The army of AI agents opening PRs and issues in my open source projects has made me close PR and issue access in my active repos. It sucks because there might be someone wants to constitute legitimately but I don't want to do the labor of figuring out if it's a human or an agent opening the PR.
I'm not against using LLMs in any ways. https://tsz.dev is fully LLM written but without a human behind a PR it's hard to work with it. I've already closed a few absolutely nonsense PRs opened by weird accounts
I was curious how this thing works and asked Claude to visualize it -- mostly to see how good Fable is and I have to say, what it made was good enough for me to get a gist of it. Posted it here
Every plausibly cool electric car innovation leads me to the same thought: “5-10 years from now, the restomod potential will be wild once these come down in cost.” For this, I am imagining retrofitting a Pontiac Fiero to reduce as much weight as possible and see if extended flight becomes possible.
As an antique BMW enthusiast, I know some people that have swapped electric motors in to 2002s and 2000 CSLs and they said it was actually a pretty disappointing experience. You lose the vintage driving experience entirely.
wtf are you talking about? I drive one of these cars every day (that I need to drive, which isn’t that often really) without leaded gasoline.
It’s the punchy little m10 motor, 4 speed transmission, and incredibly low curb weight that make these cars fun to drive. You lose all that with an electric conversion.
As an aside, the most powerful F1 engine ever put on the track was made by modifying the little 4-banger m10 found in the BMW 2002. Fun fact.
> As an aside, the most powerful F1 engine ever put on the track was made by modifying the little 4-banger m10 found in the BMW 2002. Fun fact.
Supposedly BMW lacked a dyno that went high enough so it's an educated guess what the actual hp was, but definitely over 1,300hp... from a 4-cylinder (turbo charged, and only lasted a few laps, but still) engine!
I'd throw in the manual gearbox, the noise, vibration, smell, and being able to see the mechanical workings as well. We can get a punchy, lightweight vehicle with an EV conversion, but as you've heard from others, that leaves something to be desired
But yes, the vibration, sound, and feel of the incredibly simple (as in stripped down, not unsophisticated) mechanics around you all very much contribute to the special feeling of driving these cars.
My 1976 BMW 2002 is actually my first car. I used to drive it to high school more than half my life ago, and I still drive it today. I’ve driven some other classic cars, luxury and sport, and simply nothing feels like a 2002. Just a beautiful, balanced, timeless design.
I'm pretty sure the subtext of that comment was that the benefit of electric motors in retro cars is the emission reduction (and possibly cost and maintenance reduction), and not improved specs/experience. Saying "ye but it worsens the experience" misses the point.
But this is a thread about modifying 50-70 year old cars with electric drivetrains.
There are so few of these cars on the road anymore that I think very little progress is to be made on emissions levels by converting them to electric, especially when you consider that you’re ruining the driving experience.
From the animation: "Torque grows with the cube of diameter"
That seems like a pretty cool cheat-code to get more power. Perhaps it will mean you will start having cars with more distance from the ground to accommodate the large dia motor wheels.
Number 1 problem is battery weight though. Not electric motors.
I have a 84 w123 300D, and would love to add some more power to it. Lightweight hub motors would be great, but any decent size battery would be at least 200lbs+, which is hard to do on a old chasy.
Something to keep in mind with hub motors is that they’re unsprung weight, vs the battery pack is pretty much always sprung. While that’s not a huge differentiator for efficiency, it sure cuts down on the abuse the wheels and hub motors will experience
its less hard than you'd think unless you're really going for long range.
for my sailboat I am getting rid of a 300lbs diesel and a 30gallon fuel tank with a 45lbs PMAC.
That means I have opened up about 465lbs for batteries.
Now, with a sailboat you're never truly out of range -- but the point stands : these things are so much lighter than ICEs on average that there is a lot of opportunity even with battery weight as it is (and it's getting better daily).
I guess there's always the risk for a rig failure.
I looked a bit on doing the same, but came to the conclusion that it will be expensive to fulfil racing rules requiring the boat to be able to maintain speed for 5 hours ie around 25-30 NM range.
As it is now, I have about 500 NM diesel range on my boat, which is basically 3-4 days continuous runtime. Cutting it down to 25nm and 5 hours requires minimally 100kWh.
For a blue water boat, 500 NM is not quite acceptable, but can be fixed with jerrycans for a couple of dollars. An all electric blue water boat would clock in at an unrealistic 2MWh of batteries with a weight at least 20 metric tonnes. 10x the load capacity of my boat.
This is silly, but I've also wondered if you could make a boat that can anchor and recharge batteries from ambient current, sort of like stationary regenerative braking. I'm sure it would take way too long to be worth it, but it was a fun idle thought.
Perhaps the paddle wheel[0] will interest you, the spinning is used to calculate the velocity of the boat. Probably some other propeller or similar would be more practical - like a kicker motor that's easily lowered over the side. Just spitballing.
I don't think it'd be worth it considering solar options. Even wind generators are not "super efficient" in comparison but I don't have data.
I'm sure it wouldn't be worth it, because otherwise people would be doing it, it just seemed like a cool way to supplement solar or even to allow for indefinite underwater drones. Like, imagine a deep-sea research drone that could spit out an anchor and recharge whenever the battery got low. Almost certainly a case of "Cool, wouldn't really work well"
> BYD can be lighter because they skip on safety gear and proper structural elements - in my experience.
I'd love to hear more about your experience with BYD. The ex just bought one and my kids ride in it daily. I helped negotiate the sale - I drive a Tesla and I'm very happy with the BYD.
Brazilians got a bunch of them and they are super common in Brazil. Also common - broken suspension parts from driving them in Brazil.
Also, the only cars I’ve ever ridden on that the top of my head literally touches the headliner while sitting in the back seat. Other than that, they seem good?
Top-of-the-line NMC cells have energy density around 250Wh/kg at the pack level. The newer solid-state batteries can reportedly increase this to 400Wh/kg.
So a reasonable 75kWh battery pack is going to weigh around 300kg and in future around 200kg. This is... not a lot, actually. To a point where shaving off 20-30 kg from the electric motor weight is going to result in a noticeable performance/price difference.
Teslas also don't have "crazy" weight. Model 3 is 1700kg and a comparable (in size) Ford Focus is 1300kg.
The Model 3 performance model (more equivalent to the M3) weighs over 4K lbs. The model you are quoting is the lowest range and lightest of all Teslas.
i know series hybrids aren't as efficient as parallel hybrids (thanks technology connections!), but i wonder if they'd be a good candidate for fun restomods.
drop in a tiny, powerful electric motor and a small battery (crammed in whatever location is best for weight distribution), and then wire up a little genny powered off your existing fuel tank that can jump in as a range extender
Series hybrids can compete with parallel hybrids, because the full decoupling of the engine from the wheels claws back some of the efficiency you lose through the energy conversion. It's series-parallel hybrids they can't really compete with, because those are able to do the same trick, but they also lose less energy in conversion, because the engine does some of the heavy lifting.
Series hybrids are great for packaging, though. Parallel and series-parallel commit to certain packaging decisions like having a transmission, or a long, monolithic unit, because it's the mechanical coupling that buys them smaller motors and potentially better efficiency. Series hybrids don't care about any of that, so even though you have bigger motors and potentially higher losses, you have more freedom over where things go.
Personally, I think there's a massive untapped market in converting old cars to hybrid engines. You wouldn't try to upgrade the old engine, you'd design a smaller and more power dense package and rip all of the original gear out. Because electrification lets you cut the size of the engine down so aggressively, this is probably a feasible strategy. As you pointed out, series hybrids are probably best suited to this because of their packaging flexibility. As others have pointed out, there's tremendous potential there for replicating original driving characteristics using software and the electric motor. And if we're being honest, off-road vehicles probably should get rid of the transmission and low range, because electric motor torque is just better. As is, the potential for cars is enormous, but we're getting the worst possible outcomes thanks to legislation.
Reducing the motor mass by 200 kg means you've just removed 10% of the weight of the vehicle. You could theoretically now reduce the battery pack by 10% as well.
Not true. The CdA (coefficient of drag multiplied by frontal area) matters far more for range than the weight for range. That is a smaller EV, which may very well be heavier can have a higher range and efficiency.
Yeah, my comment was hand-waving away a bit of the reality of it, but swap the Fiero engine for a battery and some of these and it's got to be close to achieving full lift.
Electric engines are already very efficient (particularly compared to internal combustion). If you go from 90% to 95% efficiency, you don't save much in terms of battery.
ETA: Internal combustion engines half a century ago had an efficiency of 20%, now they're at 40%. That cuts the fuel you need to carry in half. Electric engines are near 100%, and as I said, going from 90% to 95% efficiency cuts required battery by a bit more than 5%, so peanuts.
But the motor is not the only thing that needs to be cooled. It’s mainly the battery, which has a narrow operating range. The power electronics that convert AC to DC also need to be cooled.
So you’re halving the cooling needs of the motor, which is nice but small compared to the other two. And even then, total cooling doesn’t impact range that much compared to warming the battery in cold climates.
If we halve waste-heat generation for a practical widget enough times, then we won't need liquid cooling for that widget at all. If the trend continues and gets good enough, then maybe we can get all the way to a complete passive-cooling snoozefest.
That's pretty boring, but boring is good. The systems we use every day without a thought because they boringly Just Work are, perhaps, our greatest successes.
Yup, the visualization didn't help me understand the concept any more than plain text. Superficial in the way that you would expect from a system that has no real world reference for what it is creating.
To get something better I expect more than a one-shot is needed, and the knowledge to guide it in the right way.
Thanks for sharing. I wish it was a bit more interactive especially when there are parameters, e.g. "Widen the disc and torque rises with diameter cubed" I wish there was a slider to see that effect and thus maybe why there might be a sweet spot.
Also I have "The Way Things Work" on my desk right now and can't help but wonder, could you adapt some of the pages of the book this way? It seems like exactly the kind of content that would benefit from such 3D (interactive) visual explainers.
Oof... well thanks for sharing but that's basically unusable for me. It neatly all packed in a 2MB file containing all assets, threejs, etc.
I assumed it's based on a three.js template due to the `Rendered live with three.js · Drag anywhere to orbit the model` kind of showcase but unfortunately that's not linked. I also imagine the 3D models are more that primitives (at least the arrows showcasing the flow) but I don't know where they came from, if that are also from a template or repository or if they are generated from a tube mesh.
So... I'm genuinely grateful that you took the time to share but I don't think I can do something with this except restarting from scratch, especially if it's one-shot.
I'd suggest, if you don't mind the extra effort, that you add a ReadMe.md in the repository to clarify how you did this, at least model name, version and prompt.
I mentioned elsewhere too. This was a one-shot thing that made me wow so I thought I share. You're kind with your comments but others are just hating it, even so I said 1. I don't know this motor technology and 2. It was a one shot experiment
If I had time and making a polished web page was my goal I could probably do better but this was not the point!
Very neat. Thank you for sharing! I assume this was one shot as well -what sort of prompt did you use?
I’m sure folks would be interested even in a blog post comparing just this process with different Anthropic models if that’s something you do and need a content idea. :)
Can you make a version of this that is more in the style of "the way things work" the cool inventions book from the 90s with cavepeople and wooly mamoths and that illustration asthetic?
Just as impressive was its ability to publish the source and get the version up on my personal site. That was also a one shot but aided by context and skills I have available for these purposes.
Thanks but I'm aware of it, linked to an alternative just yesterday (but might replace by the InternetArchive one instead). A Web version would be so neat as we'd be able to link to pages, even states of the interactive explanation, rather than the whole thing but it's already great to have that basis.
Yeah so the relationship between speed, power, frequency, size (both in the direction of primary flux excitation and in the direction orthogonal to both that and the movement), and torque at nominal values of current density (for a given conductor losses are proportional to the square or this value and to the total mass of that conductor in the machine; that's independent of any of the other scaling parameters; note this is absolute power not percentage) and peak flux limitations (core saturation, permanent magnet demagnetization), are sadly not trivial if you express them in a way that is even just _valid_ for the modern days where we can support electrical frequencies up to around a megahertz at scales up to around 100 kW, and even harder when you remember that core material has severe frequency dependence of it's limits.
E.g. for example for a given electrical frequency and decent radial flux synchronous machine, power density is quite static and torque density can actually be dialed quite freely from 2-pole machine (turboset in gas turbine running on the grid at 3600 rpm (or 3000 rpm outside NA and some Pacific Islands) to 40(+) (example deployed at Hoover dam, 180 rpm).
At those higher pole counts, the center of the rotor is no longer electromagnetically active, because the magnetic field lines keep to a narrow ring only about as thick as each pole is wide.
Unfortunately it's mechanically not that trivial to handle a cylindrical shell with a small air gap (this needs to be significantly smaller (about at least 10x) than the pole width) when using substantial torque and speed.
Circumferential velocity is practically limited by hoop strength of whatever the outer region of the rotor is made of, even if it's all very nicely balanced, because eventually the magnetic armature flux source (wires or magnets) will fly out.
Higher electrical frequencies limit the field winding core's magnetic permeability (magnetic field/force strength amplification relative to vacuum, for same electrical current) which hurts efficiency by dropping the useful mechanical power component of field voltage while the voltage resulting from the current (that needs to happen to cause the magnetic field in the direction of movement that causes the mechanical force) due to wiring resistance stays. (I think the permeability gives the ratio between voltage and current for otherwise identical mechanical load conditions and winding shape?)
Thinner wires have less fill factor because the insulation has to stay the same thickness as per-winding voltage stays, but magnetically inactive terminations are less wasteful (for losses and mass) when a decent number of effective turns (>>1, think >10~50 for most of the benefits) are used.
Note while the armature necessarily has an even number of poles in it's construction (north/south), the field is not forced to that.
Indeed, the iirc most smooth torque (under practical mechanical feasibility limitations and without undue sacrifice of efficiency) results from having a prime number (of field windings, in WYE-style connection) exactly one off from the armature pole count.
Note that for low losses all these torque-smoothing techniques _require_ only a single electrically directly driven winding in each slot (per mechanical field pole) and with that only GCD(field_slots, (armature_poles / 2)) windings get to share an electrical half-bridge (one single wire going to a single voltage-output terminal on the electronics board; note mainstream BLDCs have 3 of these, classic fridge compressors have 2, and modern stepper motors (e.g. 3D printer) have 4).
Any time you have multiple windings driven by different electrical source voltages you're wasting heat in the winding because the lowest-loss would require all conductor in the slot to to perfectly evenly share current.
There's just one problem with that: you need a nearby slot with exactly opposite phase to even possibly use more than a single (half) turn of "winding" in the slot.
If the voltage is still enough to not loose too much in the connections, you can use transistors developed for efficiently powering modern computer chips from comfortable voltages like 12V, but even then a "winding" has to be much longer than an armature pole to mitigate the losses of spreading the return current sideways to where a slot carries the current in the reverse direction.
Once the voltage at the transistor is over around 10V the benefits of more precise control of the field magnetization to the armature position (and how the shapes distort the field lines from anything that would look like a sine wave) could be useful.
In theory that'd also provide direct access to electronically control the air gap (well, net force normal to the air gap "surface") which _could_ be an alternative to mechanical bearings for very thin-shell constructions.
See maglev trains for a pretty practical application of using an electric motor to also levitate the "rotor" in a place where a mechanical bearing ("train wheels + bogies") performs poorly.
Hundreds of examples of axial flux motors exist online. If you look at the visualization it shows the iron cores in a perpendicular orientation with the hub. This is correct, but loses so much of what makes these specific motors interesting. The angled nature of the grey cores and copper wrapping smoothes the transition between each magnetic field.
Basically it is a pretty version of a dumbed down partially incorrect answer. With a knowledgeable user it would be very good, but he has no idea he is wrong. I’m not sure what Dunning Kreguer with graphics should be called.
You don't. You didn't know before either. The difference is trust. How do you trust it as much as you do the hypothetical humans making such representations? That's up to you.
I think humans develop expertise and brand names and get called out when they make mistakes and if they are too wrong, their reputation is damaged.
This doesn’t seem to apply to AI for some reason. It keeps generating incorrect results after incorrect results, yet people continue to trust its output.
Human trust differs from mathematical trust. And branding / marketing abuses the ambiguity.
There is no shame in a "likely to hallucinate" model that can be instantiated 1,000 times across 1,000 different machines spread throughout our planet. So, human trust is broken by machine trust.
I've starting going back to books, either at the library or e-books. Librarians are very good at telling you if nonfiction is biased, outdated, or incorrect.
> I think humans develop expertise and brand names and get called out when they make mistakes and if they are too wrong, their reputation is damaged.
Take a look at the Forbes billionaires list and some of their statements. Or maybe at the politician fact checkers. If only being wrong damaged reputations.
The question is for you to answer, first. Gotta do that work. (My answer will differ from yours.)
Then, predictably, finding the collection of supporting details + vetting the content in question.
This is an issue we, technology-folk, ought to help guide our non-tech-co-folk through engaging with, BTW. Our responsibility is rising with tech becoming more deeply entrenched / required for society's operations.
Yeah, I'm at once awed that something like that can be auto-generated (I presume?) and disappointed that it doesn't usefully or practically improve my understanding, beyond written synopses and human videos, at all.
Stuff like this reminds me that we still need a human in the loop to edit, to improve, to advance.
Auto-from-scratch just doesn't really achieve anything of actual value.
I am not an expert, but I do know some physics and I know how to read, and I’m pretty sure this is full of BS. Also it’s a really crappy visualization.
Shameless plug. I'm writing a TypeScript checker in Rust. It's not a port. I made this with a different architecture that hopefully once is done will be proven to be a better set of trade-off
I had a few of the benchmarks left alone and was working on tech debt knowing that a new model is going to be released soon. For my project (tsz.dev) Opus 4.8 was running in circles without producing results for a while for those tasks
I have been using it for less than an hour so take this with a grain of salt of being excited for the new tech.
In a project like mine (https://github.com/tsz-org/tsz) I am constantly frustrated that models were not doing enough research and were not taking into account other situations. Again and again models would produce code that would fix one thing and break 2 other tests that were "unrelated".
With Fable it seems like tasks are taking much longer (I have not seen a pull request from Fable sessions yet) but reading the transcription of those sessions I can see how it is doing the right thing by not leaving any stone unturned.
As the article says, it's hard to communicate this "feeling" about models because it is very project specific but I thought I share
In general, sooner or later you need to restructure one thing or another when requirements are changing. Good code lets you reason about a refactoring, and experience tells you when it is necessary or appropriate. Coding agents aren’t very good at the latter.
the setup is solid. there are thousands of tests and CI won't let things to merge if tests are failing.
But overall, this is pretty normal for compilers to have this sort of "unexpected" tests failing due to some work in an area. It happened to me when I was coding everything manually back in the day too
A compiler and type checker is very special case where you can fix something in the lexer or parser and break another thing in AST walker etc. tsz is well architected but those things can happen if you're not careful and that's precisely what I meant in my original comment. Fable can think how changing parser can impact checker etc...
It seems like Fable will refuse to do any work when it comes to developing LLMs or even asking questions about topics related to LLM. Simple things like asking to explain a paper fails!
From the model card:
In light of the ability of recent models to accelerate their own development, we've implemented new interventions that limit Claude's effectiveness for requests targeting frontier LLM development (for example, on building pretraining pipelines, distributed training infrastructure, or ML accelerator design. Using Claude to develop competing models already violates our Terms of Service, but enforcing this restriction through our safeguards avoids accelerating the actors most willing to violate these terms. Unlike our interventions for cybersecurity, biology and chemistry, and distillation attempts, these safeguards will not be visible to the user.
I was wondering when something like this would happen. I got my first and only two content violation warnings in Claude Code last week when asking it about something ML related. It was a real head scratcher because I couldn’t figure out what about the requests could have violated anything.
Might be worth going back and taking a harder look at what I was asking it about if it somehow triggered a “forbidden knowledge” alert. Or maybe it was just a random bug.
This seems so wide reaching if it's catching simple things like explaining a paper. Does this also refuse to help with any already developed training pipelines?
I can kind of understand the generation of synthetic data, but nerfing the assistance of training pipelines just seems like a really shitty thing to do.
So insane to me that these ai companies are perfectly fine trying their absolute best to automate as much knowledge work as possible but as soon as this capability can be turned on them they start implementing hidden interventions to sabotage anyone trying to beat them at their own game.
Not to mention these models are all built off of clearly extremely illegal abuse of human created content, which they will apparently never be held accountable for.
I wanted to try on my biology research and it refused to talk about it and proxied to 4.8. Really, only surface level conversations about topics of interest. I know this is not a topic of broad and mass interest, but limiting it for topics like that and machine learning will probably do change how I use it.
I feel like it will have to become more finely tuned on topics of biology.
It is not just biology but is defaulting back to 4.8 for me on time series/information transfer techniques that happen to mostly have papers using the technique on neural data. Other information transfer techniques are perfectly fine, even cutting edge ones, but this one happens to be new and happens to only be discussed in terms of neural data so that is a no go.
With that said, I think it is absolutely awesome. The usage is really not bad at all compared to what I was expecting.
Yes, this stuff is really annoying when it misfires. I've had all my subsequent ChatGPT conversations biohazard-contained for several days for the crime of asking it to explain a gene drive to me.
Anthropic is really speedrunning their evil arc as fast as possible. Can't use them for basic LLM research, cybersecurity, or beyond-surface-level discussions of biology and virology, but Anthropic is allowed to sell Claude to the trump administration to kidnap maduro and to bomb iran. And don't get me started on that $100M autonomous killer drone swarm contract that they applied to and rationalized as non autonomous...
> Can't use them for basic LLM research, cybersecurity, or beyond-surface-level discussions of biology and virology
Your priorities are not everyone else's priorities. The people concerned about AI extinction risk list those as three of their biggest priorities for AI to not do. Those are the people whose culture Anthropic descends from, and by their measure, those exclusions make this the least evil path.
More like Anthropic’s priorities are not everyone else’s priorities. They are in the consistent culture of being in absolute control and dictating what is good and bad, while taking any opportunity to trash and crush potential competitors (open source models happened to be mostly developed in China). All these in the name of safety and anti-authoritarian.
The day self hosted models catch up with Anthropic’s capabilities is when they will fully lose their shit. This day can’t come soon enough
Extinction risk. From population genetics... Does Anthropic even employ biologists? It's magical thinking about a field that is poorly understood by their community.
This is super annoying and imo, really limits the usefulness of this model.
It speaks volumes about what Anthropic's position as a company and its priorities will be going forward.
I doubt this kind of gatekeeping will prevent open-models or other innovation outside Anthropic to slow down.
I would imagine these guardrails, if needed at all, should be done at a legal framework level and students should not be a part of this blanket approach to limiting the usage of these models.
I doubt that. Why would you train Mythos on its own code if you don't want it to be able to reproduce it? It's not going to add much to the overall corpus.
It also tried to force usage the paid Claude API instead of claude code usage just because there's a mention of another provider we might want to plug in (which hasnt even happened) for AI integration.
Ha funny, I was speccing out an idea for real time Claude code interaction from local apps using some tricks vs using the agent sdk when I got the popup to try Fable. So of course I gave it a go, and it triggered the sensitive content warning immediately, which I was very confused by until I put two and two together.
Fun times when “safety” means both the safety of mankind, and also the safety of revenues
A foldable iPhone will definitely solve one of the biggest problems at Apple. Foldable phones won't last 5-10 years. I can see Apple making all iPhone offerings foldable down the road
Are they really having any issue getting people to upgrade? At least here in Canada, carriers harass everyone to upgrade with deals where you can get a new phone for $0 once your contract is up to keep you renewing your contract.
My non-techie parents pretty much always get the latest non-Pro iPhone every couple years because their carrier calls them and practically begs them to take a new phone.
It's extremely rare to see anyone with a phone older than like 4 years.
I get the criticism and all your previous judgments samples are valid. I also agree that title is click-bait BUT:
I know people are desperate for a Siri that works. The convince of just talking to your phone is priceless. If Apple gets this right, this is a huge deal – which it seems they are on the right track.
People are still talking to Siri for basic stuff like timers and alarms because it works, doesn't need an app, works when phone is locked or even away from you. If this works for more complex tasks like texting and general questions Apple will have the upper hand over Meta and Google in this new way of using computers/internet.
Apple also took a very clever approach for Capex and general AI strategy. Everyone knows that the best intelligence will eventually become a commodity and Apple decided to step aside from this expensive experiment. That's worth pointing out too.
Personally I hope you can also type to Siri, which is what I'll use WAY more than voice. I work from home and live alone, but even then I don't want to basically be talking to myself all day. I also live in a major urban city and while random people talking to themselves on the sidewalks certainly isn't unheard of it's not a great look, much less on a subway or cafe.
That would be a huge deal. I’m wondering how current android is being received because as far as the reporting goes it has already delivered LLM voice assistant with personal context. But I haven’t seen any reporting about the consumer market reaction. Android phones already sell better so are thry just miles ahead now?
I'm not against using LLMs in any ways. https://tsz.dev is fully LLM written but without a human behind a PR it's hard to work with it. I've already closed a few absolutely nonsense PRs opened by weird accounts
reply