But note that the price they can charge is based on market supply and demand. If Claude is priced at $20, ChatGPT won’t be able to ask $100.

If OpenAI sent out an email today informing me that to maintain access to the current 4o model I will have to pay $1000 a year, and that it would go up another $500 next year... it would still be well worth it to me.

To you maybe. But if Claude or any other competitor with similar features and performance keeps a lower price, most users will migrate there.

You could be right, but I suspect that you're underestimating the degree to which GPT has become the Kleenex of the LLM space in the consumer zeitgeist.

Based on all of the behaviour psychology books I've read, Claude would have to introduce a model that is 10x better and 10x cheaper - or something so radically different that it registers as an entirely new thing - for it to hit the radar outside of the tech world.

I encourage you to sample the folks in your life that don't work in tech. See if any of them have ever even heard of Claude.

I don’t think people outside of tech hearing about OpenAI more than Claude is really indicative of much. Ask those same people how much they use an LLM and it’s often rare-to-never.

Also, in what way has OpenAI become the Kleenex of the LLM space? Anthropic, Google, Facebook have no gpts, nobody “gpts” something, nobody uses that companies “gpt”.

I would say perhaps OpenAI has become the Napster, MySpace, or Facebook of the LLM space. Time will tell how long they keep that title

I am surrounded by people who "ask GPT" things constantly.

That seems like the same thing to me.

Are these people asking chatGPT though? Or do they say “ask GPT” and then use other LLMs?

I feel like I hear “ask Claude” as much as “ask chatGPT”

Yeah... I think it's GPT.

I just sampled my family and nobody could name an LLM that wasn't GPT. In this very small, obviously anecdotal scenario, GPT == LLM.

They seemed vaguely aware that there are other options; my wife asked if I meant Bing.

Meanwhile, I have literally never heard the words "ask Claude" out loud. I promise to come back and confirm if/when that ever changes.

Would you be willing to share how you're using it?

I keep hearing from people who find these enormous benefits from LLMs. I've been liking them as a search engine (especially finding things buried in bad documentation), but can't seem to find the life-changing part.

1. Search engine replacement. I'm using it for many queries I asked Google before. I still use Google, but less often.

2. To break procrastination loops. For example I often can't name a particular variable, because I can see few alternatives and I don't like all of them. Nowadays I just ask ChatGPT and often proceed with his opinion.

3. Navigating less known technologies. For example my Python knowledge is limited and I don't really use it often, so I don't want to spend time to better learn it. ChatGPT is just perfect for that kind of tasks, because I know what I want to get, I just miss some syntax nuances and I can quickly check the result. Another example is jq, it's very useful tool, but its syntax is arcane and I can't remember it even after years of occasional tinkering with it. ChatGPT builds jq programs like a super-human, I just show example JSON and what I want to get.

4. Not ChatGPT, but I think Copilot is based on GPT4, and I use Copilot very often as a smart autocomplete. I didn't really adopt it as a code writing tool, I'm very strict at code I produce, but it still helps a lot with repetitive fragments. Things I had to spend 10-20 minutes before, construction regexps or using editor macroses, I can now do with Copilot in 10-20 seconds. For languages like Golang where I must write `if err != nil` after every line, it also helps not to become crazy.

May be I didn't formulate my thoughts properly. It's not anything irreplaceable and I didn't become 10x programmer. But those tools are very nice and absolutely worth every penny I paid for it. It's like Intellij Idea. I can write Java in notepad.exe, but I'm happy to pay $100/year to Jetbrains and write Java in Idea.

>> 3. Navigating less known technologies. For example my Python knowledge is limited and I don't really use it often, so I don't want to spend time to better learn it.

Respectfully but that's a bit like saying you don't need to learn how to ride a bicycle because you can use a pair of safety wheels. It's an excuse to not learn something, to keep yourself less knowledgeable and skillful than you could really be. Why stunt yourself? Knowledge is power.

See it this way: anyone can use ChatGPT but not everyone knows Python well, so you 'll never be able to use ChatGPT to compete with someone who knows Python well. You + limited knowledge of Python + ChatGPT << someone + good knowledge of Python + ChatGPT.

Experiences like the one you relay in your comment makes me think using LLMs for coding in particular is betting on short-term gains for a disproportionately large long-term cost. You can move faster now, but there's a limit to what you can do that way and you'll never be able to escape it.

> and you'll never be able to escape it.

Slow down, cowboy. Getting a LLM to generate code for you that is immediately useful and doesn't require you to think too hard about it can stunt learning, sure, but even just reading it and slowly getting familiar with how the code works and how it relates to your original task is helpful.

I learned programming by looking at examples of code that did similar things to what I wanted, re-typing it, and modifying it a bit to suit my needs. From that point of view it's not that different.

I've seen a couple of cases first hand of people with no prior experience with programming learn a bit by asking ChatGPT to automate some web scraping tasks or spreadsheet manipulation.

> You + limited knowledge of Python + ChatGPT << someone + good knowledge of Python + ChatGPT.

Substract ChatGPT from both sides and you have a rather obvious statement.

> Respectfully but that's a bit like saying you don't need to learn how to ride a bicycle because you can use a pair of safety wheels.

How did you learn to ride a bicycle?

Respectfully, this is an incredibly weak take.

You are making a lot of assumptions about someone's ability to learn with AND without assistance, while also making rather sci-fi leaps about our brain somehow being able to differentiate between learning that has somehow been tainted by the tendrils of ML overlords.

The models and the user interface around them absolutely will continue to improve far faster than any one person's ability to obtain subject mastery in a field.

Just to clarify, I didn't say anything about the OP's "ability to learn". I know nothing about that and can't tell anything about it from their comment. I also didn't say anything about how our brain works, or about "the tendrils of ML overlords".

If you want to have a debate, I'm all for it, but if you're going to go around imagining things that I may have said in another timeline then I don't see what's the point of that.

Your comment said, in many more words, that to embrace LLMs for learning or knowledge enhancement might pay off in the short term but will leave you stunted and operating in a local maximum of potential and possibility.

You could be right, but I strongly suspect that you are actually wrong.

Nope. I said that refusing to learn something will leave you stunted. I did not say that using an LLM will leave you stunted.

The OP said:

>>> For example my Python knowledge is limited and I don't really use it often, so I don't want to spend time to better learn it.

And I replied:

>> Respectfully but that's a bit like saying you don't need to learn how to ride a bicycle because you can use a pair of safety wheels. It's an excuse to not learn something, to keep yourself less knowledgeable and skillful than you could really be. Why stunt yourself? Knowledge is power.

"It's an excuse to not learn something (...) Why stunt yourself?".

> Respectfully but that's a bit like saying you don't need to learn how to ride a bicycle because you can use a pair of safety wheels.

Do you care how a car engine works to drive a car?

Yes. I generally like to know how stuff works and why.

Do you think rest of the world cares?

These who don’t care need help for every little problem. And possibly get charged a lot more than it actually cost in a car shop for any service. So while one can live without knowing it’s ALWAYS better to know

Sorry, is this going somewhere?

You did not answer the question

I actually disagree with this. One of my few really helpful LLM experiences is working with technologies that I rarely touch. nginx configs are probably a good example, or a library I haven't touched in 5 years.

The AI is pretty good at answering "I know this is possible, remind me what it's called or what the syntax is."

For a lot of these the reason I want to look it up is I don't use it enough to build muscle memory. If I did, I'd already have the muscle memory. And in practice, it's just a faster version of looking for examples on stackoverflow.

I'm long time programmer and I learned quite a lot of languages along the way. 20 years ago I programmed Perl and I spent some time learning its quirks. While some of that knowledge definitely helped me all these years, especially regular expressions, I wouldn't write a single Perl line without Googling today.

So learning syntax nuances of yet another programming language which I'll forget soon is wasted time. There's some "fundamental" knowledge, like understanding what functional value is, but the specific syntax that Python invented for lambda - that's not something valuable when I use Python few times a year.

I agree that learning syntactic nuances is not that useful, but it's something that comes naturally from contact with the language.

Rereading your comment I think I got the wrong impression from it. You said:

>> I don't really use it often, so I don't want to spend time to better learn it

And I focused more on "I don't want to spend time to better learn it" than on "I don't really use it often". I also don't use Python often and I have to look up its syntax, same with e.g. R. I guess in that case it makes sense to not spend too much time learning the ins and outs of a language. On the other hand, Python is a staple, like Java was back in the day, and Perl and so on. It's useful, like speaking English. I don't think I would ever feel comfortable saying "I don't want to learn" it; or anything else.

I’ve got a local llama 3.2 3B running on my macOS. I can query it for recipes, autocomplete obvious code (this is the only thing GitHub Copilot was useful for). And answer simple questions when provided with a little bit of context.

All with much lower latency than an HTTP request to a random place, knowing that my data can’t be used to trading anything, and it’s free.

It’s absolutely insane this is the real world now.

What do you use to interface with llama for autocomplete? and what editor do you use?

Not wanting my data to be sent to random places is what has limited my use of tools like copilot (so I'd only use it very sparingly after thinking if sending the data would be a breach of nda or not)

ollama + VSCode + Continue extension

+1 for sure running LLMs 3.2 3B is super fast and useful. I have been pushing it for local RAG and code completion also. I bought a 32B memory Mac six months ago, which I now regret because the small local models are now extremely useful and run fine on old 8B memory Macs, and support all the fun experiments I want to do.

I can't even summarize how much GPT4x has helped me teach myself engineering skills over the past two years. It can help me accomplish highly specific and nuanced things in CAD, plan interactions between component parts (which it helped me decide) on PCBs that it helps me to lay out, figure out how to optimize everything from switching regulators to preparing for EM certification.

And I could say this about just about every domain of my life. I've trained myself to ask it about everything that poses a question or a challenge, from creating recipes to caring for my indoor Japanese maple tree to preparing for difficult conversations and negotiations.

The idea of "just" using it to compose emails or search for things seems frustrating to me, even reading about it. It's actually very hard for me to capture all of this in a way that doesn't sound like I'm insulting the folks who aren't there yet.

I'm not blindly accepting everything it says. I am highly technical and I think competent enough to understand when I need to push back against obvious or likely hallucinations. I would never hand its plans to a contractor and say "build this". It's more like having an extra, incredibly intuitive person who just happens to contain the sum of most human knowledge at the table, for $20 a month.

I honestly don't understand how the folks reading HN don't intuitively and passionately lean into that. It's a freaking superpower.

> I honestly don't understand how the folks reading HN don't intuitively and passionately lean into that. It's a freaking superpower.

It is difficult to get a man to understand something when his salary depends upon his not understanding it.

Many of us here would see our jobs eliminated by a sufficiently powerful AI, perhaps some have already experienced it. You might as well. If you use AI so much, what value do you really provide and how much longer before the AI can surpass you at that?

Only time will tell. In the meantime, I am not losing any sleep.

There's a lot of people in technical roles who chose to study programming and work at tech companies because it seemed like it would pay more than other roles. My own calculation is that the tech-but-could-have-just-as-easily-been-a-lawyer cohort will be the first to find themselves replaced. Is that a revealing a bias? Absolutely.

Actual hackers, in the true sense, will have no trouble finding ways to continue to be useful and hopefully well compensated.

Maybe because my salary doesn't depend on any of this that I think this whole framing is science fiction.

A calculator needs direction on what to calculate because the machine doesn't have agency.

We are going to a higher level of abstraction and at that higher level you probably need a different skill set than before. Humans aren't going to get cut out of the loop but you always have to level up your skills.

I think there is just a generation of tech workers that got lazy and thought they were immune to change some how. I didn't even know anyone that had been on the internet when I graduated high school. Someone older and wiser though told me I would have to constantly reinvent myself, always learn new skills and will have several different careers. They were so spot on. Guess what? Nothing has changed.

If we ever find each other IRL, I suspect that we will be fast friends.

A tip of my hat to you, sir/ma'am.

I suspect many people are simply not that intellectually curious about the world.

It a super power for the intellectually curious while the highly specialized that thought they were done learning can only see the threat to their intellectual moat. They aren't wrong.

I feel like I level up on a weekly basis practically.

YESssss

I feel like this has become one of the most taboo things to even wonder about out loud, because it drives a different part of the Venn diagram into a freaking rage. How dare you force me to consider that perhaps I am only slightly above average in a population of billions?

Yeah I stopped my subscription during 4 thinking it wasn't super useful, but 4o does seem to handle programming logic pretty well. Better than me if given the same timeframe (of seconds) at least.

It's helped me stay productive on days when my brain just really doesn't want to come up with a function that does some annoying fairly complex bit of logic and I'd probably waste a couple hours getting it working.

Before I'd throw something like that at it, and it'd give me something confidently that was totally broken, and trying to go back and forth to fix it was a waste of my time.

Now I get something that works pretty well but maybe I just need to tweak something a bit because I didn't give it enough context or quite go over all the inconsistencies and exceptions in the business logic given by the requirements (also I can't actually use it on client machines so I have to type it manually to and from another machine, so I'm not copy pasting anything so I try to get away with typing less).

I'm not typing anything sensitive, btw, this is stuff you might find on Stack Overflow but more convoluted, like "search this with this exception and this exception because that's the business requirement and by these properties but then go deeper into this property that has a submenu that also needs to be included and provide a flatlist but group it by this and transform it so it fits this new data type and sort it by this unless this other property has this value" type of junk.

That being said, 4o is functionally in the same league as Claude, which makes this a whole different story. One in which the moat is already gone.