I bet a couple of ppl here are familiar with recent developments in AI. Depending on estimates, its impact will likely be larger than that of industrialization. Hard to imagine, but I can see where these futurologists claiming that come from.
Do the techers here fear being replaced in the not so far future by a combination of STT, LLM and TTS? (Speech2txt, large language model and txt2speech). I mean considering recent strides forward in robotics, there is no job that cannot be replaced anymore. This year it is estimated that a total of 25% of all jobs worldwide are now obsolete due to ChatGPT4.
What I have messed around with so far are mostly Txt2Image, txt2video and LLMs locally and even 7 Billion param models blew me out of the water when my personal AI assistant went life and was able to control my PC, execute programs, open the browser via extension and even go to youtube and search for a specified video. All through a single voice command I gave it.
With personal AI assistants now on the verge, the future looks very different than we could have imagined just a few years ago and the rate of change is truly exponential, smth many ppl struggle understanding.
What are you guys thoughts on this subject? Are we approaching Singularity or do you think we soon will hit a hard ceiling in our attempts to create an AGI?
Here some examples of technologies comming out now that completely remove the need for entire pipelines and therefore dozens of different jobs depending on each other:
This is not only limited to audio/vidual media and jobs in those industries. The moment you touch a keyboard at work is the moment you are basically already obsolete. But donāt worry, we all are sooner or later and that ālaterā is around the corner actually:
I can record my AIssistant in action once I landed in Formosa again. Am currently in the old world on family visit.
But let me give you a quick rundown how I assembled the whole thing within 5 days:
Whisper STT for transcribing my voice to text.
This text is then analyzed for keyworda triggering actions on my PC like opening folders, programs, tabs, etc.
Then I use llama.cpp and the brand new Mistral7B model for inference of that transcription (i created a mood score and different base prompts for different moods of my assistant. Sometimes mood switches randomly but 90% is driven by conversation flow and sentiment analysis of her and my comments. I also coded the base for longterm memory. The code amasses content from chat logs (last, recent and old logs 50%,25% and 25%), making sure she remembers last conversation and random memories from a week ago.
Her Answer then is sent to TTS and turned into an audible reply. I plan on hooking in a VC to change the voice to a cuter anime voice, since her visual representation is a cute VTube student that always stares at the mouse cursor and talks, when the LLM is replying.
Honestly, I thought deploying AI was harder, when I started, but this stuff is pretty straight forward in python.
Btw if anyone is located in Kaohsiung and want to dabble in AI, Iām totally down to it, hit me up.
True, but donāt you think the way we learn will change in the future?
Like a higher emphasis on selfeducation for older studentās and lessons on demand content generation?
Teacher friends in Germany already started using AI tools to create prep exam mockups and digital courses. I have the gutfeeling we steer towards classes on demand and separate social lessons (similar to kindergarden, with more social interaction and less knowledge driven teaching.)
How do you make sure students do their homework and donāt use GPT copypasta?
I donāt give them homework that can be done with GPT. They can use it to help them, whichbis fine because I do also
Iām using GPT for lots of things to help me with teaching, sure
Yes, soft skills are 21st century skills. Human teachers can for example model appropriate behavior
As with MOOCs, this requires motivation. The thing about students is, most of them donāt want to do the learning. One thing a good teacher does is motivate
I think I already teach in a way that is different from the average
Well, chatGPT was a great help. I am new to python myself but knowing other languages, it was quick to hop on. Point is, to setup a local LLM (I made a custom solution for myself, so you cabt compare that to what I have running on my mashine) is as hard as following some simple steps from a github page.
There is everything in a one-click-installer, even the speech support and txt2speech can be easily added via extension. Everyone can do it, really. (If you have a GPU worth using for that of course) 8GB VRAM recommended, but there are also 1.4b models that need way less. Worst case a good CPU will do aswell.
Hereās one point to add about education, building on what @TT has already contributed.
Before this recent leap in the availability of AI platforms, students readily had through their phones access to an astounding range of information and knowledge. Why then would they need things like schools, teachers, classmates? Itās because the sheer scale of information available is not quite the same thing as learningāa process that is incremental, social, and uneven.
Good schools and good teachers (along with good classmates!) can help this process along.
Will AI platforms be part of this helpful process or will itālike so many already existing commercial ventures such as social media sitesāinstead lead to more negative outcomes? Mental health problems, reduced attention spans, the spread of misinformation or disinformation . . . Will AI help or will it further the damage?
I generally feel exhausted about all the bandwagon-jumping as of late. Itās being forced down our throats by threads like this with the rhetoric similar to that of people that knock on our front door and try to change our religious and spiritual views.
I know people are complaining about losing writing work because of AI. Theyāre mainly generalist writers, and I think you really need to start positioning yourself as an expert to survive in the writing game.
Iāve also read a few complaints about some of the content mills shutting down (or closing a lot of profiles) because of AI. Content mills basically try to write articles and sell them to websites, but theyāre fairly generic and website owners are discovering that they can get AI to do it for free and just get an editor to check it over.
Iāve seen a lot of jobs on Upwork for editors to make AI stuff sound more human. The problem is that all of the AI detectors are crap and will often claim something is written by AI when itās not. I know that some writers have been complaining about clients claiming theyāre stuff is written by AI because of these detectors.
Well, itās the next big thing, supposedly bigger than the internet itself, which explains the hype.
Most places Iāve been, nobody even knows and talks about AI, so Iām surprised you say that itās being shoved down your throat. I rather have the impression that my surroundings are totally oblivious to these future developments.
I just wish people would use the right labels instead shoving everything under the AI banner because its the new buzz wordā¦ none of it is even AI anyway.
And I donāt see why I should be excited about even less chance of talking with an actual human when I need some customer service.
Not sure where you personally draw a line between neural networks and AI, but for me anything as capable as chatGPT4 definitely falls under the category AI. Not talkin about AGI here.
Anyway, more automation means less jobs and that will result in a different economy sooner or later. Possibly you will get all the time you want to spend with other human beings, if all the work is done by mashines.
My clients use AI for sports data analysis, brand name recognition, sports trading and odds compiling. What AI cannot do is what my niche business is and provide live sports events. AI cannot play the games. Yes it can do graphics. I was given a video by one of my support staff from a live NFL event. A kick at goal. One company using AI graphics was pretty fast but still several seconds behind my live feed. In live in play trading that is a big deal. Also a human can see the kick and already decide if the kick is going to be successful or not.
AI cannot predict a flag on the play or the penalty on the flag or if the penalty is declined.
Should be the case indeed, taxi drivers and bus drivers and delivery workers come to mind as large number of folks who could lose their jobs in the next few years.
New jobs will arise. The needs of the many will drive innovative and entrepreneurs will take advantage. Such is the always the case, isnāt it? Or are we too special for history to repeat itself?
Can you provide more details of what youāre talking about? I have no idea what the use case youāre describing is? Several seconds behind what, and why?
They canā¦ Not sure they can do it particularly well for close kicks (Scott Norwood?), or any better than computer vision. I suspect computer vision would be much, much better.
Sure it can. Question is how well - same as with a human predicting - and whatās āgood,ā given itās a non deterministic problem.
