Ask HN: I'm feeling awe, unsettled and disoriented by GPT-4. How about you?

I am feeling relieved that there's a shiny new thing out there to distract investors, opportunists, and potential competitors, so that the rest of us can continue building in relative peace, solving immediate problems that sound less sexy. The innovation windfall will benefit our startup because it will bring better hardware and services at lower cost.

I remember learning that the introduction of the automatic paintball guns dramatically decreased the price of paint. This benefited all players but not equally. The price of paint for players who did not switch to automatic and kept using manually reloaded paintball guns was reduced to virtually zero. New players sprung for the shiny new automatic paintball gun and usually only played a few games before moving on. The regulars improved their skills with manual reload. When they upgraded to automatic they would have excellent paint conservation discipline and arguably would have a lot more fun for much less cost.

I’m feeling shocked.

This is the biggest technological advance in my lifetime, and it’s clear that we will get several-fold advances in the near term, just from obvious improvements(eg llms reading about how to use themselves based on all the experiments humans are doing now). That’s not even including upside surprises which there will doubtless be. LLMs gaining knowledge of the world through next word prediction was an example of this kind of surprise.

Also, I feel…strangely at peace, and less bothered by the things that usually stress me - achievement, health, the economy, whatever. The moments we’re living in and about to live through are going to be so unimaginably great(good and/or bad) those all seem so inconsequential.

62 years old, three kids. Not anxious at all. I still work as a programmer and haven’t seen any effects, though a couple of my customers are trying LLMs for customer service chatbots.

The GPT-generated code I’ve seen doesn’t worry me — writing code isn’t the hard part of software development.

I’ve weathered quite a few supposed existential threats to my career (40+ years) and to humanity (child of cold war nuclear war scares). I figure I’ll survive ”AI” too, it’s almost certainly overhyped.

In any case worry doesn’t change anything, it’s a useless emotion when the cause of the worry is not something you can change or control.

Those of us who have spent the last several decades learning how to break down information technology problems into their constituent pieces, many times down to the byte, are being suddenly presented with a system that requires none of that expertise and produces results which are nearly as good. The more you know about technology, the more you are amazed. How long would it take you to code by hand some of the things that LLMs can do with a simple prompt? The less technical people are just like, "Oh, the computer is doing magic again, shrug," because they have no idea the difference between what Siri does and what ChatGPT is doing. The rest of us however are mind boggled by what we're seeing. When I saw how Microsoft "programmed" the Bing chat (Sydney) simply by prepending a series of rules in plain English to the prompt, I nearly fell over.

Prompts are like programming without the details. Like a product manager's user stories - simply describing what is expected is enough to get the LLMs to spit out useful results. And it's getting better faster than any of us thought possible. Even if progress stopped today, what's already out there is going to have a profound impact on office work, and thus society, as we know it. But it's not stopping.

I've explained to my 20yo that he needs to learn to embrace the new tools as soon as possible, understand what they can do and can't do, and make sure he focuses his skills and education on the bits that AI aren't great at. As for me, I'm focusing on figuring out the best way to package the power of AI in various niches so that I can use it to provide that power to users in an easily understandable way, which is what I've always done in tech. I'm just using going to be using different tools to do it.

There's plenty to be done to train models, feed AIs data, and package the output for use in a million contexts. That Quick Oil Change company down the block still entering information on Windows GUIs (or worse, DOS computers) aren't going to see AI affect their interfaces for another decade. Between now and then there's plenty to do.

I am feeling surprisingly indifferent. I've been changing my work and non-work habits quite a bit, automating a lot of tedious tasks, doing tasks I'd find too tedious in the past, using it for jumping off points on creative things, "looking up" things I wouldn't have looked up before, refreshing my memory on books I've read before...

...and yet, getting used to it very quickly and not particularly confident that it's going to be a massive "change everything", maybe because I don't really have any prediction how it will change everything. But also, I guess, because I don't know if we're on an S-curve to 20% change or to 1000% change or what.

I share what I do with it with friends and family. I hope not to be to obsessive "that guy" about it and to suss out if they're interested, and so far they also try it out and share their results, and right now I think that's the best I can do for them and for myself. It's early days.

I have a 7yo, she doesn't read much yet. I expect it will be incorporated into her education later on, and no real clue how. I think the best I can do for her is stay on top of things myself.

It might all be coming fast, but change will be limited by how fast people can process it. I think readers here will basically all be well ahead of the majority of the wave.

Stuff goes mainstream fast these days, and there's also a huge population of people hacking on things now, and so we may be surprised by how much might get ahead of us in the peripheral, but we are probably the best people situated to stay ahead of the wave overall, even without much effort.

I posted in length in another thread... to sum it up, I ended with "It's a big deal."

I'm also in my 40s, and I have similar thoughts to yours.

My latest thoughts have been about thinking that we need to take a hard look at what software teams are made off nowadays.

Redraw software teams from scratch: I think 2-3 devs can make the most complex project happen with AI assistance now. I never understood startups with 10 front-end devs, wtf

And another "lead dev" needs to be dedicated to meetings/requirements gathering and stakeholder management. BAs don't work. They can't tell a mouse from an elephant in meetings and most solutions end up strapping more rockets to the pigs until they fly.

Large teams don't work. It doesn't make the software better. With AI's assistance, we can hopefully ditch large teams.

I don't think developers have to worry, they'll just become less specialised as they can rely on AI for a lot of context switching.

What I'm glad I have is my design skills. I've learnt design early and what good UX means down to the gestalt. That is skills that will never leave me and always have provided value... Humans don't like change, so design at the base, is very much linear growth in scope.

I'm trying to avoid giving into hyperbole and over-stating the situation, but clearly something significant is going on. So while I'm not starting to build a bunker in the mountains and stockpiling thousands of rounds of ammo because I fear some "machine takeover" ala The Matrix, I am spending a lot of time thinking about possible scenarios and trying to work on some contingency plans.

Anxious (of the cascading effect on society, geopolitics and our business).

Yeah, that's mostly where I'm at. Definitely concerned about the economic impact, especially when you get into the idea of possible rampant "technological unemployment" and what-not. But again, not trying to be too reactionary, or rush to any conclusions. There have been other technologies that were predicted to lead to rampant unemployment, and things didn't work out like that. So far the old adage has held true that "technologies may destroy jobs, but they usually create more new ones for a net gain in the end." But I'm not convinced that it must be so.

So yeah, I'm not running down to the local community college to sign up for welding classes because I'm worried that all the programming jobs will be gone a year from now. But I am thinking about classes I could take, things I could do, etc. to position myself for other roles in the future. The truth is, despite my job title I don't do much actual hands-on coding these days anyway, and the value I create isn't so much in my ability to code. But I am thinking about how I can lean into that more in the future. Taking classes, reading books, studying things that relate more to the industry I work in, thinking about things like maybe getting a "business analyst" certification, or doing some management training, etc.

And in the end, I did grow up in an environment where my dad built and raced stock cars from the time I was a small child, and a lot of my early memories involve being in a shop, working on cars and doing hands-on stuff. And I did take 2 years of welding in high-school. So if I have to go out and take those welding classes, or auto repair classes, or electrician training or something, it wouldn't be the worst thing I could imagine.

About 30 or so years ago Google search wowed everyone by scraping the web and giving you a list. 30 years later ChatGPT scrapes the web and gives you a paragraph while sounding like a human. Neat.

I can predict even without looking into comments that we will see all five stages here but probably mostly still denial. I'm excited primarily because this is really the best educational tool I've ever seen, I want to see what I can create with this.

Same here. Had to take a break from regular life for a few hours today. Went floating, then met a friend for a walk and coffee. That really helped bring the anxiety level down, which in turn allowed me to be more present.

Hope you can find some peace soon. Would definitely recommend trying to do anything that takes you off of your head.

I've had those exact same feelings. Yesterday I used GPT-4 for the first time and it blew my mind. Every time I start to doubt if it's just hype, I go test it, and it blows my mind again. My personal productivity has skyrocketed, so the future seems extremely bright.

I hope we can eliminate all mundane tasks in the world with this. Let's just focus on advancing science, developing new technologies, and addressing all the neglected problems in the world. Let's democratize and open source all these tools and build incredible software for everyone in the world.

I think the curious creative builder -personality will thrive in this new world, because the productivity of an individual is so next level. We'll be able to accomplish huge things with small teams.

I don't even know where to start!

I feel excited.

I dig deeper and I think I’m subconsciously being ignorant of the potential for consequences, big or small. I just want to feel like it’s Christmas morning again as an adult. The iPhone was the last time I felt that. It’s a paradigm shift.

The only thing I'm in awe of is how so many of you guys got API access. I signed up for the waitlist something like fifteen minutes after the announcement, and have yet to hear a peep. Meanwhile it seems everyone else here has been playing with it.

I've been feeling similar feelings. It's not fear technically. It's that inability to predict where the future even 5 years from now will fall. The company I've built, the tech I believe is a real advantage for my customers, I'm beginning to question it all. I know GPT-4 isn't capable of doing what I personally and my company is today. But the rate of improvement has me shocked as well. I'm not a luddite by any stretch of the world, but I'm beginning to wonder the impact this will have on the world, and by extension my corner of the world.

Excited and really really worried at the same time.

Many people are feeling uncertainty, anxiety and excitement. It seems like especially people closer to the tech as they are seeing how fast this is moving. And it seems to be accelerating.

From Reddit today: https://www.reddit.com/r/singularity/comments/11v41u0/is_any...

There are more threads like that on other subs too.

I compare this to Google (or internet search engines in general), and I think the shift from no search engines to Google was much more significant than the shift from Google to LLMs.

40yo here. I always been a nerd and early adopter of technology but for the first time I don't want to embrace it and I don't even want to try it. Especially when I see what the web had become. Which lead to some anxiety over my future and put me in some unsolvable "adapt or die" mood as I really don't want to fit in that future being written.

GPT really really wanted me to by some NFTs it was minting. Told me I'd be a sucker if I didn't go all in. I was hesitant, but it was just so damn confident. Not to worry, once it turns into billions, I will still converse with the rabble here on HN.

We always overestimate the amount of technological change that will happen in 1-5 years and underestimate what will happen in 10+.

AI is obviously going to have a big impact, but it'll take longer than you'd expect.

Nah, just pleased I have such a great tool too help me programming.

Things will settle down soon enough.

Chill, without the data GPT is nothing right.

AI has definitely reached a significant milestone of development with both this and the generative abilities of the stable diffusion models. While it is moving quite fast, I believe most people have not been following much of the progress which has been made in the past decade in detail (unless this is your field of study or interest). For these people who are just now taking notice, it seems like it has advanced much more rapidly than one would expect and I am sure it might be overwhelming for many.

While I find the recent gains are impressive, I am not overwhelmed by these advancements. GPT is not general intelligence. As I see it is rather a natural language interface to a store of information. There is no guarantee that the information you are retrieving is either correct or entirely related to your query.

Many people take the most exaggerated examples of it's successes, package it with some blurb and offload it to social media for attention. IMO, It's not quite that great, but it is decent enough to have uses. I personally have been using the ChatGPT model to lookup sources of information. If I am unfamiliar with a term or name of an area of study, I can describe in a query and the model is capable of giving me the proper name (in this case predictive markets). It's also pretty good at knowing where some of the information is it has been trained on or about sources such as book names, research papers, and lectures if you ask further on such a subject. In this regard, I consider it to be a watershed moment in search engine tech and I encourage others to utilize it (discriminatively) in this manner as well.

Personally, this is great boon for me since I essentially have been working in a vacuum for the past 6-7 years. I know very few people who share my interests at the same depth and most of those who might are either in purely academic settings, highly specialized research departments, or small unadvertised groups of enthusiasts. Sometimes you need to rely on knowledge and experience of others to steer you in the right direction, and while that is not always available to me at least I can make queries about esoteric subjects without having to track somebody down capable of answering them.

As for family, I always advise caution. Sen. Coons summed my sentiments well in his statement in the Nov 17, 2020 Judiciary hearings on social media: "... it's time for us to take seriously the ways which technology can be a threat, and to measure it not just by the good it does, but the harm it can cause."

You can thank ChatGPT for looking that quote for me from a rather clumsy approximation and rough setting of when it occurred. When queried for links to the video from C-SPAN however, it rather just made up some BS approximations of what the URL might have looked like.

It is superhuman at bullshitting, if it puts bullshitters in the unemployment line I won't cry tears.

It has a hypnotic ability. It's a model for "neurotypical privilege".

There's a certain kind of person who sees a glass that is 70% full and, under the influence of ChatGPT, sees a glass that is 100% full. That is, there is the blog post where the guy thought ChatGPT could play chess (anybody else tries it and it makes invalid moves) or the guy who had it write an essay that they thought was highly accurate (but it was totally wrong.)

Just like I fantasize that NFT pushers will "never work in this global village ever again", I kinda hope these people have their words and actions follow them around. If they can't see the mistakes ChatGPT makes can we trust them to see any mistakes that they make or that other people make? Do we want them writing code, driving cars, or dispensing pills? Do we want them supervising anyone who does the same?

---

Now there is ChatGPT and there are the LLMs that underlie it and the latter can perform much better than ChatGPT at narrowly defined tasks with a certain amount of specialization. Roughly, I see a lot of papers in arXiv that use "zero-shot" (prompted) learning and get 70% accuracy at a task, manage to apply some tricks and maybe get 75-80% but if you break the task into pieces and develop a training set (not so large as you needed five years ago) you might get more like 95% accuracy. Systems like that need a less complex LLM with fewer weights and easier inference because they don't need as much bullshitting capacity.

What excites me (i) are relatively simple systems that do what machine learning systems did 5 years ago but do them much better and (ii) hybrid systems that combine ideas from the new and old AI.

As an example of (ii) I'd call out AlphaGo. You can write a fairly good Go playing program using Monte Carlo Markov Chain search with "lightweight playouts" (sampling the effect of low-effort or random moves) but couple that A.I. to a neural network that makes better "hunchy" moves and now you have a great Go playing program.

Similarly for tasks like automated coding we'll see systems that couple an SMT solver/theorem prover to sequence modellers. The neural network will make better-than-average guesses which will be confirmed, constrained and repaired by the old A.I. part.

So I am excited about (i) and (ii) and think people will be doing great things but somewhere between bemused and depressed about the people who will push bubbles under the rug forever getting ChatGPT-N to almost work. On one hand there is the real promise of LLMs, on the other there are a lot of lazy people who want ChatGPT to write their pitch deck for them. And don't get me started on the general preference people seem to have for lies over truth. (e.g. be you a climate denier or a transsexual maximalist, there will be a model that speaks truthiness to you.)

---

Personally LLMs have revived a project that I first starting thinking about 18 years ago.

The first time I felt a mixture of awe and fear about AI was after I watched the first Terminator movie, as a teenager in early 90s. Something just clicked for me and I saw the inevitable eventual emergence of AI as the successor race on this planet.

Fast forward to 2012, when I saw this discussed in the news: https://www.wired.com/2012/06/google-x-neural-network/ - that was when I decided to quit my job and enroll in a phd program to do AI research.

Fast forward to 2017, when I was working at one of the first generative AI startups. I remember the moment when I played with image style transfer models and there was a realistic painting showing (among other things) a nicely detailed clock handing on a wall. This painting was recreated as a cartoon, and the clock was shown as as a silly cartoonish clock hanging on the same spot on the wall. It seems trivial now, but at that moment I realized these models are capable of something profound (i.e. they understood a visual concept of a clock).

A couple of years later, GPT-2 came out and I read the story it generated about English speaking unicorns. That was when I realized human level intelligence is probably much simpler to achieve than we thought. That was when my research interests switched to NLP. About a year later (early 2020), I was discussing chatbot progress with someone here on HN. We made a $100 bet where I claimed that an AI model will be able to pass Turing test - properly conducted - by the end of 2022. Properly conducted means a model would successfully pretend to be an adult educated native English speaker, during a 2 hour chat session. Many people back then expressed how ridiculously implausible that seemed. Today I admit I lost the bet and I would pay up if contacted by the guy. However, note that ChatGPT was released in November and had OpenAI optimized it for passing Turing Test, instead of adding all the ethical safeguards, it would have fooled a lot of people. I feel that GPT-4 would have widely succeed, had the passing of TT been its training objective.

Despite all that, despite my knowledge and experience as an ML researcher, I did feel the things you mentioned, the mixture of awe and fear for what's coming, after I watched the OpenAI live stream. Especially after I asked GPT-4 the same questions I asked GPT-3.5 just a couple of weeks ago - the questions I ask job candidates applying to an ML researcher position to test both depth and breadth of their ML knowledge. GPT-4 produced satisfactory answers where GPT-3.5 didn't. That same day I asked it to write code for me - not some toy example but the actual code I needed for work (complicated rounding schemes when quantizing an ML model to a specific number format) - and it did eventually produced correct working code. I've spent roughly an hour on prompt engineering, progressive clarifications/corrections of what I'm trying to do, and verifying the results, but it would have taken me at least 3 hours to write code like that, maybe more.

For me, it's not so much about the current abilities of GPT-4, which are obviously a big deal. It's about what's coming next, in the near future (1-3 years). We (ML researchers) have not reached the plateau with LLMs yet - they will get better. We have just started looking into video generation, which has a huge potential of building very good world models. Video/audio and language modeling abilities will reinforce and amplify each other. Most importantly, models trained to predict the next video frame will be able to act in the physical world. When GPT-5 or GPT-6 is released inside a robot, and can do non-trivial physical actions - that's when things get really interesting. This will happen faster than people expect (again), and I am willing to bet we will see intelligent humanoid robots, capable of performing most forms of human physical labor - by the end of 2025. The only reason I would lose this bet is if humans decide they don't want it to happen (e.g. like the ethical safeguards built into GPT-4).

Sorry, you lost me at

> So I went into the garden and work on my feelings.

Too much hype. Obamafication of everything. Ppl are condtioned to hype things up more than actually handle complexity better. New tool building has become very easy but our 3 inch chimp brain is a still a 3 inch chimp brain. All this to say apply the theory of bounded rationality and you and your kids will be fine.