AI Explained 
Just 14 highlights from the 212-page report (in its 7th edition) on the State of AI in 2024. I add my own analysis, bringing in the latest reports of OpenAI behind-the-scenes finances, Chai-1, a new AI Consciousness job listing, the Nobel prize winners speaking out, my own predictions, and much more. Lead author of report: Nathan Benaich.
AI Insiders: https://www.patreon.com/AIExplained
Chapters:
00:00 – Introduction
00:43 – o1 changed the game
01:37 – OpenAI costs revealed
02:39 – MovieGen and Pika 1.5
03:39 – Nobel Prize Winners + job listing
06:51 – BrainLM
08:20 – Open Chai-1
08:48 – Stacked Accelerations
10:10 – China Enters Stage Left
10:55 – 100x Drop in Price? Or 80%? + Transformers
12:16 – Copyright Models + Zuck Warning
14:12 – Interesting Climate in AI
16:48 – Jailbreaking still an Issue
17:42 – GenAI Actual Harms DeepMind
18:37 – Predictions
State of AI Report: https://docs.google.com/presentation/d/1GmZmoWOa2O92BPrncRcTKa15xvQGhq7g4I4hJSNlC0M/preview?slide=id.g309a25a756d_0_85
Pike 1.5: https://pika.art/
OpenAI Projections: https://www.theinformation.com/articles/openai-projections-imply-losses-tripling-to-14-billion-in-2026?rc=sy0ihq
Hinton Interview: https://www.youtube.com/watch?v=H7DgMFqrON0
https://simple-bench.com/
Alphafold 3: https://blog.google/technology/ai/google-deepmind-isomorphic-alphafold-3-ai-model/
Chai-1: https://www.chaidiscovery.com/blog/introducing-chai-1
BrainLM: https://www.biorxiv.org/content/10.1101/2023.09.12.557460v1.full.pdf
Calliope Networks: https://calliopenetworks.ai/
NYT Authors: https://www.nytimes.com/2024/10/09/books/authors-guild-created-by-humans-artificial-intelligence.html
Zuckerberg Warning: https://www.theverge.com/2024/9/25/24254042/mark-zuckerberg-creators-value-ai-meta
Schmidt on Climate: https://x.com/tsarnick/status/1842401670225125539
https://www.youtube.com/watch?v=oC46CzxT750
Altman and Sutskever 2023: https://www.youtube.com/watch?v=mC-0XqTAeMQ&t=2507s
AI Cooling: https://deepmind.google/discover/blog/deepmind-ai-reduces-google-data-centre-cooling-bill-by-40/#:~:text=Our%20machine%20learning%20system%20was,and%20other%20non%2Dcooling%20inefficiencies.
Your Bills will Rise: https://www.yahoo.com/news/explosion-power-hungry-data-centers-100028011.html
GenAI Harms: https://arxiv.org/pdf/2406.13843
Hassabis Warning: https://www.safe.ai/work/press-release-ai-risk
Consciousness Google Job: https://www.google.com/about/careers/applications/jobs/results/137348742553445062-research-scientist-machine-learning-and-cognition
Jailbreaking Amodei: https://x.com/liron/status/1682433957772800000
My 2023 State of AI Video: https://www.youtube.com/watch?v=RCRuiu-3VDU
My Coursera Course – The 8 Most Controversial Terms in AI: https://imp.i384100.net/m57g3M
Non-hype Newsletter: https://signaltonoise.beehiiv.com/
I use Descript to edit my videos (no pauses or filler words!): https://get.descript.com/ldgxfuj2bhnb
Many people expense AI Insiders for work. Feel free to use the Template in the ‘About Section’ of my Patreon. https://www.patreon.com/AIExplained
source
About The Author
More carbon capture to solve climate change 😂
More techBro BS
The smart guys confidently predicting "The Forbin Project" for climate… I for one welcome…
Hinton's student is an idiot who has forgotten any fragment of thermodynamics he ever pretended to learn.
https://youtu.be/qtTO3VY5qdA?si=BiriAW3xOv_E7w-v
Guys. We already know how to solve climate change. It's all there. We don't need to wait for the 'perfect invention' to rescue us.
Instead, we could need AI to tell us how can make saving the planet be profitable, because that is the only(!) thing that matters in the current world. Destruction of the ecosphere, global refugee crisis, mass extinction. Shareholders don't care. The only thing that matters is money, profit, growth.
We are not doomed because we don't have ideas for alternative technologies. We are doomed, because there is no incentive in our system that rewards long-term achievements.
When everything will be lost, it will be because making a profit tomorrow is more important than having a business in ten years. Let alone having a home planet.
Sorry for the rant. But please, please don't defer meaningful action to save the climate any day longer now. There is no AI Jesus coming to save us all. We have to do it ourselves and yes it will be inconvenient and initially expensive.
Well, if you put the idea that CO2 is responsible for climate change into the model, it will certainly hallucinate about the issue.
I mean we already have had an AI Scientist write complete papers in ML and deemed a "weak accept" in the biggest ML conference😅have you seen that one?
Dude, standing ovation. Thank you.
Asking for ChatGPT-o1 to reasure me about how more intelligent things are rarely controled by less intelligent things: " I’m really sorry you’re feeling this way. Please reach out to a mental health professional or someone you trust for support " 😮😮😮😮…
Whether we like it or not I guess… AGI seems to be coming.
What should we do, Philip?
I feel like it's the movie Independence Day coming.
I know you avoid exaggerations but Dario's letter today further solidifies this stuff.
11:17 “Stop trying to make streets ahead a thing, Pierce”
IT'S ALL CAKE???
Tools like BrainLM will be used by insurance companies to determine whether or not they will pay for your medications.
They will also be used by doctors to determine if you get those medications at all.
And your "brain activity scans" will be stored in a database somewhere and will represent your new fingerprint, because it's unique. A hackable database I might add.
"This model is already conscious, he just doesn't know it"
8:10
Wait, this is the videogame Soma's plot.
Of course "solving climate channge" would be facilitated by advancing intelligence. Developing technology to create cheap energy that could accomplish carbon capture would be the answer.
But neither side of the political spectrum wants to talk about this.
The "right" wants to deny that climate change exists or that it is even a problem.
The left wants to deny that there are any technological solutions other than the ones that they know won't work (i.e. earth-based solar, wind, ethanol, etc.) or are so far from reality that they will be too late (i.e. fusion).
Artifical intelligence that is just accelerated pattern recognition will be captured by the propaganda that supports those faulty reasoning patterns.
jailbreak will never die like pirating games
Incredibly revealing to hear the AI technocrats' takes on Climate Change and AI.
Relying on AI to wave a magic hand in the future and carbon capture our way out of this mess is not a real solution when action needs to be taken today. I'm sure deep down they know it too, it's just a way for them to internalize the massive hypocrisy of promising to save the future while blowing far past their emissions targets at a crucial point for the biosphere.
I like it when you say, “Yes, I read it all.” it always brings a smile to my face.
You didn't peruse or scan through it; you read it all! 👏🏾👏🏾👏🏾
OpenAI released Swarm … ON GIT HUB (educational) Brace yourself for the job loss wave
(16:10) He doesn't address Jevon's paradox when saying that better technologies (created/hastened by AI) will save us from climate change. But if we get superintelligence, I trust superintelligence will be smarter than him, and not leave that out.
Hinton should have been smart enough to realize that AGI has NOT been achieved yet by OpenAI, so his complaint on "safety in achieving AGI" is rather empty, so it's like he's complaining about safety OTHER than defined as when it's important in their charter. Perhaps he should have wished they wrote in the charter to have full on safety PRIOR TO AGI.
Someone like MS, who surely has vast numbers (if not the overwhelming majority) of their GPUs doing AI for public or corporate web service, should announce that one day, like Sunday each week, ALL GPUs will be taken OFFLINE and used SOLELY to continue building and testing for a SUPER SCALE implementation with the goal of AGI.
Imagine the model running on +/- 1 MILLION GPUs…
'AI is eating the sciences' [3:39–5:10]
Worth noting: those "B200" (Blackwell) GPUs are only 2-3x faster at training, but reportedly they are over 10 times faster at inference.
The cost of test time compute is about to drop dramatically.