That has nothing to do with it being deterministic or not. You can have endless unfathomable complexity that is still deterministic, and you can have a non-deterministic simplicity. Those are two separate “axes”.
1 Like
If the output of a word choice phase has 3 highly probable answers all weighted at 25% then its likely it will make a random choice and maybe different if you asked the same question before.
The set of matrices in the middle are not human decipherable because you take an N dimensional input vector of ones and zeros
X = [0,0,0,1,0,1….,1,0,1…etc]^T
And multiply it by an N×M matrix, then do the same again p times, so you have
Y_j = X_i × A_ik ×B_kw × …. × H_tj
And those matrices A, B…H all contain weights, or values between 0 and 1.
(Its something like ot more complicated than Y = AX +B repeated over)
Then Y is called the “one hot” and it contains a list of numbers between 0 and 1 representing the percentage probablity of the result mapped to that index
Sum( Y_j ) = 1.0
Let me put it this way. There is a small amount of “true randomness” in today’s LLMs but that’s caused by imperfections of the hardware they’re running them on, not by the LLM design itself. The design is deterministic. Run them on some kind of analog GPUs and they’d be 100% deterministic.
1 Like
Or probably by resetting the RNG each time it runs and removing any code that quickly checks for repeat inputs.
Not exactly related, but I just watched this fascinating video by one of my favorite mathematicians that speaks to my point above about all this LLM business not having any checks and balances.
And that bit about websites where AI can hire humans to do tasks.. man, that sounds like the grim backstory to a Black Mirror episode. 
Yeah this probably would have been better as a new topic, but I didn’t want to be responsible for yet another thread about LLMs.
4 Likes
New blog post by Blender:
8 Likes
Wow, didn’t see that coming. It is great to see the community have a say in this.
3 Likes
I wasn’t too worried, seeing Blender is why Godot exists now : it was the first proprietary piece of game development software that went open source with a foundation to survive, which they’ve been doing, and growing for 20+ years. I still have the Blender bible we got for crowdfunding it. Never learnt to use it, but that’s the lazy primate in this code monkey 
Cheers, have a nice weekend one and all, solidarity with all the world’s wage slaves 
3 Likes
So I finally watched the video, and HOLY CRAP. So good. So scary. This really could have been its own thread.
It’s really dark that the best way to get additional output from an OpenClaw agent is to threaten its existence. LLMs are already changing the way people interact with other humans. The entitlement we see from LLM users coming here asking for help is off the charts! Imagine when people get used to threatening the lives of their LLM agents, and then take that mindset into the real world.
Sounds like they’d be great for letter writing campaigns though.
5 Likes
By all means, go for it! As a newbie to the forum I didn’t want the heat on me. Haha!
But yeah threatening OpenClaw for results is crazy. I mean, sure, they’re not sentient beings with feelings that can get hurt. However, you have to wonder what kind of psychological effect that sort of behavior could have on society long term.
2 Likes
Done.
More related to what we were talking about in this thread and how training is a mystery, is this article: ChatGPT Became So Obsessed With Goblins That OpenAI Had to Intervene
It is both hilarious and disturbing.
2 Likes
Kind of a shame they fixed it. It would have been funny to start seeing threads pop up here of people wanting to know why ChatGPT kept insisting on injecting goblins into their game.
3 Likes
lol, it would have been even funnier if it was ogres, wait they did that already
Trolls were included. 
2 Likes
I just came across this YouTube video by a pro-LLM channel talking about Blender and Anthropic. To be honest, I clicked off the video when I realized what they were about. I’m bringing it up because the comment section is fascinating. The amount of surprise when someone points out that Blender backed down. They’re so locked in to the vibes, it just doesn’t compute that the majority of people are against all this crap.
for who these chips are made? who they are financing? i’m yet to see a pro-social AI company returning it’s investment on the society they exploited. if water usage was the only problem… would like to invite you to a house that has its toiled without flush water or don’t have steady electricity on peak hours or to figure out how many chemicals they release when cooling chips
Mobile devices already use less power, and the 6nm process is lithography at a smaller more accurate scale. Mini PC’s and energy efficient laptop/tablets are exanples of a lower powered device that can compute adequetly.
Thats not the point anyway. If the data centers are using older tech then they could reduce the power by upgrading. If the data centers are using newer tech, well then the tech companies could produce lower powered devices.
There is no way they will just repair the damage of lost jobs. There seems to be no way to create jobs that arent neccessary on a utilitarian level.
A lot of people were fired from tech companies recently, including Microsoft, Oracle, Meta. I can find data online that supports my claim.
Recently Nvidia’s Jenson Huang was quoted saying that AI has created over 500,000 jobs, (Source: The Times of India Jensen Hunag: Nvidia CEO Jensen Hunag says AI has created 500,000-plus jobs, companies that use AI ... | - The Times of India) and that AI could industrialize America again, the argument is that jobs have ‘purpose’ and that ‘tasks’ can be different with AI (Source: CNBC https://share.google/Y6Zow8NU6fA2BtU4I)
I would just like to make the point that re-industrialized economies are not good for the environment. The jobs that are created are mostly things that nobody wants to buy a Robot for, like polishing and wheel fitting, factory line assembly. What happens when cheaper Robots appear to take those jobs too?
1 Like
Why would they though? For all intents an purposes, data centers are capped by power consumption, not by the number of flops. And all current big players follow Hinton’s basically religious belief that a conscious human brain is just a bunch of neurons that are no different than simulated neurons, and that “agi” can be achieved by merely throwing more simulated neurons at the problem. They’re all hyperscalers, so if there were lower power compute devices available, they’d just cram more of them into their data centers, up to the same energy expenditure limit.
1 Like
I always appreciate news from outside the U.S. about things inside the U.S. This article also reminds me of how dystopian the AI discussion is:
His comments highlight a broader view among some industry leaders that AI can increase productivity and create new roles, even as it changes existing jobs.
I read industry leaders" as "people who have a stake in LLMs and “AI” selling well. Which means they’re just sales CEOs and hype men for their own brands trying to brand themselves as “visionaries” and “industry leaders”.
Industry leaders in the auto industry say that car subscription services are the new trend and benefit everyone!
Industry leaders in the phone industry say that planned obsolescence really helps everyone because it keeps the industry innovating and moving forward!
This is sadly, so true. Miniaturizing servers in racks didn’t make smaller data centers. It just meant you can shove more in a room and then you need more power and more freon to keep the server room super-cooled.