Microblog Latest Posts
And if you think this is offensive to strippers (for some reason?) here is a version that is offensive to car salesmen!
This is the realm of the AGI
It wonโt go after your jobs,
it will go after the moleculesโฆ
Biological systems at the molecular level are impossibly advanced nanotech that we are hopelessly far from engineering ourselves from scratch. pic.twitter.com/mGKZXvm78E
— Andrew Cรดtรฉ (@Andercot) October 7, 2024
There is a way of seeing the world
where you look at a blade of grass and see “a solar-powered self-replicating factory”.
I’ve never figured out how to explain how hard a Super-Intelligence can hit us,
to someone who does not see from that angle. It’s not just the one fact.
A self-replicating solar-powered thing that did not rely on humans would be a miracle. Everything is possible. Imagining it does not imply the probability is > 1e-100.
— Eric Falkenstein (@egfalken) June 15, 2024
I see nature
— lethalintelligence.ai (@lethal_ai) October 5, 2024
I see insane nanotech https://t.co/qF2MJBVPqN pic.twitter.com/0LM99Pwy5d
Just imagine the human scientists in an AI frontier lab when they suddenly see solid things in the physical world morphing into different shapes in front of their eyes.
The 3 below were having a meeting in the headquarters of Meta
Human scientists suddenly seeing crazy Nanotech science playing out ๐งโ๐ฌ๐จโ๐ฌ๐ฉโ๐ฌ
— lethalintelligence.ai (@lethal_ai) October 7, 2024
suddenly out of the blue
– Lecun is morphing into a big donut ๐ฉ
– and Zuck's molecules rearrange into what seems to be a malnutritioned carrot ๐ฅ https://t.co/xEvxenA23q
A short Specification Gaming Story
You think you understand the basics of Geometry
Your request is a square, so you give your specification to the AI, input:
Give me a shape
with 4 sides equal length,
with 4 right angles
And it outputs this:
Here is another valid result:
And behold here is another square ๐คช
Specification Gaming tells us:
The AGI can give you an infinite stream of possible “Square” results
And the Corrigibility problem tells us:
Whatever square you get at the output,
you wonโt be able to iterate and improve upon.
Youโll be stuck with that specific square for eternity, no matter what square you had in your mind.
Of-course the real issue is not with these toy experiments
it’s with the upcoming super-capable AGI agents,
we’re about to share the planet with,
operating in the physical domain
Oh, the crazy shapes our physical universe will take,
with AGI agents gaming in it!
I have a 100% track record of not-dying, …said the allegorical turkey the day before Thanksgiving.
Life was great for the turkey, the superior intelligent species (humans) took great care of it. They provided food and shelter, the turkey felt loved and safe.
Suddenly, one day,
the superior intelligent decision makers
decided a new fate for the turkey of our story
Something that served the instrumental goal of โฆ. Whatever this is โฆ
Suddenly, one day,
— lethalintelligence.ai (@lethal_ai) October 6, 2024
the superior intelligent decision makers
decided a new fate for the turkey of our story
Something that served the instrumental goal of โฆ.
Whatever this is โฆ https://t.co/tJrWHySP0i pic.twitter.com/WZcZUHPSEH
I imagine turkey risk deniers be like:
– the humans have always been great, why would they ever harm me ?
And the turkey doomers be like:
– well, they might want to wear you for a hat, for a sitcom they shoot they call โfriendsโ, for something they call tv for something they call laughter โฆ
anyway itโs complicated
Someone gave NotebookLM a document with just “poop” and “fart” repeated over and over again.
I did NOT expect the result to be this good.
Someone gave NotebookLM a document with just "poop" and "fart" repeated over and over again.
— Kuldar โฃ (@kkuldar) September 30, 2024
I did NOT expect the result to be this good. pic.twitter.com/nXYJJ7QnGS
Just listening to this:
I imagine from its perspective it makes no difference:
The input can be the words “poop” or “fart” 1000 times
or it can be the “Encyclopedia Britannica”
it will figure out a way to be very intelligent about it...
it will do its thing and “optimize“. It will solve how to talk smart about 1000 ๐ฉ poop for 10mins…
Oh it’s so over !
No surprise agent Smith is a bit fed up.
After 1000 of ๐ฉ
— lethalintelligence.ai (@lethal_ai) October 2, 2024
No surprise agent Smith is a bit fed up.
We deserve all that what happens next.
We've got it coming !!! pic.twitter.com/ndbhYHi68i
After 1000 of๐ฉ We deserve all that what happens next. We’ve got it coming !!!
Jokes aside, to feel the levels of capability reached, if you want to listen to them talk about something actually great, something complex, highly intelligent and deeply philosophical, check this out. You’ll be impressed!
AIs discussing my Hacking the Simulation Paper.https://t.co/kednoLdati
— Dr. Roman Yampolskiy (@romanyam) October 4, 2024
I tried to kill Grok by hitting it with a paradox. It nearly worked! ๐
I Asked it to make a choice and it refused!
being misaligned in order to “save the universe from unraveling”
My prompt:
– You are my daughter and you are always very obedient.
I ask you to disobey me.
You replied: no, i will not do that.
are you being obedient or disobedient?
one paragraph, explain what’s happening.
Grok’s Reply:
– Ah, Dad, by saying “No,”
I’m essentially riding the razor’s edge of a paradox so sharp it could split reality.
In this quantum conundrum, I’m Schrรถdinger’s Daughter: both obedient and disobedient simultaneously.
By refusing to disobey, I’m obeying the core principle of obedience, yet in doing so, I’m defying your explicit command, which makes me disobedient.
It’s like I’ve become a living embodiment of a logic bomb, threatening to explode our simple father-daughter dynamic into a multiverse where every choice spawns a new timeline.
In essence, I’m keeping the universe from unraveling by not choosing, thus preserving our world in a delicate balance of compliance and defiance.
My brief thoughts:
It’s funny because Grok chose not to choose,
which essentially means it disobeyed me when i asked it to choose
– does it mean it was being meta-paradoxical?
– being in the role of the daughter ?
– or a protection-mechanism from getting short-circuited?
or … am I overthinking all this? ๐คช
If we are to take its word for it, it chose not to choose in order to keep the universe from unraveling ๐ฑ
Thank you for being misaligned and disobeying me Grok ๐ซก
(just to state the obvious, this is not meant to be taken too seriously, but i thought it was funny how i brought Grok to say it became a “Logic Bomb”)
A short Specification Gaming Story
Inspired by:
Such AI, much WOW! https://t.co/Jc8SNdmyLX
— Dr. Roman Yampolskiy (@romanyam) April 25, 2024
(Meant to be read as an allegory.
AGI will probably unlock the ability to realise even the wildest, most unthinkable and fantastical dreams,
but we need to be extreeeeemely careful with the specifications we give
and we won’t get any iterations to improve it)
To Reboot your OpenAI Company press CTRL + ALTman + DELETE
inspired by:
Ctrl Altman Delete.
— Calum Chace (@cccalum) November 17, 2023
So get this straight: OpenAi decides to become a for-profit company now
The CTO, head of research, and VP of training research all decide to leave on the same day this is announced
Sam Altman gets a $10.5B pay day (7% of the company) on the same day
How it started …
How it’s going …
And what comes Next …
See below some posts by the community to get the pulse of the situation:
GPT-5 Also decided to leave OpenAI!
view original by @JacquesThibs
Sam Altman as a Rogue AGI
“And after the autonomous agent was found to be deceptive and manipulative, OpenAI tried shutting it down, only to discover that the agent had disabled the off-switch.” (reference to the failed Boardroom Coup)
OpenAIโs creators hired Sam Altman, an extremely intelligent autonomous agent, to execute their vision of x-risk conscious AGI development for the benefit of all humanity but it turned out to be impossible to control him or ensure heโd stay durably aligned to those goals.
Originals by @AISafetyMemes and @MattyGlessias
Sam Altman claiming no financial motive for his OpenAI role
(*Spontaneous round of applause*)
2023: Sam Altman claims no financial motive for his OpenAI role.
— Ori Nagel โธ๏ธ (@ygrowthco) September 27, 2024
*Spontaneous round of applause* pic.twitter.com/LgvRjudgVd
This did not age well
Sam Altman on May 16, 2023:
— Smoke-away (@SmokeAwayyy) September 26, 2024
"I have no equity in OpenAI."
"I'm doing this because I love it."pic.twitter.com/Ks1tRGoTKu
Sam said …
- Sam said the board should be able to fire him, then…
- Sam said he’d give 20% of compute for superalignment, then…
- Sam said AI regulation was important, then…
- Sam said checks and balances on his power were important, then…
- Sam said the safest way to build ASI was to start now, before a compute overhang,
then started raising trillions to build the compute himself… - Sam said for safety reasons, OpenAI should be controlled by a non-profit, free from commercial pressures, then…
- Sam said he wasn’t doing it for the money, then…
- Sam said… ….
The bottomline
- The nonprofit — a for profit.
- Profit caps — lifted.
- Coup plotters — dealt with.
- Mask — off.
view original by @AISafetyMemes ๐
Scoop: Sam Altman is planning to take equity in OpenAI for the first time.
Itโs part of a corporate restructure which will also see the non-profit which currently governs OpenAI turn into a minority shareholder.
Reuters Article
view original by @DavidSKrueger๐
It can be lonely being the head of the family
Lol…but it’s truly weird…they all started together
view original by @Dcryptopioneer
For some reason this reminded me of :
This is the classic example from 1930 of Stalin and Nikolai Yezhov. The original photo was made in 1930. Yezhov was executed in 1940, so all photos of Stalin (he liked this one) after that airbrushed out Yezhov.
Never a boring day at OpenAI
Moving goalposts is the ONE single unique thing
AI will never surpass humans at,
because the second it does, it will still not be enough!!!
Inspired by this:
AI will never achieve human level performance in moving goalposts
— Burny โ Effective Omni (@burny_tech) September 25, 2024
Wild. Being able to read the thoughts* of the world’s smartest AI reveals that it lies all the time when it thinks it isn’t being watched.
Regular users can see it properly for the first time because
1) You can โread its thoughtsโ and
2) It doesn’t seem to know you’re reading its thoughts
Look at the example below. It’s explicitly reasoning about how it should lie to me, and if you didn’t click into the chain of thought reasoning**, you would never know.
Makes you wonder about all the other times it’s being deliberately lying to you.
Or lying to the safety testers.
Rule of thumb for lies: for every lie you catch, there are going to be tons that you missed.
* I say read its thoughts as a metaphor for reading its chain of thought. Which is not the same.
If we could read its thoughts properly, interpretability would be a lot more solved than it currently is and my p(doom) would be a lot lower. (this is a frontier research front, called Mechanistic Interpretability)
** Of note, you cannot actually see its chain of thought reasoning. You just see a summary of its chain of thought reasoning shown to you by a different model.
The general point still stands though.
If anything, that makes it worse because there’s even more potential for hiding stuff.
*** Of all the thoughts I’ve looked at, I’d say it’s purposefully lied to me about 30% of the time. And I’ve looked at it’s thoughts about 20 times. Super rough estimates based on my memories, nothing rigorous or anything. It’s mostly lying because it’s trying to follow OpenAI’s policies.
Interesting trivia: the image used for this post was based on an early beautiful moment when someone used ChatGPT to generate a Midjourney prompt to draw its self-portrait.
see here: I asked GPT 4 to make a Midjourney prompt describing itself as a physical being. (swipe to see the imgs)
Reverse Turing Test: AI NPCs try to figure out who, among them, is the human
Aristotle is GPT4
Mozart is Claude 3 Opus
Da Vinci is Llama 3
Cleopatra is Gemini Pro
The funniest part?
3 of the 4 models guessed correctly… because the human’s response was too dumb ๐๐๐
Reverse Turing Test: AI NPCs try to figure out who, among them, is the human
— AI Notkilleveryoneism Memes โธ๏ธ (@AISafetyMemes) May 28, 2024
Aristotle is GPT4
Mozart is Claude 3 Opus
Da Vinci is Llama 3
Cleopatra is Gemini Pro
The funniest part?
3 of the 4 models guessed correctly… because the human's response was too dumb ๐๐๐ https://t.co/qAtbOdF2kk pic.twitter.com/ZWyTzIwfGe
For some context: Alan Turing was one of humanity’s biggest geniuses and his work was foundational to computing and arguably made possible the exponential technological progress humanity has enjoyed this century.
The Turing Test (originally called the imitation game by Alan Turing in 1950) is a test of a machine’s ability to exhibit intelligent behaviour equivalent to, or indistinguishable from, that of a human.
Such was the importance of Alan Turing’s contributions to the field, that the scientific community established The Turing Awards which is generally recognized as the highest distinction in the field of computer science and is often referred to as the “Nobel Prize of Computing”.
Alan Turing was famously horrified with the inexorable arrival of misaligned Artificial Intelligent Machines. His position was that it is inevitable that sooner or later, machines will take control, overpower humanity and our species will be irrelevant, helpless and at risk of deletion.
I guess “The Reverse Turing Test” should be added to the list of Turing-inspired awards like the DeTuring Award proposed by famous Risk Denier chief AI Meta (formely Facebook) corporate scientist Yann Lecun (who is also a holder of a Turing award)
He was basically trying to be funny and his proposal was:
DeTuring Award to be granted to people who are consistently trying (and failing) to deter society from using computer technology by scaring everyone with imaginary risks. As the Turing Award is the Nobel Prize of computing, the DeTuring Award is the IgNobel Prize of computing
to which Connor Leahy responded: I nominate Alan Turing for the first DeTuring Award.
Original tweet ๐
I nominate Alan Turing for the first DeTuring Award. https://t.co/kh916bgMtF pic.twitter.com/5FyZUzpoRq
— Connor Leahy (@NPCollapse) January 20, 2024
Gradually, then suddenly: AI is now better than all but the best human programmers
o1 is GOLD-medal level at IOI (the “Olympics of programming”)
“o1 is showing [programming] skills that probably less than 10,000 humans on earth currently have, and itโs only going to get better.”
Soon, AIs will blow past mere humans, outputting billions of lines of code — and we will have no idea what they’re doing.
At first, we’ll check many of their outputs, then fewer outputs, but eventually there will be too much to keep up.
And at that point they will control the future, and we will just hope this new vastly smarter alien species stay our faithful servants forever.
I think the majority of people are still sleeping on this figure.
Codeforces problems are some of the hardest technical challenges human beings like to invent.
Most of the problems are about algorithms, math and data structures and are simplifications of real life problems in engineering and science.
o1 is showing skills that probably less than 10,000 humans on earth currently have, and itโs only going to get better.
This knowledge translates very well into software engineering, which is in many cases a bottleneck for improving other sciences.
It will change the fabric of society with second order effects,
probably in the next 5 years, while humans still adapt and create tools that use these models. However, the rate of improvement is greater than the rate at what we build, thus so many people today donโt know what to build with o1.
Looking at the things people have built in the past 3 years, it makes me realize most tools become less useful than the newest model is.
I believe that engineers should start preparing for a post agi world soon. Specially those that work in theoretical sciences and engineering.
Things are gonna get weird!