I've been using git spice (https://abhinav.github.io/git-spice/) for the stacked PRs part of graphite and it's been working pretty well and it's open source and free.
You literally just shift the window over by to the next token once you reach the max amount of tokens you want for context window, NOT with what you train on, (only limited with memory now)
This has obvious issues since you're now losing information from the now unseen tokens which becomes significant if your context window is small in comparision of the answer/question you're looking at. That's why companies try to give stupidly large context windows. The problem is they're not training on the large context window, they're training on something smaller (2048 and above). Due to how attention is setup, you can train on a small amount of context and extrapolate it to any number of tokens possible since they train via ROPE which trains the model because on words and their offset to the neighboring words. This allows us to effectively x2,x3,x10,x100 the amount of tokens we generate vs train with with some form consistency BUT still cause a lot of issues consistency wise since the model approaches more of a "this was trained on snippets but not the entire thing" situation where it has a notion of the context but not fundamentally the entire combined context
That’s a very basic way to keep the LLM inferring past the context window size (there’s better, smarter ways) but that’s not at all what the question was which is how they train a 2M token length window. My understanding at a basic level is that you need corpuses that are >2M in length for training data which is where the problem comes in for - there’s only so much long form content and it’s swamped by all the smaller stuff. I think there’s probably tricks now but I suspect it’s still largely an open problem.
AFAIK nobody does that. They train on much much shorter text but with use tricks in the position encoding steps that can be extrapolated by the LLMs. Lile ROPE and YARN etc.
AFAIK (not much) it definitely helps to train on longer sequences even with rope/yarn and is needed if you care about long context performance (and not just the long context capability).
Funny I'm a professional engineer and happily call myself "vibe coding" when writing code these days, it started as tongue in cheek, but now I've embraced it.
Being good at vibe coding is just being good at coding, the best practices still apply. I don't feel we need another term for it. It'll just be how almost everyone writes code in the future. Just like using an IDE.
If you’re looking at the AI-generated output then you’re not Vibe Coding. Period. Let’s not dilute and destroy the term just as it’s beginning to become a useful label.
People of course often do read (and even modify) the model-generated code, but doing so is specifically not “vibe coding” according to the original definition, which was not meant to encompass “any programming with an LLM” but something much more specific: https://simonwillison.net/2025/Mar/19/vibe-coding/
The whole point of the term is to convey that you are only looking at the output and not looking under the hood. If vibe coding a UI, you only look at the UI, not the CSS.
> The developer does not review or edit the code, but solely uses tools and execution results to evaluate it and asks the LLM for improvements. Unlike traditional AI-assisted coding or pair programming, the human developer avoids examination of the code, accepts AI-suggested completions without human review, and focuses more on iterative experimentation than code correctness or structure.
likewise, for a lot of frontend I "vibe code" it. I mostly don't look at the code anymore while doing it either, after I get where I want, I will look through the code and maybe clean stuff up. But a lot of the code is fine. Works really well I find. (using Augment Code with Claude Sonnet 4.5).
Not sure why you're downvoted. I've tried Robotaxi a few times and has been great. They still have a safety driver these days and wait time is a big high though.
How are you writing your prompts? I usually break a feature down to smaller task level before I prompt an agent (claude code in my case) to do anything. Feature level is often too hard to prompt and specify in enough detail for it to get right.
So I'd say claude 4 agents today are at smart but fresh intern level of autonomy. You still have to do the high level planning and task break down, but it can execute on tasks (say requiring 10 - 200 lines of code excluding tests). Any asking it to write much more code (200+ lines) often require a lot of follow ups and disappointment.
This is the thing that gets me about LLM usage. They can be amazing revolutionary tech and yes they can also be nearly impossible to use right. The claim that they are going to replace this or that is hampered by the fact that there is very real skill required (at best) or just won't work most the time (at worst). Yes there are examples of amazing things, but the majority of things seem bad.
I have not had a ton of success getting good results out of LLMs but this feels like a UX problem. If there’s an effective way to frame a prompt why don’t we get a guided form instead of a single chat box input?
Coding agents should take you through a questionnaire before working. Break down what you are asking for into chunks, point me to key files that are important for this change, etc etc. I feel like a bit of extra prompting would help a lot of people get much better results rather than expecting people to know the arcane art of proompting just by looking at a chat input.
I am just a muggle, but I have been using Windsurf for months and this is the only way for me to end up with working code.
A significant portion of my prompts are writing and reading from .md files, which plan and document the progress.
When I start a new feature, it begins with: We need to add a new feature X that does ABC, create a .md in /docs to plan this feature. Ask me questions to help scope the feature.
I then manually edit the feature-x.md file, and only then tell the tool to implement it.
Also, after any major change, I say: Add this to docs/current_app_understanding.md.
Every single chat starts with: Read docs/current_app_understanding.md to get up to speed.
The really cool side benefit here is that I end up with solid docs, which I admittedly would have never created in the past.
You can ask it to do this, in your initial prompt encourage it to ask questions before implementing if it is unsure. Certain models like o4 seem to do this more by default rather than Claude that tends to try to do everything without clarifying
I mean if you ask Claude code to walk through what you should do next with you it'll ask lots of great questions and write you a great TODO.md file that it'll then walk down and check the boxes on.
You don't exactly need to know prompting, you just need to know how to ask the AI to help you prompt it.
I feel this way too. It was scary the first couple weeks, but I'm glad I gave it more of a chance. Over time you learn when to trust it and what situations it you need to take over, and then it just becomes a normal part of driving with less stress on me while my brain energy is spent looking for problems instead of keeping up with traffic or in the lane.
Uber has used this tactic many times in their early days. It mostly worked because citizens got used to cheap rides and got mad at their government for taking it away.
It is probably the most reliable yeh, tho spending time here I’ve grown increasingly aware that the great firewall is more than aware of this vpn traffic, even if it’s wrapped up to look like normal traffic. They periodically will seem to ‘dial down’ the internet, especially at politically sensitive times. They are fully aware great swathes of the populace and visitors use VPNs, and they choose to allow it. They’d rather control and monitor than inspire even more opaque channels.
You're right - but also very wrong. People can be optimistic while still having an existential crisis about what they see/feel is happening. That's what makes them believe they can influence and change the world, and why they even try.
What we're seeing in tech today feels like it started in the 80's, and before that point computers, etc. were viewed by some as the downfall of humankind and by others the saviour.
Oh, man, I can still remember how when barcodes started showing up on products in the late 70s that there were people saying it was the mark of the beast or a precursor to it or that it was some government ploy.
i'm sure there's lots of tech optimists in absolute numbers, but my personal experience in the bay area has been that a vast majority of my friends in tech and tech adjacent have become highly pessimistic of the current tech landscape, AI, VR, etc.
All you have to do to see this is to look at how some of the biggest tech CEOs and moguls like Steve Jobs, Bill Gates, Zuckerberg, Elon, Jeff Bezos, etc. treat giving their children screens to babysit them. Pretty sure most of them don't, because they know how bad that is.
They explicitly share that they don't let their kids use so much screens etc. And they have always been huge advocates of this because they want responsibility for their technology's effect on children to be shifted to parents. So, by themselves modeling how to set boundaries for their kids, they bolster the claim that good parenting is the issue, not the apps designed to be addictive
Yeah, the dreams of what computers could do around Windows 95, and what the Internet could do around Windows 98/Windows 2000... it felt amazing as a teenager who wanted to go into computer science. IMHO social media heralded the beginning of the end, although no-one knew that at the time.
A lot of the 90s nostalgia is just the same rose-tinted glasses as all generations experience, but I think in this one dimension it truly felt a lot better back then.
I have a serious problem with calling everything social media and (more importantly) how it spells doom for this that and the other
If you want to criticize specific companies - yeah. But I literally do not understand what people are talking about when employing the usual "Social media was a mistake" type stuff
I think it boils down specifically to the social media pivot towards algorithmically curated feeds designed to prioritize engagement above all else. Platforms that did not make this change in the '10s still feel much healthier than dumps like Facebook and Xitter.
One of the big selling points of Bluesky right now is that it does not do this and that is why it feels so much like 2010-era Twitter.
People typically refer to the data brokerage economy that sprung up thanks to social media. As well as being experimented on (A/B tests) for dubious reasons.
But, the data brokerage economy was the only way such a thing could connect people from different economic backgrounds, etc.
Without that, to pay for social media, the options would have been paid apps, which would never have had the escape velocity to be what current social media is today
> the data brokerage economy was the only way such a thing could connect people
For people profiting from that economy, of course. For everyone else? Not at all.
> what current social media is today
...the social media of today is undesirable. People would absolutely pay for social media if the product was good enough. In fact, I believe we're heading towards that future[1].
"Free" search exists. However, so does paid[2].
The idea that companies need hyper growth, "escape velocity," or whatever to succeed is outdated.
> For people profiting from that economy, of course. For everyone else? Not at all.
How else can such a site exist, grow to such scale, and not fall over if it doesn't make money?
> People would absolutely pay for social media if the product was good enough. In fact, I believe we're heading towards that future[1].
People won't do this. A very very small percentage of people will do this. Time and time again that's been proven to be true. Which is fine, but the globally connected social media could never have existed by charging every person on Earth.
Mark my words, paid search will not overtake free search. The people who value paid social media over free social media enough to actually pay are always going to be a very small minority (albeit well represented in this forum)
(apologies for the ridiculous length of this, and totally understand if you pass up reading the comment, it's also a bit incoherent/lazy/quickly put together)
----
To me, the real problems are the following. All related to "social media" but not social media in itself.
1) The Stream (timeline, whatever you want to call it). Instead of static single "pages" that linked to each other as separate entities with their own internal logic, everything became one flow, and more importantly "impermanent" (e.g. the way we think of "content" is no longer about an archive or repository of things, but rather, moment to moment). A reverse chronological sorting isn't bad in itself. Even the old long form setups like blogging and livejournal etc all supported this method of viewing content - but add in the following issues for a perfect storm.
2) Limited text length (blame Twitter and Facebook, and item 3). By limiting the length of text we create another "short attention span" mindset. Twitter, by it's origin, did have a specific need for short text due to its SMS origin. W/FB - the old 420 char entries was limit to what one can say and encourages short, quick, non-discursive entries. This may have been prudent both to reduce data retention costs and keep people "engaged" (though I don't think the side-effects of this were intentional at first)... This helps to feed the short dopamine burst/feedback. Neither of these were intended for deep discourse.
3) iPhone (and Android; But, frankly it was the iPhone that really got the phone and modern style of apps/ecosystem to where we are now). Having a smaller screen that was "always on" helped to push short text & images instead of longform content, plus the instant feedback loop, wherever you were.
4) "Always On" internet via small devices vs "Log In" standalone bulky systems. Instead of coming home to a computer as a separate space that you had to intentionally log into (in dial up days; or at least turn on in cable/fiber days), we no longer had to set aside a sort of "sacred place". The space no longer being localized, it was everywhere. While this is part of item 3, the form factor of 4 meant that longer form content is still easy to produce whereas with the form factor of 3 means interaction is limited and more difficult to use due to the limitations of onscreen text/keyboards (though advances made it easier, it's still not the same as sitting down and writing a big ass blog post).
4) The death of "blogging"/Google Reader. While blogs exist and come and go in fashion, Google killing reader really was a sort of deathknell. This was more due to the rise of "social media" though, not a precursor. But it helped to cement that long-form content was "dead". Blogs still exist, but aren't near as popular.
5) Images and Embedding/Walled Gardens: The removal of easy interactivity between platforms locking people in to the systems. While in some ways this is a nice convenience (take a URL and past into FB and BAM there's the image all formatted nicely (usually)) - unlike say Livejournal where you still had to type in the img src tags (or use the visual editor that would pop up a window for you to paste the url, that would, after saving/posting the page, would render the image). Sharing memes and other content with a "share" button meant you no longer had to create the content, you could just share what someone else said if you thought it was funny, were angry about it, etc. Meme replication become much less intellectual and lazy. "Engagement" didn't come to mean actual long thoughtful replies, but just an emotive "click" on an emoji to signal ones preferences towards said content. No need to discuss. Sure the option to do so was there, but who has the time when the internet is always on and there's always more content.
6) "The Algorithm" - that which keeps the "engagement" happening and as fast and deep as possible. Just one more click. One more hit of interaction. Keep the people "engaged" as long as possible on a platform.
Of course, long form still exists, and forums like HN allows for a better more thoughtful input on these things. But HN isn't really "social media".
I'd consider blogs and similar forms a proto-social-media (or, with livejournal, honestly I'd consider it the original social media, but without all the algorithmic bullshit that let you organically find people with similar interests).
Finally - I think the drive of FB to have everyone put everyone they know IRL and the ease of "add all your friends from your contacts" makes that also much easier. By adding IRL people, it changes the dynamic. When a lot of us came online, we were looking for an escape FROM the people IRL. We wanted a new space, away from "out there". Eternal September, the normies invaded. They were everywhere, and at all times. And people add them because it's rude not to. And the escape that the online world provided with similar minded people you were hoping to find was reduced back to the friction of "the real world" and the same fights. Note that alone isn't bad in itself. A bubble is also a bad thing.
But further:
7) Ideological Techno-utopianism that doesn't take into account all the corporate power behind the algorithms, the data sharing, and the means of keeping engagement (by rage-baiting, bad-faith-actors, etc...) I used to be one of them, I used to crow about JPB's "Declaration of Cyberspace Independence" in the 90s. Even until maybe 2015 I had hopes. But slowly I began to realize these architectures amplify the worst tendencies of social interaction, purposefully or not. My point with the 90s utopianism is that this Red Book reads like it's 20 years out of date, even by the time it was written, but maybe not. Maybe it just took time for the iPhonization/SocialMedia effects to take root and alter society on the level it has.
"Convenience is the enemy". I don't know the answer, frankly. I know there are a lot of people who try to bring back "the old days" with things like indieweb movement and small shell accounts for kids to try out what the old dial up world was like, and people get interested, but the mass pull of the large places is too great. The power of having an audience (which is the magic of social media - that stream makes it EASY for everyone to just dip their foot in and out whenever they want another nibble from the Dopefish^TM on their feet). There's the tipping point where everyone's friends finally jump ship, this happened with Livejournal and Myspace over to Facebook. Or now Twitter/X to Bluesky. Early adopters watch as the masses come to their little place, sometimes they are happy about it, but other times they see what they thought of as "their" space now invaded by an outside culture. Bluesky probably not so much since it's still quite like Twitter, and most of the people coming in are from Twitter, but for other places there are cultural shifts. Another big change I can think of is how Reddit went from a pretty technical place early on to one full of lazy bro joking when Digg users migrated en masse, and the vibe shifted radically even though Reddit was more bountiful with content.
Some of this will always be a tradeoff (it's called Eternal September for a reason after all), but some of this are design choices. I wish I could say that those designing these things honestly have the best interest of society at heart, but after watching enshittification play out the past decade+ it really feels like good intentions don't mean shit and a lot of hard work needs to go into making a productive place.
It helps when engaged community members are helping run things (whether good reddit mods, or mods on metafilter, or people like dang here on hackernews). Free for all media isn't bad, and I'm not saying every site needs to have mods, etc... But... It does seem like people who have an intentional community and a vibe they are trying to retain can help inculcate a better social atmosphere than shoving everyone into a clowncar walled garden and expecting civil society (especially when algorithms feed on making those dopamine hits for engagement).
I agree with all of this, but would add:
The "social media" we see isn't the entire landscape. There are lots of small communities hanging out in spaces like unpopular subreddits, random internet forums, discord servers. It's just the 10GW burning sun that is popular social media can feel like it drowns that all out. And, I think there's a bit of a nagging feeling of irrelevance that can creep in when you're in these niches.
In the old days, you were small groups exploring the uncharted. You were in the minority because you were part of a group of pioneers, and I think there was a sense of optimism and excitement around that. I think a lot of the magic is gone precisely because tech is now ubiquitous and mainstream.
I can't help but wonder if it's my nostalgia as well, but I feel like 1995 - 2005 was the golden decade. It was at the point that technology was actually good enough to enable a lot of stuff (broadband, large storage media, etc), but it was still enough in its embryonic state that the novelty was still there.
Good point. The early exploration did make it a lot smaller. I think moving to where people end up adding IRL friends instead of "random strangers with similar interests" (I still think Livejournal had the best way of finding online people and building those relationships as people and not just meme-sharers/reacters to dopamine). I use another age-old online community that's been around since before 2000, and with good moderation and community support, while it's been up and down, generally "keeps the faith" with the users. And that's a pretty good timeline I think 95-2005.
Thanks/YW. Yeah, a lot of those old 90s libertarian/community building/better world idealism hit me right when I was coming of age. I'm a pessimist now, but those early days were inspirational in what could be.
From the viewpoint of developer culture I think Facebook is better night and day than Google.
Google's culture is hire 15 geniuses from the Ivy League with 130+ IQ and make them fight with a 40 minute C++ build and a balky Kubernetes culture because "we only hire the best"; YouTube and the advertising system are a money printing machine, the team works for 3.5 years at something that get canceled at the last minute.
Facebook is much more oriented towards greasing the skids with the goal that a fresher developer would be able to push a change to a shared development system the first day on the job.
Look at React vs Flutter.
Like Microsoft, Google is thrashing around looking for "the next big thing", sometimes like a mindless beast. I've met person after person who told me that they went there because they believed they could make an impact and came to the conclusion within a year that they couldn't make any impact at all.
Facebook on the other hand is still run by the founder and it is pushing hard to develop a technology that he believes in even if the rest of us don't. It's a riskier strategy than Microsoft or Google who are likely to stumble on another multibillion dollar business despite themselves.
When I started at Facebook one of things everyone did during orientation was to put up a diff changing the default text of the search bar at the top of the homepage. This was a good way to familiarize people with the end to end flow of making changes. They didn’t say this explicitly but I think it was also intended to give people a “holy shit” moment when you realize that your silly code change is one button press away from 10 figure page views.
The idea was you make the change, take a screenshot of the result on your dev instance, get it stamped by your “mentor” and then abandon it. AFAIK this had been going on for a while before I got there.
Fast forward a few months and I see a sev pop up. “Default text on www.facebook.com search bar says ‘I am a search bar!’”
Was thinking the exact same thing. My hypothesis is that Facebook is behind because just as the Red Book suggests, its overall culture has been to build things primarily for people (freeloaders), and not for enterprises (who are actually the ones paying) like Microsoft (originally) and Google (eventually) do. If you look at Microsoft and Google, a lot of their garage projects are to build products that are useful to enterprises first and are hence more successful because it's easier to bring in cash with them. Their end-user products, meanwhile, eventually end up getting extinguished.
I worked at a Microsoft shop circa 2007 on ASP.NET/Silverlight systems and we had a subscription to MSDN. This was a pretty good value in terms of the headliner products like SQL Server, Visual Studio, Sharepoint, etc. We were getting discs for everything in MSDN and had a cabinet filled with discs and had an employee whose job it was to catalog it. I was amazed how many discs we had of enterprise software from Microsoft or some company that Microsoft had bought that I'd never heard of we had.
I have mixed feelings about marketing to the enterprise market. On one hand you can build some large and interesting things that deliver a lot of value, particularly in the semi-custom area. For many reasons I can understand the viewpoint of a salesman on commission. On the other hand I was really depressed after I'd talked with about 20 vendors in the "enterprise search" space and found that none of them particularly cared about search relevance and didn't regularly do evaluation work unless they were participating in TREC to gain industry visibility. Sometimes enterprise products have a lack of refinement or even basic quality compared to consumer products.
If I was going to get back into business development I'd do it with a keen understanding that getting the politics right and the software wrong is better than the reverse when it comes to making a living. I think I'd find myself hard to motivate in that situation.
I think Microsoft's end user area where they are the most pathological now is XBOX, Game Pass and all thought. Looking from the outside it looks like Dr. Evil has decided to buy the whole game industry to put it out of business and force people to pick up another hobby.
I agree, but you don't get to weigh everything out on a scale and be measured by the balance - even if you could. How would you compare say, GraphQL, with providing a mechanism for 24/7 cyber bullying, or sharing photos with grand parents? Which side of the ledger does React even go on?
Giving corporate welfare,tax cuts, and bailouts to mega corps. While not equally helping small businesses, in fact quickly shutting them down when they fail to pay taxes or get permits. Regulatory capture and certain regulations too, preventing people from ever breaking out and owning their own business. It's technically possible but practically not. So they (many blue collar workers) get stuck as working poor.
None of that has anything to do with Cuba specifically keeping salaries for their government workers and most industries below 1 USD a day, a common bar for poverty.
This has always been the case, particularly for Facebook. Even when they acquired Instagram, the vast majority of the comments were negative (the top comment called it well though).
I do think that HN has gone from indie entrepreneur/ real startups to Big Tech and then back again over the time I've been hanging out here.
