I'm not sure what promises you are talking about, but I've found LLMs to be extraordinarily helpful for both my job and daily life. They are excellent at translation, summarization, troubleshooting, and brainstorming. I've used OpenAIs API to translate an entire epub, including the HTML so images are retained and the results were shockingly good after some prompt fiddling. With Claude I've received some excellent advice on decorating my living room, organizing my schedule, and quick hypotheticals. There are no pinky promises here, it already works.
For general Q&A they can hallucinate, but so long as you are using it to augment your productivity and not as a driver this isn't any different than using stack overflow, or any other kind of question you might ask on the internet. It's basically a non issue too if you upload a document into its context window and stick to asking questions about that document though.
>I'm not sure what promises you are talking about,
AI wiping out programming as a career. AI wiping out writing stories. AI replacing the need for doctors to diagnose illness. AI generally replacing all white collar jobs.
LLMs are useful assistants, but they are nowhere near the hype flooded everywhere a year or so ago.
Did everyone think it would take two months and all the doctors in the world would lose their jobs to ChatGPT?
AI is a societal shift that will take place over the next 20, 30, and 40 years, much like what happened with personal computing. This is a time horizon that impacts investments right now. Professions that existed for thousands of years will cease to exist. That is an unbelievably big change.
you should celebrate they are people in this world that think like this… as long as they are around we can capitalize on this :) like the people who were still riding horses when fords started rolling around… :)
FWIW, I think LLMs make better stories than quite a lot of the human writers on Reddit.
Not that many of the Redditors were ever going to go on to be successful novelists of course (and I say that as someone who is struggling to finish writing this darn book for the best part of a decade now…)
Honestly, and this is not personal, I doubt your ability to determine a bad summarisation or translation outcome. My wife is a professional translator and spends a good deal of time picking up the steaming wrecks that LLMs have left after someone went for the "cheap" option first. And we're talking best of breed stuff like DeepL here.
As for the other points, I rather like to spend some time thinking on them personally. If you're not connected to the decision yourself, what are you?
> My wife is a professional translator and spends a good deal of time picking up the steaming wrecks that LLMs have left after someone went for the "cheap" option first. And we're talking best of breed stuff like DeepL here.
Just so we are clear, for Japanese to English translation, DeepL is hot garbage compared to a top class LLM with the right prompt. DeepL translations are basically unreadable, and regularly just cut sections out entirely! So I wouldn't call DeepL "best of breed" by any means, it's not even at the starting line. Can't comment on English <-> French/Spanish/German/etc though, never tried it with those.
In my case the epub was technically a replacement for a fan translation I was reading, which was decent enough, but with a simple script and instructions to keep the vibes of a light novel, it got very good, I remain impressed. Next I plan to convert it all to markdown to see if I can help encourage it to structure paragraphs properly, the html tags have so far limited it to line by line translation.
When I've experimented with officially translated works, meaning cases where I've translated the raw and compared it to the official, it's still not up to par, but good enough in my opinion. I'm not aware of any payed service that streamlines this yet though, not sure why. It's nothing like traditional MTL.
> As for the other points, I rather like to spend some time thinking on them personally. If you're not connected to the decision yourself, what are you?
What? It's a dialogue, a conversation, I bounce ideas off it and ask for advice to help guide the direction of my thinking, have you ever even used an LLM? I do this with my friends and co-workers too, do you not do this?
This comes off as a bit presumptuous, an LLM lacks executive thinking, if I'm not directing the conversation then the LLM has nothing to give.
Just to note my wife is Japanese so she's aware of that. She does German / French as well and it's fine for that. But still needs a lot of work cleaning it up.
For general Q&A they can hallucinate, but so long as you are using it to augment your productivity and not as a driver this isn't any different than using stack overflow, or any other kind of question you might ask on the internet. It's basically a non issue too if you upload a document into its context window and stick to asking questions about that document though.