anyone who has decided to use "slam fire" firearms to harm someone is clearly unconcerned with safety, so they could even make several to dangle from their belt to increase their effective rate of fire.
While commuting can be chaotic, I don't think what you're calling out is the central point of the article.
Spaces that are less chaotic - coffee shops, gyms, even elevators - are places that people used to be able to strike up random short conversations with each other. Sometimes these would become longer. Sometimes they even turned into friendships.
It still happens of course, but I, like the author, am saddened that casual socialization seems to be on the decline.
I like to say Hi to people in elevators. Some reciprocate. Some find it awkward. But I don't even bother saying Hi to those wearing AirPods since that would be rude, interrupting them. Could an interaction have brightened our mutual days? Could we have become friends? Who knows, AirPods preempted it from the start.
We've found that methods like this substantially increase the quality and reliability of coding agent output. The ability to run code in a sandbox, drive an interactive session using a browser or API calls or other apps, and visually confirm output via vision models all adds up to plugging a big hole in the feedback loop for agent modifying a complex codebase.
We've had agents go as far as interactively testing how our product responds in video calls by launching our full stack in a set of docker containers (app, api, db, queues, etc.), all inside a larger sandbox, populating test data, connecting the mock system to a real video call solution like Google meet, and injecting audio and video to test the response. End-to-end, like a real user flow.
It's not perfect yet, but if you are a skeptic on the ability for AI agents to productively modify a complex product, I'd highly encourage you to play with a setup like this before ossifying your conclusions.
Didn't Claude Fable do this? (and I think codex and Claude Code in general)
When Fable was around last week, I was smitten with it. I took an executable file from an old DOS application, told it to port it to the Mac. From that single prompt, it was able to set up a test rig with Dosbox to execute the application after already disassembling and gathering as much info as it can and then continuously refine the output application while testing it against the original file. 15 minutes later it had an 99% identical looking and functioning application running natively on the Mac. Sone final refinements got that to 100%.
To be fair though, models might be changing the calculus for what constitutes a vulnerability that is too small / too obscure to care about.
If AI is reducing the cost of using the long tail of small vulnerabilities or is making possible chaining them together into something more profound, then those small, less-concerning issues might requiring addressing in a way that was previously not required.
Kinelo | kinelo.com | San Francisco | Onsite (with a bit of flexibility)
Kinelo solves "context myopia": autonomous and semi-autonomous AI agents don't know what they don't know, so they can't search or find the background information required to accomplish tasks accurately and effectively. The result looks like slop but it's not due to model performance, more due to workflow integration. From there, we're rapidly moving into organizing humans together with autonomous "AI coworkers" into workflows that span both. Our goal is to change how organizations are run and structured.
We're hiring for two roles right now:
- Product Engineer / Senior Software Engineer (https://job-boards.greenhouse.io/kinelo/jobs/4205846009) -- do you have the software engineering wisdom not yet encoded in an LLM? and do you have product sense and the willingness and desire to think several steps ahead in the market?
- Founding GTM Lead (https://job-boards.greenhouse.io/kinelo/jobs/4215784009) -- there's no playbook for go-to-market in the category we're creating + the landscape is changing daily. Come architect a strategy and then move mountains to execute on it and help us grow insanely fast.
Competitive salary, meaningful (~first batch of hires) equity, founder previously exited to Apple, strong fundraising history (~$10m).
Please email us at jobs@kinelo.com.
If you want to stand out:
- For the product engineer role, tell us how you think the job of a software engineer is changing.
- For the GTM role, tell us about a time you got massive results with limited resources, all through your ingenuity, creativity, and hard work.
The key is consciously making that choice. What people don't want is to be forced into a way of living by others. People can accept what they have a hand in discovering and choosing.
reply