@colin_fraser Also, your patience in this thread is to be commended.
English
Simon Rogers
4.2K posts
Simon Rogers
@sdrogers
AI data scientist at NHS National Services Scotland. Hon Lecturer in Comp Sci, University of Glasgow. Author of https://t.co/gUVMPYIpCj Views my own, etc.
iPhone: 57.586338,-4.435981 Katılım Ocak 2009
946 Takip Edilen601 Takipçiler
If the missing piece is that you make occasional arithmetic errors in the small multiplication or addition steps then a calculator solves this. But if the missing piece is that you lose track of long-running reasoning processes then you have bigger problems.
English
“What’s the point of this? Can’t you just give it a calculator?”
The point is that if you have the small times tables memorized, the ability to reliably add, and you can follow a sequence of steps, then you can do this with 100% accuracy. If you can’t, then what’s missing?
Yuntian Deng@yuntiandeng
For those curious about how o3-mini performs on multi-digit multiplication, here's the result. It does much better than o1 but still struggles past 13×13. (Same evaluation setup as before, but with 40 test examples per cell.)
English
@colin_fraser Each step is to some degree stochastic though, right? So, even if the correct token at some step is really really likely, then as you increase the number of steps, the probability of not picking the correct one becomes very very high.
English
@GaryMarcus @HZoete I guess they'll also have a universally agreed definition of consciousness to determine it's been reached too...
English
@HZoete That’s a silly slide. No science behind those made up numbers.
English
As I head home from five days in Paris at the AI Action Summit, some thoughts on what we learnt about the international picture.
* What worked
* What didn’t
* What needs to change.
A 🧵
English
@colin_fraser Absolutely. And pretty easy, once that tool exists to train a model where beating that test is part of the objective.
English
I maintain my position that this is basically impossible in general
nature@Nature
Scientists are closing in on a tool that can reliably identify AI-generated text without affecting the user’s experience go.nature.com/48iKc38
English
@thefulltoss @asianick85 It was doing a lot today. I don't think any team would have survived / won. So, would agree that toss important. But if they'd blocked they would have likely lost by more.
English
Lose the toss, lose the match. Or play crap, lose the match?
English
Enjoying this podcast series featuring @MelMitchell1 on intelligence -- highly recommended
santafe.edu/culture/podcas…
English
@robinsall I'm not sure "all", but plenty. Especially when it's really hot.
English
@hganjoo_153 I guess if DLS wasn't good at predicting outcome, it also wouldn't be good at defining targets.
English
The remarkable simplicity of cricket -- a simple Duckworth-Lewis based prediction model works almost as well as Cricinfo's Forecaster on average. The game state (overs and wickets left) matters the most, and dwarfs all other minutiae.
English
@jessRmorley @BBCr4today @aleksk @Kevin_Fong And the data quality. Lots is sitting in random xlsx files...
English
@jessRmorley @BBCr4today @aleksk @Kevin_Fong Enjoyed your input there, thanks! 100% agree on the less trendy backroom tasks that could make a big difference.
English
@GeoffreyPetty @LRB That didn't scan. Basically, it's goodness seemed to be that it could produce bigger lists of targets. And humans could feel like they weren't responsible for the choice.
English
@GeoffreyPetty @LRB That, yes. But more the fact that there's no way of evaluating its accuracy, the justification that it could identify more targets with humans able to delegate any responsibility.
English
Three articles in the @LRB this month that taken together are quite chilling: the wholesale swallowing of the religion of AI (article on Blair) + the use of AI in generating kill lists + the hollowing of the regulatory landscape (in context of grenfell)...
English
@LRB Mind you, could you make the personal ads more entertaining again please?
English
We are a good newspaper, not a good news newspaper.
Simon Rogers@sdrogers
Three articles in the @LRB this month that taken together are quite chilling: the wholesale swallowing of the religion of AI (article on Blair) + the use of AI in generating kill lists + the hollowing of the regulatory landscape (in context of grenfell)...
English
Uh... what?
The Nobel Prize@NobelPrize
BREAKING NEWS The Royal Swedish Academy of Sciences has decided to award the 2024 #NobelPrize in Physics to John J. Hopfield and Geoffrey E. Hinton “for foundational discoveries and inventions that enable machine learning with artificial neural networks.”
English
@rcolvile Can live very comfortably on that, and I'm confident that enough very skilled applicants would see it that way.
English
