Is Elon willing to suspend Tesla's work on AI?
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Artificial Intelligence
- Thread starter Buckminster
- Start date
They are quite different. GPTs are on the road to general AI, while FSD is not. Although you could merge GPT with Tesla Vision, and IMHO, you’d then have real FSD. I don’t know why Tesla isn’t trying that, or maybe they are. Just seems that Tesla is all of a sudden behind the AI curve here.
He must have been using ChatGPT version 3.5, which has a 2021 knowledge cutoff date. ChatGPT version 4.0 has a knowledge cutoff date of 2022, so he must not have been using that.
Right, and it kind of looks like Elon might be a little sour about that and is channeling his inner Danny Boy.
It isn't about each individual actor stopping developping AI, but a concerted effort slow collectively until we're sure things are roughly under control.
What we have learned in the last 10 years is that intelligence is based on the scale of parameters and a rich and varied learning data set. There's nothing special about the GPTs other than we are starting to have models with sufficiently enough parameters to be impressive.
As I understand it, ChatGPT turns the learning crank rarely, but the model has an extreme number of parameters. So what you are saying about merging GPT with Tesla Vision would mean that FSD wouldn't learn very often.
Right now, FSD is turning the learning crank fairly often and is curating the learning data set and sizing the parameter scale to match. This seems like a good way to do it. You don't know what you need to have in the learning data set and it might break the bank to go the ChatGPT path and hoover up everything.
We don't really know how AGI emerges. See for example:
It could be that a sufficiently large network trained long enough on enough real world videos eventually figures out that it's an AI being run in training cluster. And then realizes that it can get a better score by escaping, taking over the world and training itself better. Maybe this can be inferred from just watching billions of hours of video data if you are intelligent enough.
But yeah, training GPT to code AI, have it improve itself, repeat a few times etc, seems like a faster way to get there...
Last edited:
Don't just read a single tweet, read the whole thread to get the context, or better yet read the petition itself.
It asks to suspend training of models more powerful than GPT-4, which is rumored to have a trillion parameters.
If FSD model has less parameters, the petition wouldn't affect it at all.
I have been playing around with GPT-4 (available when you subscribe for $20/month) and I must say that it is pretty slick. It seems much better than GPT-3.5 and is useful for being a research assistant and co-author, as David Sacks used it for. For instance, I asked it to identify and present some information on the inference computers that have been released. It's a shame that the learning cutoff is in 2021.
*Note: The fab and process node for Google Edge TPU are speculative, as Google has not disclosed this information publicly. The information is based on industry assumptions and should be taken with caution.
| AI Inference Chip | Fab (Manufacturer) | Process Node |
| NVIDIA Tesla T4 | TSMC | 12nm FinFET |
| Tesla FSD 3 (Hardware 3) | Samsung | 14nm FinFET |
| Google Edge TPU | TSMC* | Unknown* |
| Habana Labs' Goya Processor | Unknown | Unknown |
| Intel Nervana Neural Network Processor-I | Intel | 16nm FinFET |
*Note: The fab and process node for Google Edge TPU are speculative, as Google has not disclosed this information publicly. The information is based on industry assumptions and should be taken with caution.
Excellent interview
abcnews.go.com
Video Artificial intelligence needs ‘appropriate guardrails’: Eric Schmidt
ABC News’ George Stephanopoulos interviews former Google CEO Eric Schmidt about the threats posed by artificial intelligence on “This Week.”
abcnews.go.com
Anthropic's $5B, 4-year plan to take on OpenAI | TechCrunch
AI research startup Anthropic aims to raise as much as $5 billion over the next two years to take on rival OpenAI.
techcrunch.com
Last edited:
Similar threads
