andreybaskov

As someone who built a niche SaaS I can see both sides of this.

If you are looking to replicate the exact same feature set of an existing product - buying would almost always be a better choice.

But it's rare for avg customer needs to perfectly match product features. Most often they need 20-40% of the product + some custom, business specific logic that's missing and is later fixed with spreadsheets/integrations. In that case it depends on how mission critical this software is.

I'd say investing into the core software that runs your business might very well be worth the effort, even if it's 10x between build vs buy.

Curious, how did you deal with inspections and building codes? Or is it in a county with no building codes?

Finally, now I know I'm not the only one! These sharp edges constantly cut into my wrists to the point I was thinking of doing the same, or glueing some kind of kind soft padding to the edges. Great someone did it. I wonder how far can you cut them?

Sadly, memory bandwidth is abysmal compared to Apple chips - 273 GB/s vs 614 GB/s on M5 Max for similar price. Even though fp4 compute is faster, it doesn't help for all the decode heavy agentic workflows.

Reading this was a roller coaster for me.

Because of a bad habit reading comments before the link I knew it was AI. I read it regardless, and... I still enjoyed it!

I'm very much not a writer or a critic, so my definition of good writing is likely very low. Yet I can't shake off this weird feeling that I truly enjoyed the writing and felt the emotions, _while_ knowing it's LLM.

I'm guessing that human after touch is what made it pleasant to read. I'd love to see the commit history of the process. Fun times we live in!

Just wanted to say that I've felt the same about the images. To me it's likely was the text that for some reason had AI-feel to it. Great story though, I was in awe learning it was AI generated.

In basic scenarios agree it’s possible. I’ve similarly toyed with building a small language and LLM picks it up. But that’s an additional context and it’s never felt like it worked “out of the box” vs say Typescript. Just saying that it’s an adoption barrier established languages don’t have.

Synthetic codebases are certainly an option, but if top models remain closed I don’t see them building datasets for every new language.

That's cool! Do you have any info on how manufacturing is done in the US?

It's a legitimate question to ask about any new language post AI - given there is no training dataset, any other language would work better with AI.

The bigger problem is maintainability over the long term, Deno is built by Node.js creator and is maintained for half a decade now, that's hard to compete with. In a way it's much more about social trust rather than particular syntax.

Great summary of the current state of AI coding, without all the hype.

I wonder though if we are fundamentally limited by LLMs architecture to come up with some novel architecture. Or is it just a limitation of current prompting and tools?

E.g. could your brute force the problem by asking "come up with 100 innovative, not yet tried ideas in compiler architecture" and run 100s of this experiments.

I feel like the answer to that is basically that "spark" that we have as human intelligence, but also... sometimes you get it by trying and failing many many times.

I'm testing ground heat rejection system that dumps heat into the ground. It needs a lot of space, but you need it for panels anyways.

Solar-powered data center in a desert.

Fully off-grid using solar, batteries and Starlink for uplink. Focusing on AI inference at the beginning. Currently building our first prototype and testing cooling solutions.

https://solarcube.com

Original Apple II manuals written by Chris Espinosa and Jef Raskin are a treat to read. Would highly recommend, just to get a sense of what it was like to get onboarded on Apple II back in the day.

And then obviously Programming the 6502 by Rodnay Zaks.

Out of all things I'm actually surprised they went straight to custom silicon, but gotta respect that decision. It's likely the only way to compete with Tesla right now.

Thank you for saying this. It's almost like others are saying we should stop trying things because they are hard and challenging.

I wish we could dream a bit bigger rather than coming up with reasons something will fail.

I don't think we need that for 1,000x. We can building more solar, nuclear and there is still room for at least 10x improvement in efficiency for the chips. We are far far away from maxing out our compute capability as civilization before we start shooting satellites into the sun.

I agree scaling alone is not enough, and transformers itself is a proof of that - it was an iteration on the attention mechanism and a few other changes.

But no matter what the next big thing is, I'm sure it would immediately fill all available compute to maximize its potential. It's not like intelligence has a ceiling beyond which you don't need more intelligence.

What high quality data sources are not already tapped? Synthetic data? Video?

Where does the next 1000x flops come from? Even with Moore's law dead, we can easily build 1,000x more computers. And for arguments about lack of power - we have sun.

I see LLMs in a similar way - a new UI paradigm that "clicks the right buttons" when you know what you need, but don't know exact names of the buttons to click.

And from my experience there are lots and lots of jobs that are just "clicking the right buttons".

Say we discover a new architecture breakthrough like Yann LeCun's proposed JEPA. Won't scaling laws apply to it anyway?

Suppose training is so efficient that you can train state of the art AGI on a few GPUs. If it's better than current LLMs, there will be more demand/inference, which will require more GPUs and we are back at the same "add more gpus".

I find it hard to believe that we, as a humanity, will hit the wall of "we don't need more compute", no matter what the algorithms are.

Thanks! That's a great way to analyze it by comparing to open source models. Though I wonder if they use the same hardware for gpt-oss-120b and Claude Opus.

Does anyone know or have a guess on the size of this latest thinking models and what hardware they use to run inference? As in how much memory and what quantization it uses and if it's "theoretically" possible to run it on something like Mac Studio M3 Ultra with 512GB RAM. Just curious from theoretical perspective.

Exactly. It's software - `curl` or LLM. It's a function that accepts input and produces output. One is much more sophisticated that the other, but it's made out of the same machine instructions, there is no magic.

What's the criteria that makes one function "autonomous" and the other one "manual"? I feel it really boils down to this.

Apologies, I didn't mean to spam, haven't seen the other thread that picked up more votes and was really curious about where the line is.

I completely understand private businesses having a right to refuse a service without a cause. But as others pointed, the question is to what degree "delegation" is acceptable if I'm acting in a good faith?

I'm guessing the answer is "to a degree it doesn't impact our business".

I know its there. But we _prefer_ to have a single account to simplify tracking and picking up packages. I'm curious if from their point of view (or their ToS) I'm even allowed to share my credentials with anyone else.

1Password shared vaults are there for a reason - people share credentials all the time, business or personal.

Interesting. Amazon ToS actually has a section about agents - https://www.amazon.com/gp/help/customer/display.html?nodeId=...

And they even provide a definition of what an Agent is:

"Agent” means any software or service that takes autonomous or semi-autonomous action on behalf of, or at the instruction of, any person or entity.

Though to me it raises even more questions. What is a software that takes "autonomous" action on my behalf. Is curl "autonomous"?

Me and my wife share the same Amazon account. Should I open a business account to do grocery shopping?

Originally posted this in another thread, but very curious what others think.

Can I ask my partner to buy a product on Amazon?

Can I ask my personal assistant to buy a product on Amazon?

Can I hire a contractor to buy products on Amazon?

Can I communicate with a contractor via API to direct them what products to buy?

What if there is no human on the other end and its an LLM?

Same issue with LinkedIn. I know execs who have assistants running their socials. Is this legal?

Like, where do we draw the line? In the future, would the only way to shop on Amazon be with approved VR goggles that scan your retina to verify you are a human?

Agentic browsers raise a lot of questions that were hanging there even before LLMs.

Can I ask my partner to buy a product on Amazon? Can I ask my personal assistant to buy a product on Amazon? Can I hire a contractor to buy products on Amazon? Can I communicate with a contractor via API to direct them what products to buy? What if there is no human on the other end and its an LLM?

Same issue with LinkedIn. I know execs who have assistants running their socials. Is this legal?