Great work! Though I see some people criticizing the usefulness of this. Are they being sarcastic are just really not understanding what is being discussed here? I can't tell. Maybe as an interesting follow up you could train the transformer on something with a more limited vocabulary. Spoken language is complex but a transformer can work on less complex domains like music or PET-BASIC code.
Thanks! The training corpus and code are in the repo if you want to try... Training takes just a couple of minutes on an RTX 3090. Don't get your hopes up too high, though. I can imagine that code would be harder, not easier. Even modest sized transformer models struggle with proper GOTO targeting. It would look like BASIC, but essentially it would be friendly gibberish too.
The general take here seems to be "everything eventually passes". That isn't always true. I wonder how many people have a primary computing device that they don't even have full control over now (Apple phones, tablets...). Years ago the concept of spending over $1k on a computer that I didn't even have the right to install my own software on was considered ridiculous by many people (myself included). Now many people primarily consume content on a device controlled almost entirely by the company they bought it from. If the economics lead to a situation where its more profitable to sell you compute time than sell you computers then businesses will chose to not sell you computers. I have no idea if that is what ends up happening.
It's worth keeping an eye on this HP-rental-laptop thing.
Personally I think it will be a big headache for HP, people can be hard on laptops and HP is already not excited about consumer support (i.e. mandatory 15 minute wait time for support calls). But if they make it work, I think there's probably a good number of people who feel like they need a laptop but don't care so much about the specifics and want to keep their costs low (as all of their costs appear to be rising right now).
Rental seems to be about corporate laptops. Companies just want things to work at a predictable cost. They are already replacing laptops after 5 years even if they work. They are already replacing a few laptops that break in less than that 5 years. In short they are already renting the laptops, they are just paying the price upfront and then using accounting to balance it out. Rental just moves the accounting, but otherwise nothing changes.
For consumers who don't replace their laptops on a schedule it makes less sense.
I'm also very skeptical of "everything eventually passes" as it pertains to hardware prices. Right now, prices are high because supply can't keep up with demand. But if/when supply increases to meet demand or demand decreases, there's no reason for companies to drop prices now that consumers have become accustomed to them.
Exactly. Production of RAM, SSDs, etc is spread out enough that no one company/country/fab has a stranglehold on the market. Right now anyone with a memory fab has a money printer. More people will build fabs, just like they did last time. It takes a bit but they'll get built.
> I was under the impression that only ASML could build SOTA fabs.
You're correct that this is true, but wrong in that it is not relevant.
SOTA fabs are used for cutting edge CPU/GPU chips. You do not need a SOTA fab to create DDR5 memory or SSD memory. There are a number of companies able to create the lithography machines to do so, and dozens of actual operators of those machines.
Why would that concern you unless you are working on the cutting edge and the very limits of that hardware?
The current generation is insanely fast. I am planning to get a gaming PC for my wife and a mix of gaming + workstation PC for me (or maybe just base it off of the Ryzen 9950x3D and call it a day). We plan to hold on to them for 10 years.
I don't care if anything 6x faster comes out. For what I need the current generation is even an overkill.
I'd even go as far as to say that it would be quite OK if that's the very last generation and no further hardware development ever happens.
I am on the edge of current available hardware and do feel the desire to upgrade. As stated before, I am unhappy with the current maximum when combing frame generation, resolution, and graphics quality.
My dream spec is UE5 at 120hz on an 8k oled. I think that sounds like a super sick experience I would buy tens of thousands of dollars of hardware for.
Not necessarily. Many people grew up with PCs and laptops but now mostly use their phones, because outside of specific jobs or hobbies, everyday computing needs are heavily optimized for mobile-first.
(A large factor here is, obviously, the cloud. With photos, documents, e-mail, IMs, etc. all hosted for cheap or free on "other people's computers", the total hardware demands on the end-user computing device is much less. Think storage, not just RAM.)
It's true even in tech; half a year ago I switched my phone to a Galaxy Z Fold7, and I haven't used my personal laptop since then, not once. I have a separate company laptop for work, and I occasionally turned on my PC, but it turns out that a foldable phone is good enough to do everything on personal side I'd normally use a laptop for. So here I am, with my primary compute device I don't have full control over - and yes, I'm surprised by this development myself, and haven't fully processed it yet.
> Not necessarily. Many people grew up with PCs and laptops but now mostly use their phones, because outside of specific jobs or hobbies, everyday computing needs are heavily optimized for mobile-first.
It's a deeply flawed comparison, because many of the things we do with a phone now wasn't something we'd do at all with the computers we grew up with. We didn't pay at the grocery store with a computer, we didn't buy metro tickets, we didn't use it to navigate (well, there was a short period of time where we might print out maps, but anyway..)
When I grew up, I feel like our use of home computers fell into two categories:
1. Some of us kids used them to play games. Though many more would have a Nintendo/Sega for that, and I feel like the iPhone/iPad is a continuation of that. The "it just works" experience where you have limited control over the device.
2. Some parents would use it for work/spreadsheets/documents ... and that's still where most people use a "real" computer today. So nothing has really changed there.
There is now a lot more work where you do the work on services running on a server or in the cloud. But that's back to the original point: that's in many cases just not something we could do with old home computers. Like, my doctor can now approve my request for a prescription from anywhere in the world. That just wasn't possible before, and arguably isn't possible without a server/cloud-based infrastructure.
Phones/tablets as an interface to these services is arguably a continuation of like those old dumb terminals to e.g. AS/400 machines and such.
> It's true even in tech; half a year ago I switched my phone to a Galaxy Z Fold7, and I haven't used my personal laptop since then, not once.
(edit: I'm broadly in agreement with your comment & observations, so I don't at all mean to come off as argumentative for the sake of being argumentative. You just got me thinking about how that situation might have been handled thirty or a hundred years ago.)
> [...] my doctor can now approve my request for a prescription from anywhere in the world. That just wasn't possible before [...]
I'm picking nits, but wasn't this more or less instantaneous approval possible before with e.g., a fax and a telephone? Or (although this is a bit of a stretch) a telegram and telegraph?
Ditto. My personal equipment includes a home server (128GB DDR3 ECC) and a tablet with a keyboard. It's honestly astonishing what you can do without a full-fledged laptop, if you're willing to go through some gymnastics to get there. And it travels light compared to a laptop! (The tablet, that is. Not the headless box. :-))
In a lot of ways the cloud is better than my personal computer, even if I'm on it.
There is a reason I have a server in my basement - it lets me edit files on my phone (if I must - the keyboard is and screen space are terrible compromises but sometimes I can live with it), laptop (acceptable keyboard and screen), or desktop (great keyboard, large screen); it also lets me share with my wife (I haven't got this working but it can be done). I have nearly always had a server in my house because sharing files between computers is so much better than only being able to work on one (or using floppies). The cloud expands my home server to anywhere in the world: it offloads security on someone else, and makes it someone else's problem to keep the software updated.
There is a lot to hate about the cloud. My home servers also have annoyances. However for most things it is conceptually better and we just need the cloud providers to fix the annoyances (it is an open question if they will)
Not my 70yo mom. She used to have a big gray PC but switched to a Chromebook (one I gave her) about 15 years ago, and now only uses her phone and tablet.
I "sold" my mother my personal top-of-the-line MacBook Pro ~2014... only to eventually discover it largely unused when we were probating her properties.
iPad awas the perfect device for her (I've touched one perhaps twice, in my entire lifetime).
The framing here is wrong, I think. My iPad has a lot of software on it that I use for music production, it all runs locally. Yes I had to install it through Apple's app store but I could disconnect it from the Internet and expect it to, at this point, work as long as the software on almost any piece of hardware it replaces.
Meanwhile my much more expensive laptop mostly interfaces with applications that primarily exist on servers that I have no control over, and it would be nearly worthless if I disconnected it from the Internet. Your central point is right, the economics are concerning, but I think it's been a ship slowly sailing away that we're now noticing has disappeared over the horizon.
If you have to pass the bouncers vibe check to get in, and your dancers have to pay him a 30% tax to work there, do you own the strip club or does the bouncer?
Windows PCs were essentially a hacker's paradise and almost all of my friends assembled their PC or let their PC get assembled by a friend.
There was so much variety in the components, in fact you carefully had to adjust mainboard and its chipset with the CPU and GPU as well as the RAM type. This was a lot of work, since all components had downsides and there were hidden gotchas, like for example the huge and looming CPU fan of later Intel CPUs.
If you not went for a large tower PC, you were essentially doomed. CPU fan and GPU could overlap, number of slots were a thing also whether there were enough host slots available.
Bus system, cable length, jumper settings - so many ways to make your hardware bite you.
But at least it felt even after many years as your system, that ran locally. There was a lot of protest, when Microsoft started online activation. This marked the inflection point at which autonomy of your system suddenly eroded. Also downloading huge driver updates in order to get your NVIDIA GPU working - you needed a permanent online connection which wasn't needed earlier.
Being in control of your own computing device was always a niche. The vast majority of people are not interested in computing itself, only in the output. For that majority, this is fine.
The niche is still there, probably as big as it was before. For example, as I grew weary of being subject to services I have little control over, I set up my own home server using a refurbished PC. It has been an amazing journey so far. But I don't think a normie would ever get interested in buying a refurbished Dell, install Debian on it, and set up their own services there.
As long as there is a niche of people interested in buying their own computers, there will be companies willing to fill that niche.
This is the correct stance - even in the era of "total personal computer control" the vast majority of people were not in control - they had a friend or child or parent or IT department who was really in control, they were just along for the ride.
And a large number of those people jump at getting them an iPad instead of the perpetual tech support required.
Yah but browser and computing are so much powerful, who needs to install software on their machine when web app and apps store is sufficient for general consumers
This looks like one of the rare instances where a company tried to balance their commercial interests with the interest of the fans of their products. I don't see why people would be complaining?
Does it seem to anyone else that author's have created a definition for 'vibe coding' that is specifically designed to justify their paper? Also that their premise is based on the assumption that developers will be irresponsible about the use of these tools ("often without users directly reading documentation, reporting bugs, or otherwise engaging with maintainers") so that it would actually be people killing open source not 'Vibe Coding'? Just a guess on my part, but once developers learn to use these tools and we get over the newness I think this will be great for open source. With these tools open source projects can compete with an army of corporate developers while alleviating some of the pressure on overworked under-rewarded maintainers.
Author here. We have the productvity-increasing effect of AI coding agents in the model (you're right, we're using "vibe coding" as a catch-all here). Our claim is that rewards to OSS developers (visibility and recognition in the dev community, future sponsorships, upsells to business users etc) fall faster than productivity increases. OSS devs lose the incentives to create software for others and respond to their needs.
Want to second this. Asking the model to create a work of fiction and it complying isn't a pathology. Mozart wasn't "hallucinating" when he created "The Marriage of Figaro".
But many artists are hallucinating when they envisioned some of their pieces. Who's to say Mozart wasn't on a trip when he created The Marriage of Figaro.
2x 3090's running Ollama and VLLM... Ollama for most stuff and VLLM for the few models that I need to test that don't run on Ollama. Open Web UI as my primary interface. I just moved to Devstral for coding using the Continue plugin in VSCode. I use Qwen 3 32b for creative stuff and Flux Dev for images. Gemma 3 27b for most everything else (slightly less smart than Qwen, but its faster). Mixed Bread for embeddings (though apparently NV-Embed-v2 is better?). Pydantic as my main utility library. This is all for personal stuff. My stack at work is completely different and driven more by our Legal teams than technical decisions.
So there is a language war going on in the industry and some of its justified and some of its not. Take 'agents' as an example. I have seen an example of where a low code / no code service dropped in a LLM node in a 10+ year old product, started calling themselves an 'agent platform' and jacked up their price by a large margin. This is probably a case where a debate as to what qualifies as an 'agent' is appropriate.
Alternatively I have seen debates as to what counts as a 'Small Language Model' that probably are nonsensical. Particularly because in my personal language war the term 'small language model' shouldn't even exist (no one knows that the threshold is, and our 'small' language models are bigger than the 'large' language models from just a few years ago).
This is fairly typical of new technology. Marketing departments will constantly come up with new terms or try to take over existing terms to push agendas. Terms with defined meaning will get abused by casual participants and loose all real meaning. Individuals new to the field will latch on to popular misuses of terms as they try to figure out what everyone is talking about and perpetuate definition creep. Old hands will overly focus on hair splitting exercises that no one else really cares about and sigh in dismay as their carefully cultured taxonomies collapse under expansion of interest in their field.
A 2009 honda civic can get an under-5 seconds 0-60 easily... however it does involve high a cliff.
Result Specs (as in measuring output/experimental results) need strict definitions to be useful and I think the current ones with have for LLMs are pretty weak. (mostly benchmarks that model one kind of interaction, and usually not any sort of useful interaction)
As a side note, while I know of several language model based systems that have been deployed in companies, some companies don't want to talk about it:
1. Its still perceived as an issue of competitive advantage
2. There is a serious concern about backlash. The public's response to finding out that companies have used AI has often not been good (or even reasonable) -- particularly if there was worker replacement related to it.
It's a bit more complicated with "agents" as there are 4 or 5 competing definitions for what that actually means. No one is really sure what an 'agentic' system is right now.
There is a very simple and obvious definition: it's agentic if it uses tool calls to accomplish a task.
This is the only one that makes sense. People want to conflate it with their random vague conceptions of AGI or ASI or make some kind of vague requirement for a certain level of autonomy, but that doesn't make sense.
An agent is an agent and an autonomous agent is an autonomous agent, but a fully autonomous agent is a fully autonomous agent. An AGI is an AGI but an ASI is an ASI.
Somehow using words and qualifiers to mean different specific things is controversial.
The only thing I will say to complicate it though is if you have a workflow and none of the steps give the system an option to select from more than one tool call, then I would suggest that should be called an LLM workflow and not an agent. Because you removed the agency by not giving it more than one option of action to select from.
Agentic AI comes out of historical AI, systems computing and further back biological/philosophical discussion. It's not about tool use although ironically, animal tool use is a fascinating subject not yet corrupted by the hype around intelligence.
I implore you to look into that to see how some people relate it to autonomy or AGI or ASI(wrongly, imo - I think shoehorning OOP and UML diagrams plus limited database like memory/context is not a path to AGI. Clever use of final layers, embeddings and how you store/weight them (and even more interesting combinations) may yield interesting results because we can (buzzword warning) transcend written decoding paradigms - the Human brain clearly does not rely on language).
However what gets marketed today is, as you say, not capable of any real agent autonomy like in academia - they are just self-recursive ChatGPT prompts with additional constraining limits. One day it might be more, but libraries now are all doing that from my eye. And recursion has pros but emphasizes the unreliability con/negative of LLMs.
That's not the definition I see most people using. Plenty of tool calling going on purely to structure an output, which could also be achieved by prompting.
For me, agentic means that at least at some stage, one model is prompting another model.
Very accurate. So much of succesful (from company PoV) real-world LLM use is about replacing employees. HN is still far too skeptical about how much this is in fact happening and is likely to accelerate.
reply