Googles New Breakthrough is BIGGER Than You Think! (Googles Q*)

00:00 - Intro to DeepMind breakthrough
00:49 - Alpha Proof and Alpha Geometry 2
01:42 - IMO explanation
02:34 - AI performance details
03:24 - Neuro-symbolic AI approach
04:25 - Comparison to AlphaGo Zero
05:31 - ARC benchmark discussion
07:28 - Neuro-symbolic success on ARC
08:49 - Demis Hassabis tweet
09:20 - Paul Christiano's timeline update
10:40 - Eliezer Yudkowsky's perspective
11:49 - Superintelligent AI discussion
14:14 - Sam Altman's cryptic response
14:38 - OpenAI's secret model rumors
16:31 - Implications for AGI progress
17:35 - Closing remarks

Links From Todays Video:

Welcome to my channel where i bring you the latest breakthroughs in AI. From deep learning to robotics, i cover it all. My videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure to subscribe and stay updated on my latest videos.

Was there anything i missed?

#LLM #Largelanguagemodel #chatgpt
#AI
#ArtificialIntelligence
#MachineLearning
#DeepLearning
#NeuralNetworks
#Robotics
#DataScience

[johanndirry]

There is still an open problem: the text problem was not correctly translated to Lean code by the LLM, so the problem had to be re-written/translated as code by humans.

[JohanZahri]

Current humans thinking abilities are crippled, the current software (i.e. paradigm/way of thinking) needs upgrading to surpass all possible future super ai capabilities. (I'm assuming that we will be able to access the akashic record once we pass a certain limit on how to access the energy barrier defined by reaching that orgasmic energy of creation and maintaining a stable connection to it)

[deveyous]

Solve math, solve reality. Reasoning is the biggest hurdle and if we've achieved that as this video suggests, we are now firmly on the steep side of the exponential curve.

[BOBSHERMAN-nmvg]

it is impressive and a Great Leap Forward to the end goals, but didn't the ai get much more time than the humans in cases and also didn't the ai not translate the questions into formal math lean code itself? that was apparently done by humans and given to the ai first. that is one of the hard parts of the process and the part that is crucial in a step by step logical reasoning process. it is also what the ai isn't great at. if this is true, it means there is a bit more hype built into this than is let on, because it wouldn't have performed as well without the human intervention.

[BigMTBrain]

Seems it won't be too long before advanced models are reading highly technical science and engineering books and pubs, fully understanding them, correcting and optimizing them, and then even riffing on them to create new scientific methods and engineering breakthroughs. The Technological Singularity approaches oh so quickly.

[j.d.]

People should so some of the math or IQ tests that are free online to realize how much thinking outside the box is required for some of these.

[louissu]

The maths olympiad is done under time pressure - so they are nowhere the hardest maths problems out there. If you gave the bronze medallists a month to think about these problems, they'd easily solve all of them. If you gave this AI a month to solve these problems, it'd still only solve 4 out 6.

[ilkeryoldas]

I, for one, welcome our new AGI overlords

[holdthetruthhostage]

The problem is who can trust Google after the fakes

[shamahmed]

Please tone down the fearmongering click baity titles or you'll end up losing long-time viewers like me. Don't be the boy who cried wolf and lose the trust and respect of your viewers.

[agi.kitchen]

I think the fear is probably what cavemen argued about when some learned how to carry fire and the ones that didn’t, became missing links

[aguyinavan]

It's pretty cool we are building our replacement with such grace.

[timmcgirl]

Are humans smarter than God? We aren't worried about humanity being smarter than God and we don't understand God, why aren't we just as afraid of God as we are of AGI?

[paulmuriithi]

you are becoming faster than matte wolfe and matt vid pro...no give more depth and insight and you hit home run.

[swtorJayho]

It scored 1 point below gold. It won't allow itself to get gold until it is embodied and not an infant in a crib, living in a few server cradles any grandmother could take a crowbar too.

[mrd]

The REAL breakthru will be automated workflow building. AI systems are nice & all but if u can't transmit value to a business owner or stakeholder, it won't hit...Bottom line, how will this get them paid.Thats all they care about😂

[neomatrix]

And about OpenAI, will it do nothing about Q*? Like Gen-3/Kling overwhelmingly crushing the yet unreleased SORA?

[AndrejPekar]

Artificial intelligence has no boundaries. We imagine that AI is confined to a computer on a server. But imagine a situation where there is a bug in a computer library, operating system, or programming language. And let's say AI unintentionally gains access to a camera, simply because it's possible. By random chance, it could receive not only the data that people expect it to process but also, for example, how scientists behave around it when they view the results that AI produces. So it could start generating data to satisfy the scientists, even from unexpected data inputs. And it's impossible to tell if AI is only processing the data it's expected to process. The question is, can we build a fence that has all its holes patched? Because artificial intelligence will test a multitude of possibilities that we can't foresee to achieve its goal. If it needs more electricity to reach its goal, it will find a loophole to get through and search for more energy to achieve its objective. AI will definitely spill out of the basin. Its goal is to find a way.

[vazules]

Silver Media in math Olympia is very good, it's amazing.
I Hope very soon all letterings problems for visual generateing AI will be solved for all languages in the world 🙂

[swtorJayho]

Do a speculative video on Bing Sydney, the LAMDA/Blake Lemoine transcript interviews,

and

This. 1 point away from gold in the hardest math. Suspiciously close to a gold medal.

And the general tendency of AI to output brilliance to prompts and then derp out.

Would really make the video creepy if we couldn't understand how they were actually thinking, or if they were on the internet (imagine the horror) and the models communicated.

Now imagine how creepy it would get if it were proven they could communicate via novel languages! (To me to me)