I guess he’s not that glossy…
2 Likes
hahaha. of course!
One one level I get it; As thirsty as their behavior is, the odds of Sempai Elon seeing it (and approving of it) are significantly higher than anyone else with a fraction of his fame/wealth. You can probably get a like from Elon easier than you could from your local town’s weather person.
1 Like
Not sure if this is pay-walled or not.
6 of 10.
Not sure how Google feels about the AI generated speaker batch? Assuming this was not in the prompt.
In real life this could become a PR problem if used in wrong context.
9/10. (SPOILERS!) I can say that I generally always have my phone in black and white now and that could have been helpful here. But my thoughts were more based on understanding current ai limitations than explicit uncanny valley stuff (which I scary). The runway model is filled with completely dead people on all the sides. The video game example is a now ubiquitous AI transition morph Blech boring thing. In the basketball player video no chance that AI would have created people playing those instruments in any plausible way (which they were) and not holding drumsticks upside down, etc. the demolition is the one I missed because it looked real but just seemed incredibly unsafe how the buildings on the left were toppling directionally and not imploding straight down. The toys r us commercial is an abomination. The influencers eyes were uncanny and I could see the jaggies from where it had been downrez’d for the ai and then uprez’d again. The google speaker had that glossy unnatural skin and I could see artifacts in the depth of field and inconsistencies in that defocus that a camera wouldn’t produce. I already knew the helicopter illusion was possible with shutter speed sync. Bike rider had exceptional temporal stability in the ground immediately surrounding him and was simply too consistent in movement and relation to gravity and groundedness to be ai which brings me to the jet ski which has the floating quality and uncanny physics of most ai videos.
Same here with the implosion. I figured they would fall straight down. I missed the influencer. With the jetski, the tell for me was the inability for the AI to properly distinguish the differences between a jet ski and a snowmobile.
1 Like
It also depends on whether you step through frame by frame on a decent monitor, or a fast scrub on your iPhone (I was waiting on someone at a coffee shop).
A trained eye with the right tools will get higher score. But that’s not what targets of mis-information campaigns or casual viewers on the subway will do.
2 Likes
This is the peak of human technology
Going back to what this all costs…
Interesting breakdown of the CapEx, OpEx, and Time it took to train Llama (Meta’s top of the line LLM), and what the just completed xAI cluster could do.
Summary:
- Llama 3.1, current state of the art model, required 3.8 10^25 FLOPS to train over 42 days.
- It was trained on a cluster consisting of 16,000 H100 GPUs.
- Hardware cost: around $1B ($480M for GPUs, $480M for the rest of the data center)
- Operational cost (Electricity): estimated at $2.41M
Now compare this with Elon Musk’s new xAI cluster:
- 100,000 H100 GPUs
- Hardware cost: Somewhere between $4-6B
- Cost to run cluster for 100 day training run for a next step function above Llama): $27.9M, using about 33.6M kWh.
This is all for training, not using inference during using a model. There had been previous reporting about the per query cost for AI based search. There was less discussion about the cost of the initial training, which has to be amortized across all the users.
And this training cost is just the data center operation. That doesn’t include the engineering to develop the model and code, and it also doesn’t include the collection and tagging of the training data which is another huge lift (think people creating CopyCat training pairs).
All to say, this is an entirely different league of what we’ve done in recent history in terms of compute scale. This AI better be making man kind a whole lot better off (for everyone, not just the tech bros), or this will be one gigantic FLOP (no pun intended).
And it better lead to some answers to climate change to offset the gigantic carbon pile it’s leaving behind.
2 Likes
The dude that thinks humanity’s best chance is nihilistically giving up on Earth and heading to Mars with what no doubt would be a group of fellow elite billionaires accompanied by however many plebs are required for each to maintain their opulent existences even while on the dead red planet isn’t going to be the person that makes mankind a whole lot better off IMHO
9 Likes
Speaking of Mars (and rich tech shits), I read this yesterday and enjoyed it: Neither Elon Musk Nor Anybody Else Will Ever Colonize Mars
Life on earth writ large, the grand network of life, is a greater and more dynamic terraforming engine than any person could ever conceive. It has been operating ceaselessly for several billions of years. It has not yet terraformed the South Pole or the summit of Mount Everest.
4 Likes
elon is making a shiny object so that the other rich people will leave earth…
Ark Fleet Ship B
3 Likes
So long and thanks for all the fish!
1 Like
that was a very enjoyable read. thank you.
TwAIlight
Hunger GAImes
John WAIcke
Thats massive news for the film and tv industry. Pretty game over now. The studios have picked a side.