Its not very AI if it would process it as a human would.
Comment has been collapsed.
A human on a 2000 cal diet consumes as much energy as a light bulb. When they build a machine that can independently analyze enough data stored in itself to answer all the questions an elite polymath could (while consuming as much power as a light buld), then you can say humans are inefficient.
Comment has been collapsed.
Well, an AI president would be better than a corrupt one.
Comment has been collapsed.
What a nice one liner, Im stealing it for IRL purposes. Thanks😇
Comment has been collapsed.
its a good idea to have it on internet record that i too love and trust ai
Comment has been collapsed.
Will AI process data like a human brain?
ChatGPT
The processing of data by AI systems is fundamentally different from how the human brain processes information. While AI systems, including machine learning models, can analyze and make predictions based on large amounts of data, they do so using algorithms and mathematical computations.The human brain, on the other hand, processes information through a complex network of neurons that communicate with each other through electrochemical signals. It has the ability to perform tasks such as pattern recognition, abstract reasoning, and creative thinking, which are still challenging for AI systems to replicate fully.
While AI has made significant advancements in various domains, it is important to note that current AI technologies are limited in their ability to replicate the full range of human cognitive processes. However, researchers continue to explore and develop new approaches, such as neuromorphic computing, that aim to emulate certain aspects of the human brain's processing capabilities.
Comment has been collapsed.
No, never.
This is an article I found a number of years ago that I really like. It's about how we use metaphors to describe our experiences, and then get confused between the metaphor and reality. During the Greek times, when they were developing complex irrigation systems, they assumed the human body worked as a balance of liquids, the four humours. Later, during the enlightenment, they assumed human brains worked like the finely tuned bits of clockwork that were the peak of technology.
https://aeon.co/essays/your-brain-does-not-process-information-and-it-is-not-a-computer
Comment has been collapsed.
We still have no idea how a human brain processes anything so it's unlikely we can replicate it in a machine.
Also "AI" is the new "VR". Sounds cool and all and the hype won't die because it gives people with column inches to fill something to say but there's no more Artificial Intelligence at this point than there is Virtual Reality.
One is a chat bot, the other is a couple of tvs strapped to people's faces.
Comment has been collapsed.
No. While LLMs can generate text and simulate human conversations (in a convincing manner), it is not thinking or understanding anything like a human. Chatbots are not sentient nor intelligent, it is just us projecting anthropomorphism on a text generator...
Think of an LLM as a very sophisticated form of auto-complete. The reason LLMs are so impressive is because their training data is basically the entire internet.
Comment has been collapsed.
I wouldn't ask about "like human", but about "as efficiently and universally as humans". The critical mass has probably already been reached and AI development increases exponentially, so I would say it's more like 5-10 years (closer to 10, than 5, IMO) before AI becomes intellectually clearly superior to us... the question still stands, will it then become self-aware and fully self-guided, i..e. a separate being living alongside us... or without us (Eliezer Yudkowsky: Dangers of AI and the End of Human Civilization | Lex Fridman Podcast #368 - https://www.youtube.com/watch?v=AaTRHFaaPG8).
My bet is, that AI thinking "like humans" will probably shortly follow, but it will not be of our design, but it will be the AI itself designing an inferior to itself "human simulator"... unless it decides, that we are so outdated, that it's not worth to run such a simulation ;(.
Comment has been collapsed.
I recommend you to take "AI for Everyone" by Andrew Ng. AI/NN uses complex mathematics to simulate human brain neuron activations.More n more work is being done(and more complexities are being added) but NNs perceive data very differently from humans......and will never be able to perceive or process data like human beings.
Comment has been collapsed.
A lot of very good stuff about current AI developments you can find in podcasts on https://www.youtube.com/@lexfridman/videos. They are quite lenghty, but those are talks between top scientists in the field, which are still able to talk about this stuff in a very understandable way for an average viewer. For the beginning I would suggest 3 talks with Max Tegmark:
https://www.youtube.com/watch?v=Gi8LUnhP5yU
https://www.youtube.com/watch?v=RL4j4KPwNGM
https://www.youtube.com/watch?v=VcVfceTsD0A
Btw, Lex Fridman running the show is really incredible and there is a lot of different stuff to be found there... he at times talks with folks like Elon Musk and Gary Kasparov, but many of his guests, that I've never before heard about, are actually having even more interesting things to say. Generally, it's "let's use our scientific knowledge to guess, where is the world going to" trip.
Comment has been collapsed.
43 Comments - Last post 2 hours ago by Myklex
1 Comments - Last post 4 hours ago by lostsoul67
71 Comments - Last post 5 hours ago by lostsoul67
76 Comments - Last post 7 hours ago by Reidor
765 Comments - Last post 8 hours ago by grimfandango8888
43 Comments - Last post 9 hours ago by Qnemes
12 Comments - Last post 10 hours ago by orono
538 Comments - Last post 12 minutes ago by galiane
421 Comments - Last post 22 minutes ago by Gawrazan
37 Comments - Last post 27 minutes ago by yoko666
2,176 Comments - Last post 30 minutes ago by Eudoxia
6,339 Comments - Last post 37 minutes ago by Oppenh4imer
3 Comments - Last post 51 minutes ago by Calibr3
7 Comments - Last post 1 hour ago by BarbaricGenie
Just interesting. I'm not an expert but think either won't or will but not soon. With LLMs hype there seems to be a trend to tune models for very specific use cases and tuning is still up to human judges
Maybe someone can suggest beginner-friendly materials to read
Comment has been collapsed.