I like llms, or lin other words, I like that we are getting better at something.
However, just want to ask; what was the initial problem llms were trying to solve, what problem did they solve so far?
Do you have any examples in your life or work, which you can clearly say “we were not able to do this before llms, but now we can” or “we were able to do it, but not good enough, it was causing us some issues, now it is a lot better”
If the answer yes; second question would be like, does the total cost of those problem at least equal or exceeding the amount of investment on these models?
Thanks in advance
Now, before you jump on me saying that AI is wrong, this is true. But at the same time, I no longer can be 100% sure that whatever SEO optimized website I land at provides accurate information. If I need solid facts, I usually double check AI with various other sources. For queries like "best keyboard for software engineers", I'd rather get a table with pros/cons from AI rather than landing on whatever affiliate related website is promoted on Google. LLM gives me a good starting point to either dig deeper into particular products, or query further to find more suggestions.
Same for coding. I used to Google "how to split a string in ruby" and land on flame war, or 19 years old, StackOverflow question. Now I can get an updated answer from whatever LLM you prefer with a reference to the official documentation. It works for simple queries, as well as code snippets.
Lastly, I use LLMs to plan trips or gift ideas. I'd just throw in my preferences, and let LLM build a rough plan, from which I can iterate further, or start doing my own research.
I finally do now.
I have done so much in the last 3 months.
1. Cleaned up my personal website and blogs 2. Built a couple of learning tools for myself - https://rfc.stonecharioteer.com and https://github.com/stonecharioteer/goforgo 3. Setup OpenWRT and Adguard+Unbound at home, with a non-trivial failover with multiple WANs.
It's helping heal my burnout, something that crippled me for years and kept me from my side projects. It showed in my career too, because I've stagnated since 2021. I'm trying to improve now, and I'm relying on Claude Code and ChatGPT (albeit on legacy models) to do so. 3.
Because it's the wrong question!
It's not that LLMs solve entire classes of life/work problems. Instead, they take some life/work task (coding, ideas generation, learning about new topics, personal reflection) and make them x% easier, y% faster, z% better.
If (the royal) your claim to fame is “I codez real gud”, you would be screwed post 2022 with or without LLMs.
On that same note, at 51 years old, if my only means of staying competitive and employable is that I can reverse a b tree on the whiteboard, I’ve done something horribly wrong with my life.
For example:
- Recommend me some books on XYZ topic
- I have this idea X; can you tell me if someone has written about something similar?
- Evaluate my argument and suggest ways to improve it
...and so on.
As to whether it’s worth the cost, societally probably not, but there’s not much I can personally do about that so I try not to think about it too much.
It’s almost like reliving the late 1990s with far more ads, more vanilla websites, and worse search engine quality.
Claude and Gemini have been very useful in helping me come up to speed on a code base written in Go (a language I have used before but not for many years). Figuring out where the business logic lies, how the dependency injection is done, how the tests are written, what overall design pattern is being etc.
Of course, I could have done all this without LLMs but it would have taken several weeks/months longer. Letting the LLM handle the boilerplate and framework jargon lets me focus on the business logic and the design patterns, and helps me contribute much faster. But LLMs do often make mistakes so it's not like I blindly trust the output. They don't replace your colleagues in terms of being the ultimate source of truth. But it has speeded up the learning process, no doubt.
Also, when writing code I provide the style guide to the LLM as context and have it review the code.
GPT5 did the differentials for the complex mishmash of autoimmune symptoms, found the unrelated cause for the cognitive stuff. The cognitive symptoms do match autoimmune diseases too - lupus and MS - but it split them out and discovered a cause anyway
My neurologist ran tests reasoned out purely by GPT5, agreed they made sense. They were the right tests to do
It also interpreted my ~50 blood tests correctly. Oh also my contrast MRI images
Unrelatedly, it made a loved one's bipolar I a non-issue (yeah really). They live a perfectly ordinary life
I’m in a leadership/operational role at a small marketing agency.
Pre-LLM I was writing a variety of scripts manually to automate and simplify processes. With LLMs, I’m still creating scripts but writing none of them. The complexity of processes that I’m able to automate has gone up significantly and the time it takes to write working scripts has gone from (at best) hours to minutes.
No longer am I limited by my knowledge of how to code. Now I’m limited by my ability to explain our business processes.
If LLMs weren’t a thing, I’d imagine I’d have hired at least 1 and probably 2 people to work on automations full-time.
2. Giving some structure to my opensource project ideas. I had a good time getting over my analysis-paralysis while writing them down.
However now all that is way easier with LLM's and stuff like Claude Code, I don't have that dread anymore because I can always just increase/decrease the amount I rely on LLM's and use them as a Hail Mary so I am not spending hours searching a super specific weird bug.
I know it means I may not be learning as much, but I see it as a worthwhile exchange because otherwise I probably would have not gone into making apps or doing anything ambitious in the first place.
And of course coding. Use case 1: replace stack overflow. use case 2: coding agent -> "perform this task for me".
Not life changing but useful.
A common argument I hear about AI is "I could just write it faster myself", well I know CompSci and general info about a lot of software things but it would take hours of getting up to speed on areas I'm not an expert in to be productive. I can just delegate that to AI and get mostly correct outputs, this is okay with me and faster than what I could do.
I think the cost is going to catch up with the AI companies running the models (not the companies building products that call AI APIs) and that is when the bubble will burst. They will need to keep increasing costs and at some threshold, fewer and fewer developers in an organization will have licenses because it may become unaffordable.
this motivated us to get him a real therapist and have a long conversation about the dangers of humanizing ai
If you don't want to be banned, you're welcome to email hn@ycombinator.com and give us reason to believe that you'll follow the rules in the future. They're here: https://news.ycombinator.com/newsguidelines.html.
