Ask HN: Is anyone else bearish on OpenAI?

But my mind kind of works like: once I'm stuck I endlessly do my best to understand it, which sometimes mean that I grind to a halt. ChatGPT stops that entirely.

The fact that it always produces 2 to 3 bugs is fine.

I use ChatGPT 4, GPT-4, and Copilot every day. It is an "average intern" at many many things. Here's how I feel it helps me:

* Its interactivity lets me learn a lot (superficially) about new topics. I can then expand that with research of my own

* It helps me think outside the box or from other perspectives when thinking about e-mails, proposals, real-world scenarios

* When exploring a new language, framework or technology, it points me in the right direction.

* For quick scripts, using the code generation/analysis feature, if I direct it right (i.e. layout "the plan" beforehand and ask it to work the rest on its own), it gets a lot of it right pretty fast, saving me some time writing the code, figuring out the right libraries and the nitty-gritty details.

* It is great at giving ideas for why something might not be working.

Real things I've done with it:

* Discuss ongoing negotiations with clients, trying to better my proposal and better understand the clients point of view.

* Learn more about managerial or "business-y" topics, by allowing me to discuss things with it and iterate on that with my own research. It is a valuable "white board" to discuss with.

* Adjust my e-mails so they are more appropriate to the situation. This can involve changing the tone, shortening them, adding more detail, etc.

* In general, i've used it to find flaws in reasoning when dealing with people. For example, it has helped me question my own client proposals or approaches, by specifying where I was lacking (e.g. because I was vague, or pessimistic, didn't give a mensurable objective, seemed to ignore a client's request, etc)

* I use a command-line utility from the shell which lets me ask it to do something and then have it do it. I now use this with some frequency to just write the commands I would have to google because I haven't memorized. Things like ffmpeg or imagemagick arguments. Or combinations of grep, sed, ls, find, git, etc. Here are some examples:

i) "merge jpgs in this folder in order onto a pdf. each jpg should fill the page 100% without margin. Avoid complex calculations".

ii) "zip this folder with maximum compression and password 12345678'".

iii) "git remove branches matching pattern sprint-* remotely'"

iv) "use imagemagick to convert to jpg lossless and rotate 90 deg clockwise ~/home-en.png"

v) "Add _en before .jpg in all files in current directory. So they become _en.jpg"

vi) The list goes on and on...

* It has helped me cleanup nginx config files

* I have thrown code at it which I suspect has a bug. With a bit of context and some back and forth, it has helped me find it.

* In languages or frameworks I don't use often, it really shines. It has helped me write several applescript scripts which I have also cobbled together to create Alfred workflows. If I need to code something in a language I don't often use, what it produces is good enough for me to iterate on.

* It has helped people at our company improve their copywriting (when used with a lot of care)

* I have used it to help me critique my own poetry and improve how I approach poetry in general. Highly subjective, I know

* When trying to figure out how to use apps I don't often use, or dealing with unexpected behaviour in them, it often helps me find the issue. Notable apps in this category include Photoshop and Excel.

* I don't often do frontend so I'm particularly bad at styling and organizing things. When I ocasionally have to do frontend, it often gives me the initial "skeleton" very well.

I have seen many people try to use these tools, and here's where I think they SHOULD NOT use them:

* For facts, obviously -- which is unfortunately what many people actually try to use it for

* For writing (without check) most of your e-mails our posts, especially with very little context

* For ideas where you need "just enough creativity". It's a very fine line. Think brainstorming ideas for UI elements in a new website according to specific brand guidelines

* For incredibly specific details or overly complex tasks. You can't often have it write your 200 line function out of thin-air!

It is clear that GPT provides me with:

1. A better search engine, which skips past loads of outdated content or SEO-laden bullshit. It's just a shame that I can only use it this way for more specific "creative" or "problem-solving questions" questions. For important fact-based info, I always have to check what it says.

2. A partner with whom to discuss ideas and iterate over them, especially on topics I don't know. If anything, it's a great rubber-duck

3. A way to forget the underpinning of some of what I do and approach it with natural language. I find myself asking it to write my bash sequences instead of thinking them up myself. Of course I check them and 99.9% understand them, but it's just so much easier to have it do it for me.

4. (Copilot is an upgrade of my IDE. And a great one!)

If modern Google (and Bing and everyone else?) weren't so shit, I wouldn't need 1. To this day I still first go to google, eventually give up because it doesn't answer me properly and then go to GPT. It's ridiculous. I think 1. brings a lot of value to many many people. Google is so absolutely shit nowadays I can't believe it -- and the more I use GPT to get a comparison baseline, the more I am shocked. 2., 3. and 4. are incredible additions to my workflow.

I truly believe we are heading in the direction of interacting with computers via a new interface. Even writing code might eventually be affected. Perhaps, like we have built high-level languages on top lower-level languages and machine code, abstraction over abstraction, we might end up with some form of language writing which uses (very well controlled) LLM technology under the hood. Instead of directly writing the code, we tell it to write it. We still need to be smart to pick the right algorithms and structures, but no longer have to worry so much about writing the nitty-gritty syntax or details. Maybe?

Using GPT is its own art. Crafting the right prompts and getting a feel for how it works is very important. It is also essential to have critical-thinking and know when to question what it says and when not to.

A friend of mine told me this: "The companies who strive on hiring loads of shit people to do work are the ones who will suffer more. The others, which have hired smart people capable of critical-thinking, will benefit. GPT obliviates the work of many small shit people. Pair GPT with someone who is already smart and has very good problem-solving or critical-thinking skills, and now you've got a team that can obliterate the teams of many small resources.

I don't know if OpenAI will succeed or not, but I do know this technology is absolutely life-changing for many, and going back to a world without it being this acessible will likely be a net-negative.

It can be true, but I don't think it is always or necessarily often true. The overall value proposition for interns (and apprentices etc) for society as a whole is that they will then go on to become the professional knowledge workers that they have learnt from. LLMs won't become that, so the value proposition is limited and localized. Remember the adage about garage mechanics, where the price goes up depending on how much 'help' the customer wants to give?