Start here: https://github.com/bendlang/bend
2,111 karma · joined June 28, 2021
Personally I disagree with this. I've been using LLMs the last few years and learning a lot about techniques for getting value out of them. Even though they weren't very good for a while they seemed worth learning to me because the rate of improvement was so fast. We've now reached a point where if you know how to use them right you can get lots of value out of them.
I really don't mean to be rude here but consider perhaps the reason why you're having trouble seeing why they're valuable is because you haven't used them enough in the last few years or kept up with the developments closely enough. Maybe I'm wrong. But consider it's possible you are too.
> This has also been said every other month since 2023.
I'm sure others have said that but my perspective has been that the models are not quite good enough yet, for a while but that the trajectory is so crazy that it made sense to make the bet and invest lots of time into learning how to use them on the assumption that they would soon be good.
Even now I would say they have a long way to go. What is hard but possible today will be easy tomorrow. But quite a lot is now possible with the right techniques. Crazy things.
It might make sense to learn now, before everyone expects that you be an expert with this tool in order to do your job. On the other hand perhaps they will be easier to learn if you wait.
There is an incredible amount of nuance to be had in using LLMs and we reached the stage, probably only in the last few weeks where all the pieces have fallen into place such that if you are skilled at using the LLMs just about all of the previous problems with them can now be easily mitigated. If you learn how. There is a lot to learn.
You gotta not take it personal. Announcing on HN is a gauntlet. People can be a bit prickly. Just a bit though. You get lots of good feedback too.
If the author of the article had done just a small amount of research about bend or it's author before writing the article they would have known pretty quickly what they were saying was incorrect.
I think the larger pattern here is that nuance is one of the most valuable commodities in the AI era. If you're hand waving stuff away without even missing , you're going to miss a lot of stuff in this cycle.
This article reminds me a lot of the famous hacker news Dropbox comment.
I bet if you set Astra codex against this task on the $100 plan it would have it done at a speed that would shock you. Potentially even by the end of the day and mostly uninterrupted.
The SOTA models can reverse engineer whole codebases even from assembly now. If you have to specify with words what you want to do it's still hard but if you have a codebase to work from, you can pretty much arbitrarily convert apps and games into other languages / platforms now as if by transitive property.
You can use the AI tools but still program by hand every day for practice and fun. Programming becomes deliberate practice to keep your skills and mind sharp. Like going to the gym, but for your brain. You can use the LLMs to push and challenge yourself, setting more aggressive and ambitious goals than you might have in the old days for learning.
I used to think of software dev like climbing a mountain. Hard, and I loved it. I loved it because of the difficulty. That was the most fun part. Getting to the top, the summit. That was incidental to me. I wouldn't even defend that as a good thing. Certainly not good engineering. It was just what I liked. Any good engineering was incidental to how fun the hard parts of the puzzle were.
Then LLMs came along and I thought "this is like a helicopter. You can go straight up to the top in a few minutes." While very practical, it also led to the feelings the author described. After months of handwringing and using the tools, having existential crises - I've arrived at a new conclusion:
It's not a helicopter. It's a jetpack. And it's still hard to use if you push it to it's absolute limit. Knowing just the right spot to grab the rock as you fly by. Taking as much risk as possible, but not too much. Pushing both the jet pack (or perhaps it's an iron man suit now with latest models) and the rock climbing knowledge as far as they will go. Screwing up, falling off the wall, and learning.
And with this jetpack you can get to places you never could have climbed to without it. Not just the top. Weird, interesting spots on the mountain.
That perspective, which I've taken to over the last month or two has helped me find the fun again. The existential crises have ended and I'm back to feeling about software dev the way I felt a few years ago. Boundless joy and excitement. Probably more now.
Maybe not. Could just be kind of accidentally ironic with the author picking up LLM quirks from using them a lot.
Funny in either case.
A lot of handwringing about the security implications but I think the accomplishments of the swarm itself are the most interesting. Next rung up on the ladder of abstraction I suspect.
If an API is exposed you can just have the LLM write something against that.
Giving an LLM a computer makes it way more powerful, giving it a kubernetes cluster should extend that power much further and naturally fits well with the way LLMs work.
I think this abstraction can scale for a good long while. Past this what do you give the agent? Control of a whole Data Center I guess.
I'm not sure if it will replace openclaw all together since kubernetes is kind of niche and scary to a lot of people. But I bet for the most sophisticated builders this will become quite popular, and who knows maybe far beyond that cohort too.
Congrats on the launch!
You'll annoy the hell out of some people, and thats fine. They can find other people to spend time with.
You can probably find a good community where you are, and if not just move to SF which is something like the autism homeland. Being autistic there is valorized and even imitated in sort of amusing ways.
Masking is a kind of hell, living someone else's life. Unmasking and living as yourself feels scary at first but the people who will love you that way can only find you if you live that way.
Been on the lookout for an open source version but they all seem kind of unessecarily bulky or otherwise poorly maintained.
Would be interested in suggestions anyone has for whole apps or libs that work well when glued together for this purpose.
After a while, the more mature Linux engineers start going the other way. Ripping out as much as possible. Stripping down to the leanest build they can, for performance but also to reduce attack surface and overall complexity.
Very similar dynamic with k8s. Early days are often about scooping up every CNCF project like you're on a shopping spree. Eventually people get to shipping slim clusters running and 30mb containers with alpine or nix. Using it essentially as open source clustering for Linux.
I use it a lot for mapping out of the initial concepts but I find one of the best use cases is after understanding the basics, explaining where I need to learn more and asking for a book recommendation. The quality of my reading list has gone up 10x this way and I find myself working through multiple books a week.
Great for code too obviously, though still feels like early days there to me.
There is a secondary market for OpenAI stock.
It's not a public market so nobody knows how much you're making if you sell, but if you look at current valuations it must be a lot.
In that context, it would be quite hard not to leave and sell or stay and sell. What if oai loses the lead? What if open source wins? Keeping the stock seems like the actual hard thing to me and I expect to see many others leave (like early googlers or Facebook employees)
Sure it's worth more if you hang on to it, but many think "how many hundreds of M's do I actually need? Better to derisk and sell"
Big tech hires mostly off leetcodes. There are other factors too.
Your journey begins with practicing leetcodes and reading all the books on teachyourselfcs.com
Study dilligently. Watch YouTube courses too but this should be considered supplemental.
Apply for big tech jobs. This will probably take several trys, especially in this market. Just keep applying and studying.
Once you're at a big tech company keep studying, learn from the smartest people you meet, and ship a lot. After a promotion or two apply to Meta (or someone else if they're paying more on levels.fyi at that point, but Meta pays especially well)
Start giving presentations at conferences. At higher levels in big tech this is encouraged and sometimes even expected as part of promo packets. Also practice your writing. Starting on writing a technical book on a subject your an expert in by this time would be good.
Keep shipping, keep learning, keep getting promoted. Once you're on that track you're well on your way to $1M/yr TC.
I would suggest following / reading people who talk about using Claude 3.5 sonnet.
Lots of people developing whole apps using 3.5 sonnet and sometimes cursor or another editor integration. The models are getting quite good now at writing code once you learn how to use them right and don't use the incorrect LLMs (a problem I often see in places other than twitter unfortunately.) They seem to get better almost weekly now too. Just yesterday Anthropic released an update where you can now store your entire codebase to call as part of the prompt at 90% token discount. Should make an already very good model much better.
Gumroad's CEO has also made some good YouTube content describing a lot of these techniques, but they're livestreams so there is a lot of dead air.