this post was submitted on 18 Nov 2024
333 points (94.2% liked)
Technology
59495 readers
3081 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
AI summaries of larger bodies of text work pretty well so long as the source text itself is not slop.
Predictive text entry is a handy time saver so long as a human stays in the driver’s seat.
Neither of these justify current levels of hype.
https://arstechnica.com/ai/2024/09/australian-government-trial-finds-ai-is-much-worse-than-humans-at-summarizing/
https://ea.rna.nl/2024/05/27/when-chatgpt-summarises-it-actually-does-nothing-of-the-kind/
Go look at the models available on huggingface.
There's applications in Visual Question Answering, Video to Text, Depth Estimation, 3D recreation from a photo, Object detection, visual classification, Translation from language to language, Text to realistic speech, Robotics Reinforcement learning, Weather Forecasting, and those are just surface-level models.
It absolutely justifies current levels of hype because the research done now will absolutely put millions out of jobs; and will be much cheaper than paying people to do it.
The people saying it's hype are the same people who said the internet was a fad. Did we have a bubble of bullshit? Absolutely. But there is valid reason for the hype, and we will filter out the useless stuff eventually. It's already changed entire industries practically overnight.
the reactionary opinions are almost hilarious. they’re like “ha this AI is so dumb it can’t even do complex systems analysis! what a waste of time” when 5 years ago text generation was laughably unusable and AI generated images were all dog noses and birds.
I think he’s talking about the LLMs, which…yeah. AI and LLMs are lumped together (which makes sense, but classification makes a huge difference here)
Even LLMs in the context of coding, I am no programmer - I have memory issues, and it means I can't keep the web of information in my head long enough to debug the stuff I attempt to write.
With AI assistants, I've been able to create multiple microcontroller projects that I wouldn't have even started otherwise. They are amazing assistive technologies. Many times, they're even better than language documentation themselves because they can give an example of something that almost works. So yes, even LLMs deserve the amount of hype they've been given. I've made a whole game-server management back-end for ARK servers with the help of an LLM (qwen-coder 14b).
I couldn't have done it otherwise; or I would have had to pay someone $60k; which I don't have, and which means the software never would have existed.
I've even moved onto modifying some open source Android apps for a specialized camera application. Compared to a normal programmer, sure - maybe it's not as good. But having it next to me as an inexperienced nobody allows me to write programs I wouldn't have otherwise been able to, or that would have been too daunting of a task.
Hell, even if you are a programmer and have no memory issues, it's a hell of a lot faster to have it boilerplate something for you for a given engine with certain features than to sit down and write it from scratch or try to find a boilerplate. Stack exchange usage has been going down regularly as LLMs are filling the gap.
It doesn't get you to third base or anything. But it does get you started and well-structured within the first couple minutes of code for any reasonably simple task.
Last year I worked on a synchronized Halloween projector project. I had the first week of work saved into my repo, but as Halloween approached, I wrote a lot of it on the server. After Halloween, I failed to commit it back and inadvertently wiped the box.
This year, after realizing my code was gone, I decided to try having copilot give me a head start. I had it start back over from scratch, asked it in detail for exactly what I had last year, it was all fully functional again in about 4 hours. It was clean, functional well documented code. I had no problem extending it out with my own work and picked up like I hadn't lost anything.
to be fair, you had already done the thing and learned from that process. you should give yourself more credit!