Somewhat a fractured theory of Neural Network is,
Width helps u with variability in features within class
Depth helps u with learning commonalities and intricate differences of cross-class features
Width helps with knowledge, depth helps with generalization
It seems that many problems with modern ML, like the need for large data, model size are the culprit of cross entropy loss, and not backprop or SGD.
The remedy for most issues (generalization... , including explainablity) is a better loss fn with some architectural updates to realize the gain.
1
337
In addition, What if the ais hack and find dirt on critical people and blackmail that these people are afraid to turn them off and become immediate slave to Ai?
Suddenly Africa is the only continent that will function normally. @grok what is the chance of such scenario?
.@paulfchristiano thinks that on the current trajectory (absent any slowing down), there is a 1 in 3 chance that we could have superintelligence powerful enough to overthrow the government within a year. (He articulates a concrete scenario).
He argues that the labs should pause further algorithmic progress until they can secure their weights and research secrets. Otherwise China could simply steal the superintelligence in the middle of the intelligence explosion, and the labs would no longer be able to slow down even if they saw their models going off the rails.
He also explains how AIs will be trained during the intelligence explosion, and why even a system trained entirely inside a data center could still have the capabilities profile such that humans would have very little chance of competing with it or controlling it.
And he discusses the plan for alignment - basically, making a superintelligence that actually does what we ask the way we mean it. Fixing reward hacks and building better graders will help up till the next few model generations, but he thinks that no single current approach has better than even odds of letting us control superintelligence. The hope is that the AIs figure out the solution to this problem during the intelligence explosion.
Paul is head of the Alignment Research Center. Previously, he was head of AI safety at the US AI Safety Institute. Before that, he led alignment at OpenAI, where he invented RLHF. He’s recently joined OpenAI’s board (though here he is speaking as a private citizen).
1
99
At least, apple isn't full of scam
Tesla to wind down Cybertruck production after years of poor sales in few markets reut.rs/4zuYVVD
3
140
Why is everyone picking on her for being 20 when the captain is a 55 year old guy?
So far, no one knows if it's a pilot error not.
When max-8 fell, same was said about the pilots who did everything by the book to save the airplane.
The co-pilot of the Learjet that crashed into the ground in Florida has been identified as 20-year-old Nathalia Filgueiras, texted her boyfriend just 10 minutes before the crash.
Filgueiras' mother said she "always knew" a plane would "take her away from me."
The pilot was flying to the Bahamas when the plane plummeted 20,000 feet per minute into the ground.
"I wanted her to live her dream, but I knew that her dream was going to be the thing that would take her away from me," her mother told the New York Post.
"I knew that a plane was gonna take her away from me. I always knew it, but I was so proud of her. I know she was living her dream."
Just 10 minutes before the crash, Filgueiras reportedly texted her boyfriend: "I’ve got turbulence, I’ve got ice."
The female pilot got her pilot license in high school. Her stepfather says she had 1,300 hours of experience.
In a social media post, Filgueiras boasted about how young she was as a pilot with her boyfriend.
"How often do you see 2 pilots under 21 with a commercial, multi-engine, instrument, CFI/CFI-I," she said.
8 other people were on board the jet when it crashed. They have yet to be identified.
2
108
Math people are going crazy already btw xitter.cf/AmanBitz/status/208062…
people are starting to realize.
there *are* no moats left. the only true moat in existence was intelligence, and now that's out of the equation.
2
107
My friend and I joked one time that there will businesses that provide invasion as a service (IaaS), where you book an invasion of a place, pay and the location will be yours.
Sadly, this is on a fast track now.
Africa's safety is once again determind by some guy's moral level who lives like 6kms away.
2
2
345
This explains why we need exponential parameter count of ai models
Each coordinate of a closed outline has its own real Fourier series: x(t) = ∑ (aₖ cos kt + bₖ sin kt), and the same for y.
A cosine and sine of frequency k make one arm of length √(aₖ² + bₖ²) turning k times per lap. One chain sets x, one sets y, and the pen sits where their guides cross. With N harmonics per axis the knight is off by 30.2 % of its height at N = 1, 0.70 % at N = 50.
At N = 1 both coordinates are pure sinusoids: an ellipse.
2
139
Its unlikely that machines will have human like intelligence. But so many of the world will be automated around them that, human intelligence will simply be irrelevant.
I believe humans have the ultimate level of intelligence.
A well trained person can solve any problem within the training scope given the right set of information and reasonable time.
If all humans, trained in all aspects, collectively can communicate and have access to all information, then there can be no smarter being.
5
234
Grifters such as @projectmano tried to fool and bully Ethiopia to get into shitcoin gimmick and their example was elsavador.
That grifter himself is now moving away from it too.
El Salvador was the first country to make Bitcoin legal tender. Five years on, the experiment's done little to turn the cryptocurrency into a mainstream form of payment — so President Nayib Bukele is turning to dollar-backed stablecoins. bloom.bg/3TcxyzP
📷: Marvin Recinos / Getty Images
4
261
JevBerta can also play chess, beating laya and tying with Kev, which is 3x bigger model.
My friend @surafel_kindu was testing JevBerta on 2048 game (that I didn't even know it could play), and it didn't disappoint. It's slow tho, which will be fixed in its next version
My friend @surafel_kindu was testing JevBerta on 2048 game (that I didn't even know it could play), and it didn't disappoint. It's slow tho, which will be fixed in its next version
JevBerta: Another Jev model
huggingface.co/leobitz/jev-b…
still work on progress and a bunch of interesting ideas
on the line. unlike jev, no RL, no noul or score any BS like that. just plain old deterministic encoder based zero shot classifier done right. check it out at @huggingface
JevBerta: Another Jev model
huggingface.co/leobitz/jev-b…
still work on progress and a bunch of interesting ideas
on the line. unlike jev, no RL, no noul or score any BS like that. just plain old deterministic encoder based zero shot classifier done right. check it out at @huggingface
One of the dumbest arguments by many Africans who try hide our laziness is "the west doesn't want and allow Africa to grow". Why is Dangote killing it then?
This quoted post is unavailable.
128
I feel like this move will result in the feared outcome and backfire.
It's better to advance, find out what could happen and push the world to get prepared.
Delaying would give others the chance to advance, make unexpected but bad decisions, an irreversible impact
I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we've had at OpenAI in recent weeks.
Committing to having independent evaluators with employee-like access is a great idea, and we will do the same. We'll have more to share soon.
121
Happy new year #Ethiopia!
113
The entire math department right now is like Lehman Brothers in 2008
We recognize that the rapid progress of AI in mathematics is disruptive. We're looking to engage with the math community more on the best way to integrate this technology and communicate its impacts. After considering concerns raised by members of the mathematics community, we’ve decided to withdraw OpenAI’s sponsorship of the Caltech Mathathon and have notified the organizers.
67






















