- My Forums
- Tiger Rant
- LSU Recruiting
- SEC Rant
- Saints Talk
- Pelicans Talk
- More Sports Board
- Fantasy Sports
- Golf Board
- Soccer Board
- O-T Lounge
- Tech Board
- Home/Garden Board
- Outdoor Board
- Health/Fitness Board
- Movie/TV Board
- Book Board
- Music Board
- Political Talk
- Money Talk
- Fark Board
- Gaming Board
- Travel Board
- Food/Drink Board
- Ticket Exchange
- TD Help Board
Customize My Forums- View All Forums
- Show Left Links
- Topic Sort Options
- Trending Topics
- Recent Topics
- Active Topics
Started By
Message
re: U.S gov will get to decide who gets access to GPT-5.6
Posted on 6/27/26 at 9:57 am to trinidadtiger
Posted on 6/27/26 at 9:57 am to trinidadtiger
quote:
I think he is looking for funding for his NGO...
0
Posted on 6/27/26 at 10:18 am to Narax
quote:
They are also more diverse than you are willing to admit, we've seen huge differences in going from FP 16 to FP 4.
The diversity almost completely disappears when applied to LLMs though. That’s the link I’ve been trying to explain (and admittedly probably doing a poor job).
quote:
You say that but FP4 doesn't even use RISC, thats all tensor core. Tensor core is why the B200 is insane.
???
Maybe you know something I don’t but all the FP-4 I’ve tested still based their instructions off of RISC-V, or some other reduced instruction model. Does nvidia even support CISC at the chip level?
CUDA vs Tensor doesn’t really matter when it comes to the on board instruction set. So I’m confused as to what you’re referring here.
quote:
Look it's an interesting theory, and maybe there is something to it, but your statements are far too broad without data to back it up
From what I’ve seen to date, there’s a ‘there’ there. Whether or not it makes it into open literature, who knows.
Posted on 6/27/26 at 11:34 am to CFDoc
quote:
The diversity almost completely disappears when applied to LLMs though. That’s the link I’ve been trying to explain (and admittedly probably doing a poor job).
So that's cart before horse, the current belief is that the similarity of data removes the diversity of outputs. (sparse/highly clustered data, limited context tokens)
Your claim seems to be that there is something specific to GPUs that that reduces the diversity.
quote:
Maybe you know something I don’t but all the FP-4 I’ve tested still based their instructions off of RISC-V, or some other reduced instruction model. Does nvidia even support CISC at the chip level?
They don't use CISC, they use SASS as the instruction for the tensor core, it's fundamentally different from RISC-V.
They use RISC-V at the higher level for board level operations, which still has some impact, but under the hood where the actual training happens it's all SASS which keeps changing in each generation of Nvidia chip. That's what controls the actual instructions for training a model.
It's at the lowest level that the sheer volume of data (the transformer architecture) is what has been giving us the leaps and bounds improvement.
That's why think harder works so well, why Seedance in 4k is far better than at 780. That carries forward regardless of chipset.
That's the whole theory behind "Attention Is All You Need", and that carries forward into JEPA, these world models like Cosmos are just transformers carried forward.
quote:
CUDA vs Tensor doesn’t really matter when it comes to the on board instruction set. So I’m confused as to what you’re referring here.
I guess this is where I'm really confused with your claims that RISC-V is important as it really has very little to do with the underlying tensor processing of matrices.
quote:
From what I’ve seen to date, there’s a ‘there’ there. Whether or not it makes it into open literature, who knows.
If it doesn't then there was nothing there.
Posted on 6/27/26 at 12:31 pm to Narax
quote:
Your claim seems to be that there is something specific to GPUs that that reduces the diversity
Yes, that is precisely the claim.
And like I mentioned, from what I’ve seen to date, there’s exploits that work when running an LLM on a GPU that do not work when running the exact same LLM on another architecture. It is the very fundamental physical makeup of the GPU that allows for it.
And no, I’m not sharing IP with you on this topic.
quote:
They don't use CISC, they use SASS as the instruction for the tensor core, it's fundamentally different from RISC-V.
It's at the lowest level that the sheer volume of data (the transformer architecture) is what has been giving us the leaps and bounds improvement.
SASS is not the only instruction set for tensor cores.
My argument has nothing to do with improving LLMs.
At this point I think we’re just talking past each other. It’s prolly time to end the conversation and just go have a beer.
quote:
If it doesn't then there was nothing there.
This is categorically false.
Posted on 6/27/26 at 12:32 pm to hawgfaninc
That's bull
Posted on 6/27/26 at 2:49 pm to CFDoc
quote:
And like I mentioned, from what I’ve seen to date, there’s exploits that work when running an LLM on a GPU that do not work when running the exact same LLM on another architecture. It is the very fundamental physical makeup of the GPU that allows for it.
And no, I’m not sharing IP with you on this topic.
I've never asked, but this conversation has curiously gone from "it's easy" to "it's a secret".
quote:
SASS is not the only instruction set for tensor cores.
Yea, PTX works at the higher level, but RISC-V still isn't used on them, both are significantly different from RISC-V...
quote:
At this point I think we’re just talking past each other.
Curious.
quote:
This is categorically false.
Eventually things get published, impactful things don't hide in some dark chamber forever.
Even exploits eventually get patched and published.
Posted on 6/27/26 at 2:52 pm to mikesliveisacheater
quote:You're acting like this is an accident? It's not. Regulator capture is the only possible moat for AI companies.
Poor AI companies. They kept hyping that their AI is "too dangerous" and then don't understand what happened when someone in the government took them for their word.
Posted on 6/27/26 at 2:56 pm to CFDoc
quote:
And chipsets based on SIMT coupled with RISC-V controllers really really really don’t have free will.
Posted on 6/27/26 at 3:06 pm to hawgfaninc
99.9999% on this message board don't need more than Chat GPT 3.0
Not sure what the fuss is about
Not sure what the fuss is about
Posted on 6/27/26 at 3:34 pm to Narax
quote:
I've never asked, but this conversation has curiously gone from "it's easy" to "it's a secret".
So I went back and read my original posts and I’ll own up to using poor wording when I said it was ‘easy.’
My intent was to describe how deterministic and low dimensional spaces are ‘easy’ to exploit. More of a generality.
At this point, you are correct that it is not easy to exploit a GPU. What I am seeing is progress in research that essentially breaks a GPU down into a low dimensional, deterministic space when operated in transformer architectures. This is news since, for some time now, the general consensus was more along the lines that LLMs were uncontrollable stochastic dart throwers.
Maybe the details will become more public in the near future. But for now, the cybersecurity firm I’m doing work for requires otherwise.
Posted on 6/27/26 at 3:43 pm to CFDoc
quote:
That was just a precursor and that is full root access to almost every single Linux distribution from the last 10 years all wrapped up in a 700 byte Python file.
How is it even possible to defend against AI finding weakspots 24/7 with no need to rest?
It’s terrifying.
On a plus side, maybe we will see the Indian and Paki scam calls and emails become obsolete.
I’m almost afraid to click anything in an email any longer as it is. I report the hell out of shite at work. We get targeted an inordinate amount of times it seems.
Maybe that’s happening to everyone.
Posted on 6/27/26 at 4:15 pm to hawgfaninc
Frontier AI will be reserved for use by select few organizations. As F-AI evolves, older models will be shared outward. AI models are doubling in capability about every 6 months right now.
This post was edited on 6/28/26 at 5:14 am
Posted on 6/27/26 at 9:05 pm to dalefla
quote:
Frontier AI will be reserved for use by select few organizations.
People are going to assume all sorts of things as to why the AI future is going to be highly controlled and limited. In reality, the math (cost and energy) required to build and maintain AI Frontier models is insane.
The Efficient Compute Frontier for AI roughly estimates that we are within about 10-15 years of a single Frontier AI model (either JEPA or GPT based) requiring roughly a Terawatt to build/maintain.
For reference, the entire Earth requires about 18 Terawatts of continuous power. So 20 AI companies in 10 years could easily require an entire Earth’s power grid worth of energy to operate.
All that to say, AI is going to be highly controlled because there’s no form of government or political system that is going to allow a product to operate at that scale without massive control parameters.
On top of that, you can throw all the other stuff like security, control, and whatever other socioeconomic theory you can come up with.
Page 3 of 3
Back to top
Follow TigerDroppings for LSU Football News