Is this true or is there just a massive gulf in application at the moment?
For one trivial example take something like indexing historical newspapers that you would really expect to be in the class of “solved problems” because it’s largely typeset, but the commercial offerings [1,2,3] are just chock full of errors and there is absolutely nothing I can apply off the shelf that gets better results without a ton of extra effort and compute.
1. Newspapers.com 2. Newspaperarchive.com 3. Genealogybank.com
It is only true if you drink the cool aid.
Speech recognition - Siri still have major issues undestanding me. Youtube text2speech routinely mistranscribes because it simply have no understanding of the language.
Machine translation is hilarious at best and dangerously wrong at worst.
Machine vision still could not spot the difference between pastry and furry animals last time I checked.
All of these are examples of non-working over hyped tech. It is not a list of "basically" solved problems.
DeepL is actually really good a lot of the time.
When was the last time you checked the status of machine vision, because the problem you suggested is not hard for it anymore.
A real human transcriber still outperforms any automated system in existence.
My company performed extensive accuracy testing and lets people choose to use a human transcriber if they want. Most people are perfectly happy to use the machine transcription. You are just seem bitter. Did an AI steal your wife?
Machine translation still makes a few mistakes but hardly more than a human.
Machine vision: I worked on a factory where a machine would approve/reject products passing through at a ridiculously high speed, and it never got one wrong. This was 15 years ago.
Your experience is completely the opposite of mine.
Isn’t text2speech the opposite of transcribing?
Speech2text would be transcribing, and text2speech would be speech synthesis.
Anyway, assuming you meant speech2text: I found YouTube‘s transcribing quite good. It even understands stuff that is inaudible to me (especially in movie snippets). Of course it’s not perfect, but neither am I.
I picked a random passage from a novel in French I am currently reading. ChatGPT translated the three paragraphs I ran it on correctly; there are no major quibbles to be had. It is good, coherent English, a correct translation, which closely follows the French original, even capturing some of the poetic imagery effectively.
I'm sure after another paragraph or two there will be a weird screw-up. And there's no consistency in a running translation of any length. Etc. Yes, it's not perfect. Not fully human-equivalent.
Still. I remember when machine translation like I just did was the realm of science fiction. And I thought it would remain science fiction for a long time. The fact that such a thing isn't mind-blowing shows how far things have come, hasn't it?
> Speech recognition - Siri still have major issues undestanding me.
I am using speech-to-text AI transcription every day. It's been revolutionary for me. I am hard of hearing. The cutting edge is Whisper, and it is leaps and bounds over the state-of-the-art just a year ago: https://github.com/openai/whisper
Have you tried OpenAI Whisper, especially with its "large" model? Siri and Youtube shouldn't be the yard sticks to judge the entire field, they both have unique hardware constraints and they're far from the state of the art.
Siri isn't even the latest models that have a much lower error rate.
But at the same time, people are flipping out because it is, well, machine learning. I was following AI, mostly as a hobby interest, though I did some undergrad courses, before it got all buzz-wordy. Just ten, even five years ago, some of the stuff being demonstrated, were the sort of thing I thought would take many decades to develop, and that I may not see in my lifetime. Speech recognition, machine translation, machine vision, for example.
For more than half a century, good machine translation and speech recognition were the holy grail of AI. An incredible amount of work went into them. Specialized systems, grammars and parsers, programmed general knowledge databases. Collecting and correlating enormous aligned bilingual text corpuses with phonetic annotation. After several decades of such work, it got us to mediocre-quality Google Translate style services that were the norm until recently.
Now, without any of that specialized work, with a much more general learning algorithm, there are translation systems that are superior to anything humans designed to specifically be machine translators. It just falls out accidentally from the large language models -- they can also, incidentally, translate, and do better than anything specifically designed to translate ever did.