We are building Cognitive Emulation, an alternative AI paradigm that is controllable, transparent, and safe. Get in touch at [email protected].
London
Joined June 2022
- Tweets152
- Following4
- Followers4K
- Likes168
Conjecture retweeted
This was a great podcast! Absolutely check it out!
Thanks so much for having me on, Dan!
Conjecture retweeted
We're kicking off day one of The Ashby Workshops with our emcees, @BriTreece21 and @AndrewFATHOM today in Middleburg, VA. We have a room full of incredible leaders across sectors (and the world!) diving into the complexities of AI and its profound impact on everything from governance to labor markets.
It's exciting to be among such a diverse group of experts coming together to explore scenario planning and innovative ways that AI can help shape a positive future. Stay tuned for insights as the conference unfolds 👀
Conjecture retweeted
Last year, OpenAI's chief lobbyist said that OpenAI is not aiming to build superintelligence.
Her boss, Sam Altman, is now bragging about how OpenAI is rushing to create superintelligence.
Conjecture retweeted
We have some exciting ideas about how to develop useful and reliable AI programs and are building the foundation for that.
If you’re interested in being an alpha tester and helping shape our direction, sign up now!
🚨Attention developers! We are looking for alpha testers for Tactics - a new type of AI framework to build powerful, safe, and reliable AI programs.
Come help us shape the future of AI: alpha.conjecture.dev/
Conjecture retweeted
We're very excited to announce that Unify is co-organising "The AI Showcase", in partnership with @AWSstartups, @TYTN_ai, and @OctopusVentures, in London on 28th November 🚀😍
Connect with London’s bustling AI community, learn from industry experts, and see live demos from the hottest AI startups in town (@boltzbit, @ConjectureAI, @getcontextai, @GlyphicAI, @PapercupAI, @XRAIGlass, @synativ, @NeuralSpace, @mysticdotai, @Quix_io, @turintechai)! 🤖🧑💻
Book your free place here 😁
aws-startup-lofts.com/emea/e…
See you there! ✌️
Conjecture retweeted
This was a great interview, thanks for having me on!
“If you make a system that is more competent than humanity… what do you think is going to happen?”
Chief executive of Conjecture, Connor Leahy, warns that if AI is not managed well, humanity could go extinct - “extremely literally”.
@Roolockwood | @NPCollapse
Conjecture retweeted
.@PatrickChristys from @GBNEWS spoke to me about the risks from AI.
Patrick started by asking a very sensible question: Why would AI systems want to deceive us?
I told him: Because it would be useful to for them to do so. It would help them achieve their goals, which is what they programmed to do.
We then discussed how AI could create biological weapons, I told him that we already have "the genetic sequence for small pox available online... you can download it right now and synthesize it in the lab", and, even worse, "Researchers from Canada have already successfully from scratch, synthesized an extinct relative of smallpox called horsepox!"
Patrick's response was refreshingly sane! He shouted: "Why!?" "This winds me up!" "What kind of nutter wants to go and revive this thing?" "We've all seen the films Connor, we know how this ends."
We do indeed know how this ends, but it doesn't have to end like in the movies! We can change the odds in our favour with real action and policy.
If you find yourself playing Russian Roulette with extinction-level risky technology: Stop playing!
Conjecture retweeted
Winning the debate was bittersweet, however.
Humanity still remains utterly unprepared for the risks posed by advanced, uncontrollable AGI. But all is not lost - we can still do this. But we need to fight.
The future isn't decided yet, but it will be soon.
Conjecture retweeted
If we build machines that are more capable than all humans, then the future will belong to them, not to us.
Everything else derives from this simple observation.
Thank you @cambridgeunion for inviting me to make the case that "This House Believes AI Is An Existential Threat."
Conjecture retweeted
.@ConjectureAI CEO @NPCollapse now talks about how the ‘Responsible’ Scaling of the AGI labs are anything but.
#AISummitTalks – at Islington, London
Conjecture retweeted
AI Summit Talks - Together with @ConjectureAI, we're proud to announce new speakers: investor Jaan Tallinn, co-founder of @cser, policy analyst Eva Behrens from ICFG, journalist @tomough, along with a special message by @RobertBuckland. More info: lu.ma/xp3l35k6
Conjecture retweeted
Our CEO Connor Leahy [@NPCollapse] gave his 'AGI in Sight' talk at @CogX_Festival.
He spoke about how AGI companies are in a multi-way standoff - racing towards the precipice.
But it doesn't have to be this way. We can just stop, and build the institutions for a good future.
Conjecture retweeted
There are only two times to react to an exponential: Too early, or too late.
A hopeful, better future is possible, but it's not easy to build, and we definitely don't get there by blind accelerationism.
We can, and need to, do better.
Our CEO Connor Leahy [@NPCollapse] gave his 'AGI in Sight' talk at @CogX_Festival.
He spoke about how AGI companies are in a multi-way standoff - racing towards the precipice.
But it doesn't have to be this way. We can just stop, and build the institutions for a good future.
Conjecture retweeted
Today in the Guardian, I said:
“The deep issue with AGI is not that it’s evil or has a specifically dangerous aspect that you need to take out. It’s the fact that it is competent. If you cannot control a competent, human-level AI then it is by definition dangerous."
It's crucial to stress this: an AI that is merely at "human-level" in terms of tasks that a person can do on a computer is already an enormous threat, unless we are very sure we can fully control it.
Human-level AI is not stable: it's AI that can already be used to automate further AI development.
It's AI that can already do all of those tasks with immediate advantages over humans: never sleep, get bored, have read every book every written, anything learned by one AI can instantly be transmitted to all copies, and it is trivially easy to create thousands of parallel copies.
This leads to tightening and accelerating feedback loops in AI development, with AI capabilities quickly zooming past humans and making them obsolete.
The risk does not come from any form of inherent "badness" of the AIs or their developers, it is simply the case that if you build something incredibly powerful, and you are not in control of it, you will cause disaster in proportion to the power, and AGI is extremely powerful.
And even if you don't think AIs would do these things autonomously, remember: companies (and open source developers) will use them for this, fully purposefully.
OpenAI's "Superalignment" plan is to build a human-level AI researcher (an AGI) and task it with solving an incredibly hard unsolved scientific problem, alignment of superintelligence.
Anthropic just raised 4 billion dollars to build AI models "10 times more capable than today’s most powerful AI", as their leaked pitch deck indicated in April.
Do you expect they'll just stop doing what they're doing once they get to that level? (If they even know when that happens...)
I do not, and they have not committed to any concrete threshold.
Conjecture retweeted
Together with@ConjectureAI, we will host an event in the runup to the first major global summit on AI safety, on Oct 10,@ConwayHall. How can policy and the upcoming Summit help to prevent risks? @romanyam @NPCollapse and others will discuss! Info, tickets: lu.ma/xp3l35k6
Conjecture retweeted
What's a Kill Switch? I explain more in this piece.
"Not a literal switch (although that would be funny…) but a system much like the ‘red telephone’ hotline that was set up in 1963 between Moscow and Washington D.C.. It would be less a physical line and more a set of safety procedures and protocols that allow us to shut down ongoing developments and deployments of advanced AI systems quickly, if it were to become necessary."
No-one should be scared of AI - they should be terrified, says AI expert... express.co.uk/comment/expres… #ai #computers #aiwillkillus
Conjecture retweeted
We’re in a world where top scientists, governments, and even AGI developers themselves acknowledge that achieving AGI could literally kill everyone on the planet. Yet, the current trajectory we’re on is one of unfettered AGI proliferation despite the risks being clear to everyone.
The other narrative, that some are pushing for, is one of AGI nationalism: where AI risk is framed as a matter of national security and competition, and countries back industrial champions to get to AGI as fast as possible to disempower everyone else.
Both of these are not stable paths towards a good future, and they will predictably lead to disaster.
We need a third way, where we build the institutions to be responsible stewards of technological progress to benefit us. This piece from @andreamiotti is a proposal for a stable future with powerful AI.
In @TIME Magazine, I make the case that we can prevent AI disaster much like we prevented nuclear catastrophe, by:
(1) Banning the development of AI systems above a certain threshold of computing power.
(2) Building a multilateral institution to house risky AI research: MAGIC (the Multilateral AGI Consortium).
We don’t let individuals or corporations develop nuclear weapons for private use, and we shouldn’t allow this to happen with dangerous, powerful AI.
The world has woken up to the extinction risk that powerful AI systems pose to humanity. An open secret among AI researchers for years - 2023 finally saw top scientists, governments, and CEOs of artificial general intelligence (AGI) companies themselves publicly acknowledge that achieving AGI could mean “lights out for all of us”.
Yet, while we’re staring at the prospect of our literal extinction, the trajectory we’re on is still one of apathy and defeatism.
AGI developers, despite acknowledging the risks, continue to race to build the technology as fast as possible. Any individual who can befriend a few billionaires or large corporations can just raise funds to set up their own AGI shop. Governments are starting to recognize the risk, but seem afraid to even dare use their own power to rein in companies. Individuals feel hopeless and powerless, doubtful of working together with their own governments to protect themselves.
It doesn’t have to be this way. Humanity has faced extinction threats in the past, and we have coordinated successfully to deal with them and survive. This wasn’t for free, it wasn’t achieved by being blind to the whims and forces of technology, but by humans taking control over their own fate.
Governments and thousands of talented individuals had to relentlessly work against all odds, even across the Iron Curtain, to control and stop the proliferation of nuclear weapons. And they succeeded! If they did not even try, we would not be here today.
We managed to not destroy ourselves with nuclear weapons, and we can secure our future again - but not if we remain idle. What we need to do now is to act, and build the institutions that can lead us to a stable future.
World leaders are finally calling for the establishment of an international institution to deal with the threat of AGI. In June, President Biden and Prime Minister Sunak discussed the idea of a ‘CERN’ or ‘IAEA’ for AI. The U.N. Secretary-General, Antonio Guterres thinks we need one, too.
We must place advanced AI development into the hands of a new global, trusted institution and create a safer future for everyone. This should be the end goal of the upcoming AI Safety Summit hosted by the U.K. in the Fall.
* * *
I propose the establishment of MAGIC (the Multilateral AGI Consortium). MAGIC would have exclusivity when it comes to the high-risk research and development of advanced AI. It would be illegal for other entities to independently pursue AGI development.
This would not affect the vast majority of AI research and development, and only focus on frontier, AGI-relevant research, similar to how we already deal with dangerous R&D with other technologies.
To make sure high-risk AI research remains secure and under strict oversight at MAGIC, a global moratorium on creation of AIs using more than a set amount of computing power be put in place. My preferred threshold is 10^24 FLOP used across a model’s training run. This is easy to monitor and enforce with existing technology, as it affects a very small number of projects and can be monitored at the cloud providers’ level.
This approach is similar to how we already deal with uranium internationally, the main resource used for nuclear weapons and energy.
Research breakthroughs done at MAGIC will only be shared with the outside world once proven demonstrably safe. The benefits that MAGIC unlocks from its research will be shared with the signatory countries, so that all countries that join in can share in the safe upsides of the technology.
The heroic work of our predecessors helped us avoid nuclear war by controlling nuclear development. As humanity faces a new global threat—uncontrolled artificial general intelligence (AGI)—we once again need to take action to secure our future.
Conjecture retweeted
In @TIME Magazine, I make the case that we can prevent AI disaster much like we prevented nuclear catastrophe, by:
(1) Banning the development of AI systems above a certain threshold of computing power.
(2) Building a multilateral institution to house risky AI research: MAGIC (the Multilateral AGI Consortium).
We don’t let individuals or corporations develop nuclear weapons for private use, and we shouldn’t allow this to happen with dangerous, powerful AI.
The world has woken up to the extinction risk that powerful AI systems pose to humanity. An open secret among AI researchers for years - 2023 finally saw top scientists, governments, and CEOs of artificial general intelligence (AGI) companies themselves publicly acknowledge that achieving AGI could mean “lights out for all of us”.
Yet, while we’re staring at the prospect of our literal extinction, the trajectory we’re on is still one of apathy and defeatism.
AGI developers, despite acknowledging the risks, continue to race to build the technology as fast as possible. Any individual who can befriend a few billionaires or large corporations can just raise funds to set up their own AGI shop. Governments are starting to recognize the risk, but seem afraid to even dare use their own power to rein in companies. Individuals feel hopeless and powerless, doubtful of working together with their own governments to protect themselves.
It doesn’t have to be this way. Humanity has faced extinction threats in the past, and we have coordinated successfully to deal with them and survive. This wasn’t for free, it wasn’t achieved by being blind to the whims and forces of technology, but by humans taking control over their own fate.
Governments and thousands of talented individuals had to relentlessly work against all odds, even across the Iron Curtain, to control and stop the proliferation of nuclear weapons. And they succeeded! If they did not even try, we would not be here today.
We managed to not destroy ourselves with nuclear weapons, and we can secure our future again - but not if we remain idle. What we need to do now is to act, and build the institutions that can lead us to a stable future.
World leaders are finally calling for the establishment of an international institution to deal with the threat of AGI. In June, President Biden and Prime Minister Sunak discussed the idea of a ‘CERN’ or ‘IAEA’ for AI. The U.N. Secretary-General, Antonio Guterres thinks we need one, too.
We must place advanced AI development into the hands of a new global, trusted institution and create a safer future for everyone. This should be the end goal of the upcoming AI Safety Summit hosted by the U.K. in the Fall.
* * *
I propose the establishment of MAGIC (the Multilateral AGI Consortium). MAGIC would have exclusivity when it comes to the high-risk research and development of advanced AI. It would be illegal for other entities to independently pursue AGI development.
This would not affect the vast majority of AI research and development, and only focus on frontier, AGI-relevant research, similar to how we already deal with dangerous R&D with other technologies.
To make sure high-risk AI research remains secure and under strict oversight at MAGIC, a global moratorium on creation of AIs using more than a set amount of computing power be put in place. My preferred threshold is 10^24 FLOP used across a model’s training run. This is easy to monitor and enforce with existing technology, as it affects a very small number of projects and can be monitored at the cloud providers’ level.
This approach is similar to how we already deal with uranium internationally, the main resource used for nuclear weapons and energy.
Research breakthroughs done at MAGIC will only be shared with the outside world once proven demonstrably safe. The benefits that MAGIC unlocks from its research will be shared with the signatory countries, so that all countries that join in can share in the safe upsides of the technology.
The heroic work of our predecessors helped us avoid nuclear war by controlling nuclear development. As humanity faces a new global threat—uncontrolled artificial general intelligence (AGI)—we once again need to take action to secure our future.