[echelon]

AI is a million times better than Google search. I don't see how it doesn't replace Google search in a few years.

AI code completion is god mode. While I seldom prompt for new code, AI code autocompletion during refactoring is 1000x faster than plumbing fields manually. I can do extremely complicated and big refactors with ease, and that's coming from someone who made big use of static typing, IDEs, and AST-based refactoring. It's legitimately faster than thought.

And finally, it's really nice to ask about new APIs or pose questions you would normally pour over docs or Google and find answers on Stack Overflow. It's so much better and faster.

We're watching the world change in the biggest way since smartphones and the internet.

AI isn't a crappy consultant. It's an expansion of the mind.

[bayindirh]

AI is just a weighted graph which stands on shoulders of a million giants. However, it can't cite, can't fact check, doesn't know when it's hallucinating, and its creators doesn't respect any of the work which they need to feed to that graph to make it to fake its all knowledgeable accent.

Tech is useful, how it's built is very unethical, and how it's worshiped is sad.

[echelon]

> However, it can't cite, can't fact check, doesn't know when it's hallucinating

Maybe some folks need this, but the way I use this tech doesn't rely upon that so much. By the time results start appearing, my brain is already fast at work processing the output to qualify whether the information the LLMs return is accurate, whether it's a good leaping off point, whether I can keep drilling deeper, expand my prompt scope, etc.

I'm using it as search. Just as old search had garbage results we had to filter out, so do LLMs. But this tool is a way more advanced query language than Google ever supported. These tools are like "Google 9000".

It feels like I'm plugged into the Matrix rather than getting SEO'd garbage. I know the results have issues, but that doesn't matter - I can quickly draw together the pieces and navigate around it. Compared to Google, it feels like piloting a star ship.

[bayindirh]

> By the time results start appearing, my brain is already fast at work processing the output to qualify whether the information the LLMs return is accurate, whether it's a good leaping off point, whether I can keep drilling deeper, expand my prompt scope, etc.

Seems unnecessarily tiring. Instead I use a SEO spam and ad-free search engine. It's called Kagi. It allows me to further refine my search via lenses and site prioritization. Also, it has zero hallucination chance, because it's a deterministic search engine.

> It feels like I'm plugged into the Matrix rather than getting SEO'd garbage. I know the results have issues, but that doesn't matter - I can quickly draw together the pieces and navigate around it. Compared to Google, it feels like piloting a star ship.

Same for Kagi, without selling my data or trawling information obtained without consent or disregard of ethics, and many other things.

Note: I don't use any of the Kagi's AI features, incl. proofreading.

[danielbln]

Modern LLM offerings can use tools, including search, and that (like most good RAG) enables citation and fact checking. If you use LLMs like it's late 2022 and you just opened ChatGPT, then that's not indicative on how you should be using LLMs today.

[bayindirh]

I think I cleared my point about tools in another reply I wrote somewhere close [0].

As I said, the network doesn't carry citation/source information. IOW, when it doesn't use a tool, it can't know where it ingested that particular piece of information.

This is a big no no, and it's the same reason they hallucinate and they'll continue doing that.

As a tangent, I see AI agents hit my digital garden for technical notes, and I'll probably add Anubis in front of that link in short order.

[0]: https://news.ycombinator.com/item?id=43972807

[danielbln]

I don't understand the point though. If you take tools away from me then I'm not particularly reliable or productive either.

[bayindirh]

But you are aware that you may not be reliable or reduced in capacity. An LLM without tools doesn't know or can't know this.

[jaoane]

I think you are a bit outdated, since state of the art AIs can cite and fact check just fine.

[bayindirh]

Give them tools, maybe, but "The network" can't do it natively. They doesn't have an understanding, can't filter out "ad absurdum". Plus they can only go so deep. Maybe they can hit snopes and check something, but I don't believe current ones even with tools do detailed cross examination with open ended research and stop when they are convinced that they have enough data.

[jaoane]

That's like saying that humans can't tighten screws because they need screwdrivers.

[bayindirh]

Nope. This is akin having no long term memory and don't able to tell when and where you learnt that water boils at 100 degrees C first, and it in fact burns very badly.

[ninetyninenine]

You’re still in denial and possibly behind. AI cites stuff all the time and has become agentic.

On the opposite end of the spectrum of worshippers there are naysayers and deniers. It’s easy to see why there are delusional people at both ends of the spectrum.

The reason is that the promise of AI both heralds an amazing future of machines and a horrible future where machines surpass humanity.

[bayindirh]

I don't think I'm a denier, or being behind [0]. I know about tools and agents.

For the third time [1] [2], I'll divide the line between core network and tools that core network uses. Agents are nothing new, and they expand capabilities of the LLMs, yes that's true. But they still can't answer the question "how did you generate this code and which source repositories you did use" when the LLM didn't use any tools.

The core network doesn't store citation/source information. It's not designed and trained in a way to do that.

geez.

[0]: https://notes.bayindirh.io/notes/Lists/Discussions+about+Art...

[1]: https://news.ycombinator.com/item?id=43972807

[2]: https://news.ycombinator.com/item?id=43972892

[ninetyninenine]

Agents are nothing new? They’ve only been around for a couple years. The rustiness is expected.

Second the question you brought up can’t be answered even by a human. It’s a stupid question right? You blindfold a human and prevent him from using any tools and then ask him what tools he used? What do you expect will happen. Either the human will lie to you about what he did or tell you what he didn’t do. No different from an LLM.

The core network doesn’t store anything except generalization curve. Similar to your brain. You didn’t store those references in your brain right? You looked that shit up. The agentic LLM will do the same and the UI literally tells you it’s doing a search across websites.

Geeze.

[bayindirh]

So, I took your word, assumed I knew nothing about Agents and Agentic AI, and started digging. Wikipedia states the following for Agentic AI:

> "Agentic AI is a class of artificial intelligence that focuses on autonomous systems that can make decisions and perform tasks without human intervention."

I can work with that. So we have agents that autonomously react to their environment, changes, or what we can say impulses. They sit there and do what they are designed to do, and do that autonomously. Makes sense. However, this sounds a bit familiar to me. Probably me hallucinating something, so let's dig deeper. There seems to be an important distinction, though:

> "Agentic AI operates independently, making decisions through continuous learning and analysis of external data and complex data sets."

So, we need to be able to learn, evolve, and analyze external and complex data sets. That's plausible, but my hunch is still lingering there, tingling a bit stronger. At this point, for Agentic AI, we need an independent "thing" which can decide, act, learn, and access external data sources to analyze and learn from them. In short, I need to be able to give this Agentic AI a goal, and it accomplishes it automatically with the things at its disposal. Fair enough.

We were discussing (software) agents and their history. So let's pivot more to agents. Again, turning to Wikipedia, we find this sentence:

> "In computer science, a software agent is a computer program that acts for a user or another program in a relationship of agency."

Again, a piece of software that acts for a user or another program. Hmm... They have five basic attributes: 1)are not strictly invoked for a task, but activate themselves, 2)may reside in wait status on a host, perceiving context, 3) may get to run status on a host upon starting conditions, 4)do not require interaction of user, 5)may invoke other tasks including communication. That hunch, though. It feels more like mild kicking. Where do I know these concepts? Somewhere from the past? Nah, I'm hallucinating. You told me that they are new.

As I skim the article and pass "Intelligent Agents" past, I see something very familiar line under "Notions and frameworks for agents" title: "Java Agent Development Framework (JADE)". I know this. Now I remember!

I have used this framework to code a platform where an agent gets orders from a client for a set of items, and submits them to another agent, where other agents send their best prices, and another agent calculates the best combination for the cheapest price. Doing a "combinatorial reverse auction" for a set of items. We had no time to implement feedback-based price adjustment strategies, but the feedback and announcement code were there, so every agent knew how the transaction went. They all were autonomous. A single agent acted on behalf of the user, and the whole platform responded to that without any humans at any step, including final decisions!

That was my Master's thesis. I have also presented it at the IEEE Symposium on Intelligent Agents, IA in Orlando in 2014 [0]!

When did I complete my Master's thesis?

Oh. 2010. 15 years ago.

Alright. This solves it.

Now, on to your second question. Let's put it right here:

> You blindfold a human and prevent him from using any tools and then ask him what tools he used? What do you expect will happen. Either the human will lie to you about what he did or tell you what he didn't do. No different from an LLM.

You're mangling my question here. The question I ask is different:

> Generate me a Python code for solving problem X, then tell me which source repositories you used to generate this code. Cite their licenses, if possible.

All of this information is in the core network for the first part of the problem. LLMs without tool capabilities can generate code, and generate it well. The source of this knowledge came from their training set, which consists of at least "The Stack", and some other data sources on top of that. So, the LLM can generate the code without any tools, but it can't know where the source came from. It's just there, in the core network.

You think the question is stupid, but it's not. This is where all the ethical questions regarding LLM training are rooted. LLMs hallucinate licenses, don't know where the code came from, and whatnot. If you ask me about a code piece in my source code, I can give you the source, the thought process, and design, citing the originality or how I found it elsewhere and got into my codebase. Lying about it would be a big problem in the light of licenses, but LLMs get scot-free because they're just fair using it. Humans can't do the same thing, why LLMs? Because their owners have money and influence? Seems so.

> You didn't store those references in your brain right? You looked that shit up.

No, when I looked that shit up, I recorded where I read it alongside other contexts, including the weather that day in some particular cases. I don't answer "I just know, I don't know how" when people ask me about the source of my knowledge.

This is the difference between humans and LLMs; this thin line is very important.

[0]: https://ieeexplore.ieee.org/abstract/document/7009456

[ninetyninenine]

Nobody is talking about your masters thesis which nobody gives a shit about.

We are talking about agentic LLMs which have been around for about a year only. Not some bs pre LLM ai chatbot or some useless thing like that? Are you autistic? No joke and more respect to you if you are but to a non autistic person I am obviously talking about AI which in modern contexts means LLMs and agentic AI obviously means agentic LLMs

Once you started getting into your masters thesis I stopped reading. Conversation is over. Good day.

[skydhash]

> AI code autocompletion during refactoring is 1000x faster than plumbing fields manually. I can do extremely complicated and big refactors with ease, and that's coming from someone who made big use of static typing, IDEs, and AST-based refactoring. It's legitimately faster than thought.

Unless you know Vim!

[bayindirh]

> Unless you know Vim!

or the IDE (or text editor for that matter) well. People don't want to spend time understanding, appreciating and learning the tool they use, and call them useless...

[layer8]

That's funny, because I have little patience with having to spend time finding out how to coax the AI into doing what I effing want, and much prefer the reliable and deterministic IDE features.

[bayindirh]

It's funny and sad, and I'm on the same page with you. I use some tools close to two decades and can do things people can't fathom in no time, for a very long time.

I don't bend the tool, even. It's what it's designed to do.

[codechicago277]

This is true. Just like a crappy consultant, AI lets you offload the repetitive, monotonous work so that you can focus your time on the big architectural problems. Of course you can write a better function if you spend a lot of time on it, but there’s magic in just letting the AI write the off the shelf version and move on.

[skydhash]

Where are these repetitive, monotonous work so I can se send a job application there.

Even on a greenfield project, I rarely spend more than a day setting up the scaffolding and that’s for something I’ve not touched before. And for refactoring and tests, this is where Vim/Emacs comes in.

[cess11]

Many people doing code for money never or only rarely used any form of code generation until it was given to them as a SaaS in exchange for copies of the code they work on.

I've been surprised by this for a long time, having seen coworkers spend days typing in things manually that they could have put there with IDE capabilities, search-replace, find -exec or a five minute script.

[echelon]

> And for refactoring and tests, this is where Vim/Emacs comes in.

I've used Vim bindings and strongly typed languages with IDEs that have strong AST-based refactoring my entire career.

Nothing comes close to changing one condition of a test and having the AI autocomplete magically suggest the correct series of ten updates that fix the test. In under the blink of an eye, too.

Everything is truly changing in big ways.

[skydhash]

> Nothing comes close to changing one condition of a test and having the AI autocomplete magically suggest the correct series of ten updates that fix the test. In under the blink of an eye, too.

But why are you doing this? Granted, you may have a longer career than I do, but I never once think: The test condition is wrong, let's update it. Oh, I wish I could update the code alongside it!.

[player1234]

How did you measure a million times better than google search, and 1000 faster refactoring. Can you please share your bench marks and methodology?

[tim333]

Google sees that and is trying to incorporate AI.

[echelon]

They have to defend 100% of it to just stay in place.