Ask HN: What are the big success stories of AI/ML so far?

OCR, handwriting recognition and other more niche computer vision. ML computer vision is a step change vs classical methods in many cases (and no better in others just a lot of window dressing)

The "bombastic news headline" stuff often supports improvements in real applications too. It's just that it ends up getting shown off with party-trick type models that look cool but don't go anywhere, adding to the boom and bust cycle (for example search HN for "generative" AI to see a mini hype bubble in full swing: https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu... )

It's not a question if AI can think, it's whether humans will.

Targeted advertisements

AI research in the 70s and 80s went on to kill a lot of people, so there's that.

Success? Idk. Openais free trial is good enough for writing Reddit arguments for when you can't be arsed to put the effort in yourself.

Covid 19 research

I know github copilot gets a lot of flak, but as a user it has been an immense enhancement to my productivity, especially when working with frameworks or languages I am not super familiar with. It is so much easier to write a comment on what I want and then get an implementation that just works than to google something, get a half-matching stackoverflow answer, and adjusting it to do what I want.

I consider it one of the few projects that use modern models to do something that is directly beneficial to me.

Voice recognition. Voice generation. Identification of music recordings. Game play (chess, go, poker, etc). Object detection, classification, and identification in images/video. Translation between pairs of natural languages (of conversation more than writing). Tracking of moving targets. Human face recognition. Flight control of drone aircraft. Driving and navigation of automobiles on- and off-road (esp. in the absence of traffic). Noise reduction in many forms of digital signals.

Deep learning using GPUs and large amounts of training data has redefined the state of the art in most uses of signal processing (voice, music, images, video, etc) and game play.

Noise suppression and audio processing for conferencing. Most people likely use it daily.

At one of the shops I worked at a guy used ML to auto route tickets to the appropriate team. Absolute champion.

We use ML for the AIs in our digital boardgame adaptations on Steam and phones. Our Race for the Galaxy AI is better than 97% of our players and our Dominion AI (still in beta) is at about the 75th percentile. Not only is it good at the games, it's super adaptable when a new expansion comes out or the designer adds errata for cards. Finally, it's been helpful with the design of new cards. A quick training session and we can put the cards in front of the designer and they can supplement their play testing. The AI has even found some 'exploits' that new card combos create that we can address before release.

Image recognition, self driving, voice to text, text to voice, text to images, source code generation, text generation, translations, image editing, classification, forecasting.

That’s now basic technology that I use regularly at work and in my personal life.

In my opinion, computer vision had many successful uses of AI/ML models. Object detection, segmentation, there are multiple uses in medical image processing.

As I am from manufacturing side, object detection and segmentation models changed product quality inspection for good.

Now multiple cameras with trained models can find defects on any surface or part as small as it can get, before it was multiple operators looking at moving parts 8 to 12 hours a day which was very tiring and repetitive job to do.

I am not sure if this counts. Also second hand information but apparently safety in mines has been improved through AI monitoring camera feeds. Alarms/warnings go off when people get to close to certain equipment. This most likely has prevented not only economic losses but probably saved lives.

Image recognition makes my life easier with Google Photos. If I want to search for a passport photo, I type in passport. If I want to find all photos with me and a friend, I can do that pretty easily.

Protein folding -- DeepMind

There are many products that incorporate artificial intelligence and machine learning that have become so ubiquitous that they no longer counts as artificial intelligence. Spam filtering and recommendation engines are two such examples.

Music separation. That is, taking a mixed song and splitting it into separate drum, bass, vocal, and other tracks.

GPT3 is magic

IMHO: DALL·E

https://openai.com/dall-e-2/

Tons of internal stuff; think of anything that is minimum wage knowledge worker does, some fraction of it is probably obvious enough to be automated. For example, categorizing PDFs (which go through OCR) into broad buckets so that the right specialist can look at them – the system might not be able to figure out 100% of them, but even taking a chunk out of the human workflow at 99% accuracy is pretty valuable It’s important understand what a massive flood of PDFs some organizations get on a daily basis