f0e4c2f7

Lots of services have no desire to ever expose an API. This approach lets you step right over that.

If an API is exposed you can just have the LLM write something against that.

Ha! This is great. I've been waiting for someone to make this.

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!

I don't like the premise of this game. If you're autistic, don't mask. Live authentically as yourself and find people who love you for who you are.

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.

This looks cool and I wish you luck but I'd probably never use something closed source for this.

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.

Front page of HN. Funny to imagine thousands of people sitting in the office crossing their eyes at their computer screen right now.

There is a funny parallel I see with Kubernetes that I also saw a lot with Linux in the early years. There are thousands of packages and tools you can install on Linux (think phpmyadmin for example) and new users sometimes go wild installing every single package they read about.

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.

Sounds like management and mentoring might be a satisfying diversion at this point in your life.

For sure. It's been amazing honestly and I feel like it has really accelerated my learning. Enough that I'm trying to get really serious about making the most out of this new leverage seemingly out of nowhere.

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.

Building using LFS has been on my list of "I really should probably do that just to learn" for about 20 years now. I'll get around to it! This year I'm finally learning Lisp (and really enjoying it).

Really cool idea. Novel, I imagine a lot of people would play this. I would suggest a social media campaign with short form videos. Seems like something that has the potential to go viral or at least make a lot of money.

There is one clear answer in my opinion:

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"

I'm a big fan of the reading list and structure of teachyourselfcs.com

Fortunately big tech jobs are available in EU too. Highest salaries might trickier without moving to US, shouldn't be too hard to transfer to US once you get a job at one if you want to. Making quite a high salary without moving is possible too (though still need to be a in a major metro that has big tech companies.)

This is fairly straightforward imo. Most money is in big tech. Look at levels.fyi to get an idea of the roadmap.

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'm not sure why but it seems like most of the high quality AI content is on twitter. On average seems to be around ~4 months ahead of HN on AI dev approaches.

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.

https://www.youtube.com/watch?v=1CC88QGQiEA

https://www.youtube.com/watch?v=mY6oV7tZUi0

There have been some papers showing that RLHF makes models more palletable to use but reduces performance on evals and in other various ways.

I couldn't find the one I was looking for but this is one of them.

https://arxiv.org/abs/2310.06452

Edit:

This tweet also has a screenshot showing degraded evals from RLHF from base model.

https://x.com/KevinAFischer/status/1638706111443513346?t=0wK...

This is a bit of a Zen thing you're describing. A lot of different approaches and theories on getting to the same thing here I would say. It's hard to think of specific advice here beyond "just start doing it" which does not seem very helpful. Maybe a good insight here is "practice will make it easier."

I think ultimately for me this mindset was cultivated at a pretty young age with some writing and art I happened to come across. I love content like that and seek it out now. I think you can become more and more growth mindset oriented with time. I'll share some of the things I've liked on the topic here, maybe that will be useful:

http://www.catb.org/~esr/faqs/hacker-howto.html

https://www.goodreads.com/quotes/7727986-mountains-should-be... (this is a quote, but recommend this whole book)

https://www.amazon.com/Mindset-Psychology-Carol-S-Dweck/dp/0...

https://youtu.be/etEJrznE-c0?si=Eaq8aycm8Yxn1v56

This book is so good. It's about building early computers but it feels just like tech does today in the way it's described. Which make it feel like this bigger context you're reading into. That even before computers were mainstream, there were still those who tinkered.

Two things are especially memorable to me. One is a casual remark in the book that they found the best way to get things done is to pair someone very experienced and cynical with someone very inexperienced and naive. Combined they would get lots done together compared to either alone. I think this is still true today.

The other thing is the intro. It's about the head of the project getting a group together and renting a sailboat on vacation. On the sailboat the get tossed and at times feel like they barely survived and it ends with someone saying "if this was his vacation...what did this man do for fun!?"

Currently listening to this series of interviews about Neuralink:

https://www.youtube.com/watch?v=Kbk9BiPhm7o

Would be interested to hear the experiences of someone who has used this.

When I was a kid there was this new thing that came out called Wikipedia. I couldn't convince anyone it was useful though because they pointed out it was wrong sometimes. Eventually they came around though.

AI is like that right now. It's only right sometimes. You need to use judgement. Still useful though.

This is an extremely good insight and piece of advice. Already we're starting to see what looks very similar to a "frontend" and "backend" kind of development model similar to the previous tech generation.

While there are advantages to going "full-stack" in this analogy most people focus on one or another

Is there a way to try this yet?

Lately I've been experimenting with using TDD approaches for LLM assisted coding. There is a little bit on YouTube and some research papers on this topic. But most of it relates to how well you understand the test suites for the given language / framework to ask for the tests to be designed in the right way.

This has been great in gpt-4o for python because it has a built in interpreter. So I can describe the way I want the API to look, ask it to write tests in Python, write the code to satisfy those tests, and then run the unit tests in it's own local python env to see if the code works.

This streamlines things a lot. I find I can get working artifacts from gpt this way when other I often have to have a bit of a back and forth where I'm the one testing and pasting in errors.

For gpt-4o this only works with python though (for now).

It might be interesting to build a platform like this, even in your local laptop that allows the LLM to shell out to a sandbox for something like React with the unit tests configured so the LLM can run them and get the output.

Interestingly, frameworks like react also have test suites around ui/ux concepts. I haven't tried it, but I think it would be interesting to use that same approach with those test frameworks to attempt to get better looking frontends (something the LLMs are not especially good at out of the box)

Exa is trying to do this. I've found some sort of interesting stuff this way but it honestly doesn't feel quite good enough yet to me.

https://exa.ai/search?c=all

Does this still work?

If you follow someone you're still more likely to see their posts. You just see other stuff too in the "For You". It's algorithmic, and at times annoying - but contrary to the article is still very valuable in my experience.

I would say twitter stays about 1.5 months ahead of every other site I know of on AI news in particular. I've usually had time to demo new tools for quite a while before hearing about them on HN or elsewhere.

Architect the program yourself in your head. Then use gpt to write one function at a time. For obscure libraries, paste in documents directly to the context window and give links to example GitHub repos. Paste errors in to troubleshoot as needed. Architect for many small libraries in their own files so it's easier to compartmentalize and let gpt work with smaller blocks of code.

This study uses a version of ChatGPT that is either 1 or 2 versions behind depending on the part of the study.

It cracks me up how consistent this is.

See post criticizing LLMs. Check if they're on the latest version (which is now free to boot!!).

Nope. Seemingly...never. To be fair, this is probably just an old study from before 4o came out. Even still. It's just not relevant anymore.

Lots of good tools for this.

One way to do it completely free is OBS + Kdenlive. The interface for both leaves something to be desired but both open source and have all the features you would want (though sometimes buried in menus)