2024 and every year going forward will have more and better AI. This intelligence explosion is attracting funding and resources exponentially and it revenue will go exponential.
AI expert, Dr Alan Thompson, expects OpenAI to release GPT 4.5 in days and it will exceed human experts in most intelligence tests.
A survey of thousands of AI experts does not expect AI to be capable of surpassing human in all tasks until about 2047 (13 years earlier than last years survey) and they do not expect AI to replace all human jobs until 2100-2150. However, most of them expect AI to accelerate technological develop by ten times within 5 years. How will technological advancement been rapidly accelerated from every month being a year of progress without complete world altering effects?
An Nvidia Expert, Jim Fan, Forecasts Synthetic Data Becoming Very Important
Synthetic data will be a significant research path in 2024. Many old techniques that were meant for board games and image generation are now revived with new meanings for LLMs.
This paper’s idea is simple: self-play + GAN. Intuitively, an LLM at iteration (t+1) plays an adversarial game against its older self at (t). (t+1) tries to tell a model’s output from real human’s output, while (t) tries to fool its opponent by generating text that looks as real as possible. Then the (t+1) parameters are copied to (t) and the self-improvement loop continues.
Most importantly, this method does not require human preference data. A supervised finetuning (SFT) dataset is enough to kick-start the self-play game.
I think this method is similar to GAIL in spirit. GAIL stands for “Generative Adversarial Imitation Learning”, which outperforms simple behavior cloning for control tasks like Robotics. It also runs this adversarial game that drives self-improvement, but with a physics simulator in the loop.
Synthetic data will be a significant research path in 2024. Many old techniques that were meant for board games and image generation are now revived with new meanings for LLMs.
This paper’s idea is simple: self-play + GAN. Intuitively, an LLM at iteration (t+1) plays an… pic.twitter.com/ycIE3FToOd
— Jim Fan (@DrJimFan) January 3, 2024
Happy New Year everyone! 2024 will be EPIC. We are riding the vertical part of the exponential curve. Every year from now on will be state of the art, by far! 👏 pic.twitter.com/nv8XnAv4Cc
— Jim Fan (@DrJimFan) December 31, 2023
Brian Wang is a Futurist Thought Leader and a popular Science blogger with 1 million readers per month. His blog Nextbigfuture.com is ranked #1 Science News Blog. It covers many disruptive technology and trends including Space, Robotics, Artificial Intelligence, Medicine, Anti-aging Biotechnology, and Nanotechnology.
Known for identifying cutting edge technologies, he is currently a Co-Founder of a startup and fundraiser for high potential early-stage companies. He is the Head of Research for Allocations for deep technology investments and an Angel Investor at Space Angels.
A frequent speaker at corporations, he has been a TEDx speaker, a Singularity University speaker and guest at numerous interviews for radio and podcasts. He is open to public speaking and advising engagements.
Did they somehow fix the ‘hallucination’ problem while I wasn’t looking?
If AI can surpass humans in every task by 2045-2050, why would companies wait until 2100-2150 to replace humans with AI ? All jobs will be lost in 2050, not 2150.
There is zero intelligence in anything coming out of a LLM. They may appear to be intelligent but that’s just an illusion. To call this AI is just crazy. If they want to call it FI for Fake Intelligence that’s fine because that is what it is.
I think “simulated intelligence” would be more accurate. There may be (and it seems to be the case that there are) a lot of applications where a simulation of intelligence is good enough and the better the simulation the more applications meet that criteria.
So much money is invested into that, so the progress will be fast.
Just look at Space Race, how much faster did space exploration progress. But this field(AI) won’t be abandoned. To convince me, AI must demonstrate true intelligence, not just fetching data from some database and use most probable, best rated answer.
Chat-gpt is far from true intelligence for now. The question is how much is state of the AI more advanced, than we know it? What they have in labs is more advanced, no doubt about that. Their private versions are way better, than what is released.