Thanks for posting, I find these updates very interesting. How will you know when the Automated Coder milestone is reached? It seems like there weren’t be a point where it’s clear AI companies would prefer to lay off their workers rather than stop using AI. Is there also a chance that AI could 10x or 100x development speeds before it is possible for the software engineers to be laid off (if for example one of the things currently holding it back is unable to solved), would you consider that automated coding?
Afterimage
Karma: 76
I’m curious are people disagreeing with the points about centralisation > decentralisation, or the points about the usa being the best/better than china. Assuming that USA or china is the country to develop TAI, (free feel to correct that assumption if you disagree about the likelihood), as a non-american/non-chinese person I’m not happy with either option but still think the USA has the edge in creating a better post-TAI world than China would. I wouldn’t have thought this claim was so blantently false as to gather 23 disagree points.
As a data point after reading your post and the apple comic I still don’t understand how primer time travel works. If forced to write a fanfic I think I would butcher it too. I’m not sure if I should be holding an LLM to a higher or lower standard than myself.
The examples are clearly favoring haiku for those who want a summary.
-Chatgpt: 674 words to conclude there is no clean answer.
Haiiku − 274 words and gives an unambiguous answer.
I have read your link and I understand you seem to have much more knowledge about statistics than I am. Perhaps I’m making a simple mistake somewhere in my reasoning. My thoughts are:
50 success horizon went from 1 to 4 hours. I interpreted your comment and post to posit that the increase is to some large degree due to training on cybersecurity to increase preformance in the 2-4hr range. However, looking at the charts it seems clear that the increase is not dominated by increased preformance in the 2-4hr range.
My point isn’t that there’s enough samples or that it isn’t possible to game the METR test, it’s that the measured improvement for Claude Opus 4.5 doesn’t seem to be primarly in the 2-4 hr bracket, if anything it’s dominated by 4-16hr.
If I’m intrepreting these charts correctly, there is a decent amount of progress in 15m-1hr bracket, a small amount of progress in the 1hr bracket, a decent amount of progress in the 2-4 hr bracket, and a large amount of progress in the 4-16hr bracket. It doesn’t look like progress was dominated by the 1-4hr range.
Thought-provoking post, especially the section about the meaning of traditional food. I’m curious about the relationship between insulin resistance and obesity. If you can keep skinny can you avoid the metabolism FUBAR you’re talking about even if your diet is still westernised?
So, I have to admit I’m still confused. Is the icecream example fairly unrelated to the introduction and first chapter? They seem to be talking mostly about pure qualia, while the ice-cream example is talking about actions.
I agree qualia is entirely disconnected from rationality, but I think anything beyond qualia such as actions or intending to take actions is rationality fair game so to speak. I don’t see an issue in Bryce assessing the rationality of Ash stopping for ice-cream, it was his communication/social skills that were lacking.
After reading your post very carefully, I think you are agreeing with the above, I just had the opposite impression upon my first and second reading. Apologies if i’m still misunderstandin, either way I find this topic very interesting so thanks for writing this post.
I also find the ice-cream example confusing because it’s your main example but it doesn’t seem like it supports your main points. For example replace Ash with a drug addict and ice cream with meth and Bryce suddenly looks like a hero trying to help his friend from making a big mistake.
I think the below example still makes Bryce look like how it seems you intended him to look even if it was talking about meth.
Ash: “Ooh, I want ice cream.”
Bryce: “Seriously?”
Ash: “Well, I’m not actually going to buy some, I just want some”
Bryce: “That’s the problem. You know it’s not good for you. You should want to be eating healthy food”
Ash: “I mean, yeah, but right now I just really feel like ice cream.”
Bryce: “That’s stupid”
For anyone under 30, 10 years of polical lying, or 20 years for journalists “lying” is a long time and could be seen as a new reality, especially seeing as it seems to be working out quite well for the liars so it’s unlikely to change anytime soon.
Can you explain more, could a lower income worker without family nearby afford child care and full time house help?
Personally the idea of no free will doesn’t negatively impact my mental state, but I can imagine it would for others, so I’m not going to argue that point. You should perhaps consider the positive impacts of the no-free will argument, I think it could lead to alot more understanding and empathy in the world. It’s easy for most to see someone making mistakes such as crime, obesity, or just being extremely unpleasant and blame/hate them for “choosing” to be that way. If you believe everything is determined, I find it’s pretty easy to re-frame it into someone who was just unlucky enough to be born into the specific situation that led them to this state. If you are yourself successful, instead of being prideful of your superior will/ soul, you can be humble and grateful for all the people and circumstances that allowed you to reach your position/mental state.
Thanks, that sated my curiosity nicely. Just so you know I’m not trying to pretend I’ve optimised my child’s upbringing, just doing the best I can like most parents I know. I reckon your kids are lucky to have you.
I’ve got a few questions, mostly curiosity and not trying to be critical at all.
Regarding the purpose of this project, would you say the main motivation is academic or just something you’re doing for fun. For example would you still have done it if you had read an extremely trustworthy study saying that starting toddlers reading at an early age had no impact on future academic success.
I’ve done no research on the effect of bribes and child-rearing, is that something you researched before designing the token system?
A bit of broader somewhat personal question—You’ve put a lot of effort into teaching reading to your toddler’s, do you also put a similar amount of into other similar abilities. What other skills besides reading have you been focusing on. Off the top of my head, thinking logically, basics of nutrition, identifying emotions, interpersonel skills, basic maths, calibrating risk. Again, not any kind of attack on you, just thinking about these things myself as I have a young kid. Obviously both parents and toddlers have limited time and energy can only go anki-droid level deep on maybe one topic/subject.
I’m curious, if you imagine someone who is more conscientious and making better life descisions than you, if they were to look upon you, do you expect them to see you as some kind of cat as well? Similarily, if you were to imagine a less conscientious version of yourself? If you can find empathy here, maybe just extend along these lines to cover more people.
Also, having a deterministic view of the universe makes it easy for me to find empathy. I just assume that if i was born with their genetics and their experiences I would be making the exact same descisions that they are now. I use that as a connection between myself and them and through that connection I can be kinder to them as I would hope someone would be kinder to me in that situation. If you have sympthy for people born into poverty, it’s the same concept.
This is important context not only for evaluating Greg Burnham’s accuracy but also for the Gold Medal headline. If this difficulty chart is accurate (still no idea on the maths), getting 5⁄6 is not much of a surprise. Even question 2 and 5 seem abnormally easy relative to previous years.
I have no idea of the maths, but reading through the epoch article it seems to me that this result is entirely unexpected.
“but this year I’d only give a 5% chance to either a qualitatively creative solution or a solution to P3 or P6 from an LLM.”
Sure it’s unreleased LLM but it still seems to be an LLM.
I bite the bullet. I do think it’s fine and actively good to have 7-year-olds and 17-year-olds in the same math classroom. Of course, if you think that learning is bad, you won’t like this plan to have kids learn
I’d be keen to hear an explanation of this bullet biting. My instincts tell me it’s a very bad idea and I imagine most people would agree but I’m interested in more details.
I find this graph useful. I think you can agree at some point that AI will be more intelligent than humans, even if AI intelligence is quite different and lacking in a few (fewer every year) areas. If this is the case then this graph is quite effective at conveying that this point may be happening soon.
Wouldn’t that imply that Opus 4.6 should have been below trend since it wasn’t a step change in model size or pretraining (as far as I know).