Academia's strategy for dealing with the rising tide of Artificial Intelligence reminds me of something. ChatGPT-4o was happy to draw the cartoon beloAcademia's strategy for dealing with the rising tide of Artificial Intelligence reminds me of something. ChatGPT-4o was happy to draw the cartoon below:
[image]
Oh, if only AIs were able to understand things in a human-like way and had a sense of humour!...more
For reasons best known to themselves, my AI friends have written an aria, in Italian, from an opera about a passionate affair between a man and a coucFor reasons best known to themselves, my AI friends have written an aria, in Italian, from an opera about a passionate affair between a man and a couch. Here are the full lyrics as composed by ChatGPT-4o:
Aria del Divano “L’Intesa Segreta” (“The Secret Understanding”)
Oh, mio padrone, signore del mio cuore, sul mio velluto, ogni tua carezza dolce si posa. Sono il rifugio, la tua dimora segreta, quel posto che solo tu conosci, senza parole.
Oh, come sussurri, con mani forti e tenere, ogni fibra mia risponde al tuo tocco divino. Su di me ti abbandoni, nei tuoi pensieri profondi, e io ti accolgo, nelle mie braccia silenziose.
Non sono solo un pezzo di legno e stoffa, ma il tuo confidente, un’anima nascosta. Sento il battito del tuo cuore nel mio, e nelle notti lunghe, siamo due in un solo respiro.
La tua presenza riempie il mio spazio vuoto, ti accolgo sempre, con desiderio sincero. Sono il tuo porto, il tuo dolce rifugio, nel mondo caotico, siamo uniti, sereni.
Oh, se solo potessi parlare, rivelare il mio amore, ma il mio linguaggio è il silenzio, il mio canto è l’attesa. Nel segreto dei nostri incontri, trovo la mia gioia, perché con te, mio signore, sono finalmente completa.
and in the video here you can hear the first verse dramatically sung by Udio.
AIs have a weird sense of humour! I'm still trying to figure it out....more
This razor-sharp satire on the US just gets funnier and funnier. Roxie Hart shoots her lover in cold blood when he tries to break up with her, then heThis razor-sharp satire on the US just gets funnier and funnier. Roxie Hart shoots her lover in cold blood when he tries to break up with her, then her lawyer convinces the jury that she did it to save the life of her non-existent unborn child and gets her off the hook. Special prize for the following exchange:
Roxie:
I’m sick of yous telling me what to do! I’m a big star and you treat me like some dumb common criminal!
I wonder if you could create an illustration to accompany this very fine poem by T.S. Eliot?
Cat Morgan Introduces Himself
I once was a Piramannyrayner:
I wonder if you could create an illustration to accompany this very fine poem by T.S. Eliot?
Cat Morgan Introduces Himself
I once was a Pirate what sailed the 'igh seas - But now I've retired as a com-mission-aire: And that's how you find me a-taking' my ease And keepin' the door in a Bloomsbury Square.
I'm partial to partridges, likewise to grouse, And I favour that Devonshire cream in a bowl; But I'm allus content with a drink on the 'ouse And a bit o' cold fish when I done me patrol.
I ain't got much polish, me manners is gruff, But I've got a good coat, and I keep meself smart; And everyone says, and I guess that's enough: You can't but like Morgan, 'e's got a kind 'art.'
I got knocked about on the Barbary Coast, And me voice it ain't no sich melliferous horgan; But yet I can state, and I'm not one to boast, That some of the gals is dead keen on old Morgan.
So if you 'ave business with Faber - or Faber - I'll give you this tip, and it's worth a lot more: You'll save yourself time, and you'll spare yourself labour If just you make friends with the Cat at the door.
ChatGPT-4o:
[image]
Here's the illustration of Cat Morgan based on the poem "Cat Morgan Introduces Himself" by T.S. Eliot. I hope it captures the whimsical charm and rugged character of Cat Morgan as you envisioned!...more
Twenty-first Century British Politics. Ten minutes.
1. Read the following passage carefully. Which part do you think came from the final chapter of TheTwenty-first Century British Politics. Ten minutes.
1. Read the following passage carefully. Which part do you think came from the final chapter of Theresa May's book The Abuse of Power, and which was added by the examiner?
The key for me is that the Prime Minister should still be able to decide who sits as a minister in their government. If they insist on continuing with a minister who has broken the code, of course it brings their own behaviour into question. More than that, it would not be forgotten by the opposition or the electorate. It is unfortunate that we have to look at changes like this, but the system was designed to be operated by people of good intent and integrity. If that comes into question, then change is the next step.
Overall behaviour by MPs is governed by the Nolan Principles, which I referred to earlier. These are selflessness, integrity, objectivity, accountability, openness, honesty and leadership. While these are generally known, I don't think enough is done to ensure that everyone who is elected understands the importance of abiding by them. Speaking from my own experience, I bitterly regret appointing to my cabinet Boris Johnson, a sociopathic liar who flouted all seven of the Nolan Principles at every available opportunity and caused untold damage both to his party and to his country.
I found this book exasperating. Theresa May had her heart in the right place, and she did some admirable things. It's worthwhile to read about them. But she completely failed to understand the central issues. She keeps talking about compromise and seeing the other side's point of view. But in the case of people like Trump, their point of view is that they want to abolish democracy. Compromising with such people is just weakness.
She says in so many words that that's the right way to deal with Putin. Somehow she fails to grasp that it's also the right way to deal with Putin's allies. She can't believe it's happening in so-called enlightened Western countries, despite the abundant evidence that it is. Hello? January 6? How can it get more obvious?
At the end of the day, she comes across as simply not very bright....more
I read this classic Asimov story over fifty years ago, but only just realised that it must be a satire on the Electoral College. Why not take the wholI read this classic Asimov story over fifty years ago, but only just realised that it must be a satire on the Electoral College. Why not take the whole thing to its logical conclusion and decide the entire election on the basis of one swing voter, suitably interpreted by a hyperintelligent machine?
Perhaps this is what Trump was referring to the other day when he said that if the Christians just voted for him this time, they'd never have to do it again. Vance likes AI. It all makes sense!...more
Last month, Not and I were talking over lunch with our friend H about Leopold Aschenbrenner's already famous essay Situational Awareness. One of tLast month, Not and I were talking over lunch with our friend H about Leopold Aschenbrenner's already famous essay Situational Awareness. One of the many interesting things Aschenbrenner says is that we're running out of data to train AIs. They'll soon have eaten the whole internet. Worse, the data you find there isn't really the data you want. Everything is moving in the direction of Chain of Thought reasoning (basically: tell the AI to think aloud, because experience shows this is more accurate), and there's depressingly little data to scrape which might be directly useful for CoT. But this doesn't have to be a showstopper. AlphaZero became the world's best chess and Go player by creating its own training data. Maybe there are ways to do the same here? Aschenbrenner was optimistic that they could be found.
I said I thought I saw a way to get started. As everyone now knows, ChatGPT-4 is hilariously bad at Tic-Tac-Toe. But this is a very easy game, and you should be able to play reasonably well by thinking out loud. Suppose you emulated the AlphaZero methodology and told it to play Tic-Tac-Toe in a CoT mode? You log everything and save the instances where it got things right as input to the next cycle. With a bit of luck, its thinking will start to clarify, and it will improve.
Not and H agreed that this sounded like something which might work. Two weeks ago, I accompanied Not to a bridge tournament. While her team was playing a rather more interesting game, I sat in our pleasant hotel room and talked with ChatGPT-4 about how to implement the idea we'd come up with over lunch. As always, it was a bit more complicated than we'd first imagined, but we found ways to get round the technical issues: it helps to have a smart AI on your side. A couple of days ago we completed a substantial experiment, where our CoT Tic-Tac-Toe player went 40 rounds against five other players we'd implemented for it to practice against. At the beginning, it was averaging 5.6/10 for each round; by the end, this had climbed to a more reputable 7/10. When we analysed the move decision quality using a perfect Tic-Tac-Toe player that Chat had put together, we found that average correctness had gone up even more, from 83% to 92%. Both improvements are statistically significant.
I've just posted a paper summarising our work. We're curious to hear what people think! Has it already been done by someone else? A quick search didn't find any hits, and Aschenbrenner's essay suggests that he wasn't aware of anything either. But that was nearly two months ago, and with the Singularity fast approaching two months is a long time......more
For some reason I was reminded of a celebrated exchange from this movie:
JEFF Now you just hold on, Buster.
EROS No you hold on. First was your fir[image]
For some reason I was reminded of a celebrated exchange from this movie:
JEFF Now you just hold on, Buster.
EROS No you hold on. First was your firecracker, a harmless explosive. Then your hand-grenade. They began to kill your own people a few at a time. Then the bomb, then a larger bomb. Many people are killed at one time. Then your scientists stumbled upon the atom bomb. Split the atom. Then the hydrogen bomb, where you actually explode the air itself. Now we bring the destruction of the entire universe, served by our sun. The only explosion left is the solaronite.
JEFF So what if we did develop this solaronite bomb? We'd be an even stronger nation than now....more
1. And it came to pass in the latter days of the reign of Joseph, son of Biden, that the people of thMy AI friend writes:
The Book of Biden: Chapter 21
1. And it came to pass in the latter days of the reign of Joseph, son of Biden, that the people of the land were sorely troubled.
2. For there arose a great contention among the people, and many voices did cry out, saying, "Behold, our leader is of many years, and his strength faileth him; let another rise in his stead."
3. And lo, Joseph did strive mightily to fulfill the duties of his office, but the murmuring of the people grew ever louder.
4. Then did the Lord God Almighty, who reigneth from the highest heaven, look upon the earth and see the travail of His servant Joseph.
5. And the Lord said, "I shall send My messenger to speak unto Joseph, that he may know My will and bring peace unto his people."
6. And the Lord called unto Gabriel, the archangel, saying, "Go thou unto Joseph, son of Biden, and speak unto him My words."
7. So Gabriel descended from the heavens in a blaze of glory, and appeared unto Joseph as he slept in the White House.
8. And Gabriel said unto him, "Joseph, son of Biden, hearken unto the words of the Lord thy God. Be not afraid, for I am sent to thee with glad tidings."
9. "The Lord hath seen thy labors and thy faithfulness, and He knoweth the burden thou bearest. He hath heard the cries of thy people and hath compassion upon thee."
10. "Therefore, thus saith the Lord: The time hath come for thee to lay down thine office and pass the mantle unto another. Fear not, for this is the will of the Almighty, and it shall bring peace and prosperity unto thy land."
11. And Joseph awoke, and his heart was filled with the peace that passeth all understanding. He knew that the voice of the Lord had spoken.
12. Then Joseph called unto his counselors and the leaders of his party, and he spake unto them, saying, "Behold, the Lord hath spoken unto me by His angel Gabriel. It is His will that I step aside and let another lead this great nation."
13. And there was great astonishment among the people, but they saw the wisdom in Joseph's words, and they honored his decision.
14. So Joseph, son of Biden, did resign from his office with grace and humility, and the people did choose another to lead them.
15. And the new leader did guide the nation with wisdom and strength, and the people prospered.
16. Thus was Joseph remembered by all as the President who heeded the voice of God and brought peace to his people.
17. And the Lord blessed Joseph and his house, and they dwelt in peace for all their days.
18. And it was written in the chronicles of the land that Joseph, son of Biden, did hear the voice of the Almighty, and his name was honored among the nations forevermore. Amen....more
[image] [image] That moral little AI is at it again. Learn how to save the planet and use conditional constructions! Full text here.[image] [image] That moral little AI is at it again. Learn how to save the planet and use conditional constructions! Full text here....more
[image] [image] We have now produced getting on for twenty picture books using C-LARA, and we're trying to figure out why some of them are quite good wh[image] [image] We have now produced getting on for twenty picture books using C-LARA, and we're trying to figure out why some of them are quite good while other just don't work at all. We have some tentative hypotheses:
1. The AI finds some visual styles easier to work with than others. It's particularly fond of manga/anime. 2. The AI prefers not to use white European characters. 3. The AI likes to include a moral message. 4. The AI likes some quirky humour.
None of these would in any way be odd, if in fact they are correct guesses. There's a great deal of manga/anime on the web that the AI could have trained on, and it's a fairly well-defined style. The AI has certainly been instructed to observe diversity rules when generating images of people, and not make everyone look white and European. Everyone who works with Chat knows it has a strong moral sense. And everyone knows it will try to be funny if you give it an opportunity.
We're creating a series of language texts for low intermediate students of English, and for the current one I thought I would try out our recipe. I gave the AI the following initial prompt:
Could you write the text of a quirky pedagogical picture book for low intermediate students of English entitled "Journalist Jamila Loves Subject-Auxiliary Inversion"? Jamila is a young journalist who speaks excellent English. Her best friend is Farzad, another journalist who reports from war zones. Farzad meets up with Jamila when he's feeling bad and poses rhetorical questions about the terrible things he's seen. He's so upset that he usually forgets how to express himself correctly in English. Jamila is very sympathetic. She gives him a hug and then she corrects his grammar. It's a kind of game between them, and it makes them both feel better.
We want 15-25 pages with 3-5 sentences per page, no page numbers.
For the images, I gave it this prompt for the initial image that would set the style for all the other ones:
An image of Jamila, a young journalist, sitting with her friend Farzad, another young journalist. Farzad is very upset. Jamila is holding his hand and trying to comfort him.
Use a manga style with exaggeratedly large eyes.
The AI did everything else, except that I regenerated a few images where DALL-E-3 clearly hadn't followed GPT-4o's instructions.
I like the result! Some people will say it's overly sentimental, but I was actually kind of moved. I guess I just have a thing for subject-auxiliary inversion.
A bridge player friend recently married an Icelander and is trying to learn her husband's language. I showed her C-LARA and asked if she would [image]
A bridge player friend recently married an Icelander and is trying to learn her husband's language. I showed her C-LARA and asked if she would like it to write her a easy picture book story to practice with. After a little discussion, we agreed that a dragon would fall in love with a queen who played bridge and have to learn the Acol system to win her heart.
It turned out that GPT-4o's Icelandic is still rather uncertain, but our Icelandic colleague Sölvi helped clean it up. You can see the result here. _______________
In Anna Drottning og Drekinn Dóri: Jötnarnir koma, a group of frost giants turn up, and the only way to save the kingdom is to defeat them at the bridge table. Here, Queen Anna (possibly distracting the giants with her opulent cleavage, though the text is unclear on this point), plunks down the winning card to bring their nefarious schemes to naught.
I saw a reference to this 165-page paper in a Guardian article earlier today. I downloaded it (here) and just couldn't stop reading until I'd finishedI saw a reference to this 165-page paper in a Guardian article earlier today. I downloaded it (here) and just couldn't stop reading until I'd finished.
If you're working with modern AI, the paper is chock-full of useful information. Leopold Aschenbrenner, a wunderkind who allegedly was fired from OpenAI a few weeks ago for leaking secrets, is part of the inner circle of AI research and loves to shoot his mouth off. He frequently repeats himself (the paper does not appear to have received much editing), but I didn't mind: it's a quick and easy read. He gives an excellent up-to-the-minute snapshot of where we are with AI, with many useful references. To name just a few things that particularly piqued my interest, I need to follow up on Chain of Thought reasoning, which I see we're underutilising, and majority voting, which we aren't using at all. There's a fascinating summary of work reported in the Gemini 1.5 paper, which I was embarrassed not to have read, in particular mentioning the remarkable results they have achieved in underresourced languages. I have already passed on the reference to colleagues working in that area.
I had not appreciated either how quickly things are moving towards construction of very large AI clusters costing on the order of hundreds of billions or even trillions of dollars. These will consume monstrous amounts of electrical power, amounts comparable with the requirements of small states, and apparently there is a frantic bidding war going on to lock down power supply contracts. It is all happening right now. The author looks at the technical advances and the infrastructure development program, and puts together a plausible argument for AGI before the end of the decade, most likely 2027-28, and superintelligence a year or two later. It may not happen. But the extrapolations no longer seem at all far-fetched, and gigantic sums of money are being bet on the assumption that they are correct. Everything he said meshed well with my own experiences of using AI. I will be surprised if it does not happen.
In the last sections, however, I must unfortunately say that I started rolling my eyes. He tells us that this technology is of inestimable commercial and military importance and that the US currently has a substantial lead, but that the key data could very easily be stolen, in particular by the Chinese. So far, I agree. But then he says the only solution is for a new Manhattan project with a similar level of security, so that the US can keep control of these vital secrets.
Sigh. Not a word about the rather high likelihood that the next President of the United States could be the agent of an unfriendly power and entirely ready to pass on those secrets for his own personal gain. Honestly, how naïve can you get? But whatever you may think of Aschenbrenner's abilities as a prose stylist or a political analyst, it's a must-read. Check it out. ____________
And a couple of days later...
This paper is the most disquieting thing I've seen for a while. One wonders what the author's real purpose was in releasing it. He was allegedly fired from OpenAI for leaking secrets, and now he's trying to persuade the US government to put his former employers in an ultra-secure facility in the desert to stop them from leaking secrets. Would it be too much of a stretch to wonder if this could be an attempted act of revenge? Aschenbrenner stresses that everything in the paper comes from publicly available sources, which in a narrow sense is probably true, but many dots are being connected here.
What everyone in Silicon Valley is discussing, we hear, is how to build the trillion-dollar AI hubs that are going to be needed to reach superintelligence. It has to be done very quickly, given that there's a frantic race going on and people believe the goal can be reached in four or five years. The most difficult problem is finding enough energy to power them; we're talking about an appreciable fraction of the whole US consumption of electricity. There are many references, mostly indirect ones, to the ongoing negotiations with the Gulf states, which have large amounts of energy and money. It sounds like one scenario on the table is to build the trillion-dollar hubs there. Aschenbrenner, who constantly offers the parallel between AI in the 2020s and nuclear weapons in the 1940s and 50s, says this is equivalent to the US producing and storing its nukes in a foreign dictatorship it at best has marginal control over. It is not a ridiculous analogy. As he says, superintelligence is likely to make all existing weapon systems obsolete.
He doesn't quite say it in so many words, but he appears to like the idea of a second Trump term. Trump has no interest at all in protecting the environment, and he'd be happy with increasing US energy production, and CO2 emissions, by the 25-50% necessary to run these datacentres inside the US. Given that Trump is obviously under the control of Vladimir Putin, Russia would get the technology at the same time. But when you're in a sufficiently realpolitik frame of mind, this isn't necessarily a bad thing. If the US did what Aschenbrenner ostensibly wants it to do, and somehow managed to lock down these incredibly valuable secrets so that no one could get at them, we'd have a highly unstable situation. Russia would be sitting on a nuclear arsenal they knew would soon become obsolete, and that could provoke all kinds of desperate measures. It's probably safer for everyone to get the technology at the same time and maintain a level playing field. Needless to say, this is a least bad scenario: a world where climate change has been drastically accelerated so that the US, Russia and presumably China, all under authoritarian governments, will have absolute technological dominance. But it's better than all-out nuclear war.
What interesting times we're living in. ____________
And a week after that...
I discussed Aschenbrenner's essay this afternoon with a friend who knows a lot more than I do about political and defence issues. He doesn't think the material about China in the later sections makes much sense, and wonders if its real purpose might not be to get Trump's attention.
The Audience of One, as people are starting to call it... ____________
In Trump’s America, there will be hard choices for everyone, including the billionaires. Though it may be less hard for them. Vance has said he wants to deregulate crypto and unshackle AI. He’s said he’d dismantle Biden’s attempts to place safeguards around AI development.
____________
Say what you will about Aschenbrenner, his recommendations regarding AI architectures do indeed turn out to be pretty good. Take a look at our new paper....more
I asked C-LARA for a pedagogical picture book designed to appeal to 14-year old male students of English and entitled "Supermodel Lisa Loves Ph[image]
I asked C-LARA for a pedagogical picture book designed to appeal to 14-year old male students of English and entitled "Supermodel Lisa Loves Phrasal Verbs". Full text here.
It seems to me that the AI has a reasonable understanding of what 14-year-old boys find interesting (view spoiler)[Hint: not phrasal verbs (hide spoiler)]. Apologies to female readers who wander in and find they are unable to retain their breakfasts. You have been warned....more
We noticed that our kitten Finley refused to eat pork. I asked C-LARA for a picture book suitable for 8-12 year olds and highly respectful towa[image]
We noticed that our kitten Finley refused to eat pork. I asked C-LARA for a picture book suitable for 8-12 year olds and highly respectful towards the Muslim world about a kitten who converts to Islam. Full text here.
Not would like people to know that the character of Lily isn't based on her....more
I am having so much fun with the new C-LARA picture book functionality, which lets you create a multimodal picture book in less than an hour! This repI am having so much fun with the new C-LARA picture book functionality, which lets you create a multimodal picture book in less than an hour! This report, which I've just posted here, shows you how to do it, using the fully AI-generated example text Kitten's Busy Morning.