
AI Policy | Social Contract Enjoyer Biggest charva at @GoogleDeepMind. Washingtonian (the one in Sunderland) 🏴
Merrie England
Joined June 2019
- Tweets540
- Following1.6K
- Followers562
- Likes3K
Pinned Tweet
Watching the eclipse through a translucent Wetherspoons sign.
This company truly is the best of Britain.
Hurling my phone under a lorry on the south circular road as Substack suggests an article to me called Longtermism for Animals.
48
Nick Swanson retweeted
thank you to @KanishkaNarayan for joining the London launch of @ukdynamism, to the team who made it an incredible night, and to the believers and builders that joined us
this land has a long history. with the work of many people in this room and way beyond it, it will have a long future
Last night, a room of believers came together to launch the @ukdynamism fund. Thank you all for coming.
My totem is just about to start counting… 🧨
Rare spotting of the Original Washington on the timeline
Whenever I see this on the TL I’m reminded how great a language English is. Look at this syntax.
“In lands to him unknown” is iambic, as are the first lines. And Latin puts adjectives after the noun — terra incognita — which is why “lands… unknown” feels so formal.
136
Got Argon to build me a little interactive front end for my task management and it is delightful.
2
76
Nick Swanson retweeted
Today we have 3 new DeepMind Institute essays:
How can we control misbehaviour in agent swarms?
How should we orchestrate complex networks of AIs and people?
The case for making AGI’s benefits equitably distributed.
Get them here -> bit.ly/deepmind-institute
Nick Swanson retweeted
"stochastic parrot" was a mimetically-fit cognitive virus that spread from 2021-2025; it temporarily blinded many gifted people to the nature of AI progress, burning up crucial years in which they could have helped think through the response to the situation.
Nick Swanson retweeted
My journey to develop AGI spans 25 yrs, including 10+ yrs thinking about technical & societal perspectives at Google DeepMind.
AGI is on the horizon - we need deeper understanding of its implications. To help, we've created the DeepMind Institute.
Prediction market idea: When will the word "shoggoth" first be spoken in either the House of Commons or House of Lords?
I reckon we're on course for it to happen this session.
1
10
389
Nick Swanson retweeted
King Midas is often used as an AI parable (because it is a very clear story which demonstrates the alignment problem) but you rarely hear the version where in the end, Dionysus had the King wash in the Pactolus river to remove the curse, whose sands, now filled with gold (according to legend) helped build the wealth of the Lydian empire. There is probably some mythological wisdom in this part of the story too.
All chat app versions of AI feel at times like you're talking to Indecisive Dave, because they eagerly jump to give you a full and comprehensive answer to often very uncertain queries or early sketches of plans. When you give them a tiny bit more context they end up idiotically and confidently acting as if they were recommending the exact opposite thing. It's striking how often conversations feel like this.
There are some UX improvements (multiple choice Q's, occasionally offering to interview me, suggested next steps) which improve this, but there is a whole class of interaction (basically everything where there are multiple possible interpretations based on unspecified information) where initial back and forth on the details should be the norm.
There is plenty of research on this as you'd expect (research.google/blog/learnin…) but it is striking how good LLMs are at searching the web and synthesising info, compared to how well they search the actual source of the context and the entire query - me, the user.
New post! Alignment doesn't just relate to a model's properties, but also the systems and institutions under which it operates. As we enter an increasingly multi-agent world, better harnesses, protocols, and mechanism design will be critical for safety. 🐜 blog.cosmos-institute.org/p/…
I genuinely love the fact that AI overviews will answer any questions at face value. I think it's lovely.
1
2
675
There is an interesting point to be made about models and systemic risk in sectors like financial services...
...but this is just the equivalent of saying "FYI, we plan to build really bad systems with no redundancy in them - cheers seeya in a bit. 😘"
theguardian.com/business/202…
316






















