I liken it to VR. That was a big hype before AI and while I really love the tech (I have 5 headsets) I could have told anyone that the expectations were insane. The investors truly believed that in 2-3 years time everyone would be doing everything with a big headset on. It was dragged into lots of situations where it didn't belong.
Then of course the hype collapsed and now even the usecases where VR shines are deemed a flop. But no, it's exceptionally good at simulation (racing/flight) and visualising complex designs while 3D designing.
I see the same with generative AI and LLM. It's really good with programming. It's definitely good at making quick art drafts or even final ones for those who don't care too much about the specifics of the output. I use it a lot for inspiration.
But it's not good for everything that it's trying to be sold as. Just like the VR craze they're dragging it by the hairs into usecases where it has no business being. A lot of these products are begging to die.
For example an automation tool using real world language. For that it's a disaster, it's inconsistent and constantly confuses itself. It's the reason openclaw is a foot bazooka. It's also not very great at meeting summaries especially those where many speakers are in a room on the same microphone.
I don't think AI will disappear but a realignment to the usecases where it actually adds value, yes I hope that happens soon.
> It's also not very great at meeting summaries especially those where many speakers are in a room on the same microphone.
It is astonishingly poor at this. My intuition was that it should be good at this (it is basically a translation problem right? And LLMs are fundamentally translation systems) but the practical results are so poor. Not just mis-identifying speakers (frequently saying PersonX responded to PersonX) but managing complete opposite conclusions from what was actually said.
I'm genuinely intrigued as to what approaches have been taken in this space and what the "hard problem" is that is stopping it being good.
I mean it is a tough problem, you'd really have to voiceprint each speaker. But I'm sure this is technically possible considering voice cloning is pretty commonplace now.
And yeah the transcription quality also drops a lot. Where humans are still quite capable at reading it. Sometimes when I read the transcript I'm quite surprised it manages to make any intelligble minutes out of it at all.
I just don't understand how Microsoft place this feature as a minute-taking replacement when it's not ready for really super common usecases.
Same. Plus I have kids and a mortgage. My company just had a 5% layoff and I survived. Seems like it will be the first of many rounds. It's a really stressful time.
To be clear, I'm confident the impact of AI is going to be massive, and that massive impact is already underway rather than years away. But, separate from that, having seen it up close Block was bloated as hell
Yeah. We are coping. Just today, I had a simple bug where the data received was throwing undefined because it was in 2 alternate formats.
I showed ChatGPT(free-tier) the API response and the part of the code reading it, and it fixed it in 5 seconds. Would've been pretty short either way less than 30-40 mins but it's very good for simple tasks like these. The solution is just correct.
Even if they are right about quality, people on here vastly overstate the value of quality. From socks to dishwashers to airfares, slop is a valid product as long as it is cheap. Security from a business perspective has been proven not quite optional, but it is hardly catastrophic if it fails.
I recognize that my experience may not be typical, but I spend the vast majority of my development time improving the quality of the systems I work on, in response to specific customer demands for it. The last time I had multiple consecutive weeks of greenfield development was in 2021.
This thread is a great discussion and I have kept coming back to it over the last couple days to read more of it when I have a chance. I’m kind of disappointed that it artificially ended. I think at some useful comment threshold level you just have to let it go.
>But it shouldn't matter if he gave 5 bullets to Chat gpt that expanded it to a full page with a detailed plan.
The coworker should just give me the five bullet points they put into ChatGPT. I can trivially dump it into ChatGPT or any other LLM myself to turn it into a "plan."
I feel the same way, if all one is doing is feeding stuff into AI without doing any actual work themselves, just include your prompt and workflow into how you got AI to spit this content out, it might be useful for others to learn how to use these LLMs and shows train of thought.
I had a coworker schedule a meeting to discuss a technical design of an upcoming feature, I didn't have much time so I only checked the research doc moments before the meeting, it was 26 pages long with over 70 references, of which about 30+ were reddit links. This wasn't a huge architectural decision so I was dumbfounded, seemed he barely edited the document to his own preferences, the actual meeting was maybe my most awkward meeting I've ever attended as we were expected to weigh in on the options presented but no one had opinions, not even the author, on the whole thing. It was just too much of an AI document to even process.
If ChatGPT can make a good plan for you from 5 bullet points, why was there a ticket for making a plan in the first place? If it makes a bad plan then the coworker submitted a bad plan and there's already avenues for when coworkers do bad work.
How do you know the coworker didn't bully the LLM for 20 minutes to get the desired output? It isn't often trivial to one-shot a task unless it's very basic and you don't care about details.
Asking for the prompt is also far more hostile than your coworker providing LLM-assisted word docs.
Honestly if you have a working relationship/communication norms where that's expected, I agree just send the 5 bullets.
In most of my work contexts, people want more formal documents with clean headings titles, detailed risks even if it's the same risks we've put on every project.
On this topic I think it’s pretty off base to call HN a “well insulated bubble” - AI skepticism and outright hate is pretty common here and AI negative comments often get a lot of support. This thread itself offers plenty of examples.
> there are about 230 billion* links that need visiting
> * Thanks to arkiver on the Archive Team IRC for correcting this number.
Also when running the Warrior project you could see it iterating through the range. I don't have any logs handy since the project is finished but they looked a bit like
https://goo.gl/gEdpoS: 404 Not Found
https://goo.gl/gEdpoT: 404 Not Found
https://goo.gl/gEdpoU: 302 Found -> https://...
https://goo.gl/gEdpoV: 404 Not Found
They are useful for putting URLs in print materials like books. Useful for sharing very long links in IRC and some other text based chat apps (many google maps links would span multiple IRC lines if not shortened, for example). They are good for making more easily scannable QR codes.
reply