No. 1 Superforecaster & AI 2027 Author Eli Lifland - On Our Differing Timelines to Superintelligence (New Podcast Series Potentially!)
Added 2025-06-19 10:07:39 +0000 UTCA debate on whether superintelligence will arise by 2027, with No. 1-ranked RAND Superforecaster & AI 2027 co-author Eli Lifland. Set to be a new series, where we discuss on the podcast how progress is going with respective to our divergent timelines.
AI 2027: https://ai-2027.com/
Key Timeline and Eli Update: https://ai-2027.com/research/timelines-forecast
Eli Profile: https://www.elilifland.com/
Key METR Analysis: https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/
Eli Lifland Retweet: https://x.com/dwarkesh_sp/status/1926364571104653555
FrontierMath: https://epoch.ai/frontiermath
Epoch on SWE-Bench Verified: https://epoch.ai/blog/what-skills-does-swe-bench-verified-evaluate
RE-bench: https://arxiv.org/pdf/2411.15114
Terence Tao Fridman: https://www.youtube.com/watch?v=HUkBz-cdB-k
GSM8K-Platinum: https://gradientscience.org/gsm8k-platinum/
AI Explained Video on AI 2027: https://www.youtube.com/watch?v=wOBqh9JqCDY
AI 2027 Fundraiser: https://ai-2027.com/about
Comments
I’m 3:30 in, with that said: this feels like interviewing a doomsday prophet preemptively backpedaling on their wild clickbait-grabbing prediction. There’s a quasi-religious fervor around AI. Some hope we’re creating omnipotent god-like entity that will solve humanity’s problems and deliver us to paradise. Doomers loose sleep over impending destruction of humanity at the hands of the wrong deamon summoned by the first group. And then there’s there a cottage industry of prophecy high on their 5 minutes of fame. With 25+ years of SW engineering experience I learned this lesson the hardest: delivery date estimation is a crystal ball exercise to appease management so that sales can keep on twisting the clients to pay our bills. Nobody knows what they’re doing. So, I am calling BS. These guys’ personal excuse might be their young age and short experience, but as SW industry we should get real. Let CEOs pull wool over investors’ eyes, they will gamble at the stock market casino no matter what we do. If you’ve shipped any real software, you’ll known the 80% effort gap between a demo prototype and shippable product. If you haven’t, just think of fully self-driving cars. Tesla’s been “selling” these since 2017. That’s 8 f#%king years and counting. So in 2-5 years, there will be superhuman AI software engineers. Based on what? Current tech, because we understand it thoroughly and know which kinks to iron, right? Oh, you meant the unforeseen breakthrough that will finally unlock everything? Because you’ve seen the signs of miracles occurring more frequently lately?
Pavol Vaskovic
2025-07-16 06:03:11 +0000 UTCI am of the mind that several of the problems we face -- hallucinations, blackmailing, and probably some others as well -- can be resolved or at least greatly reduced in a rather straightforward way at the cost of additional compute at runtime. That is, more time to think.
Jason Dowd
2025-06-30 06:42:48 +0000 UTCI think real-life common sense and usefulness make normal people very sceptical. These models may pass some tests, but when you ask a simple question like "What time will a plane that just took off from city X land in city Y?", they start producing nonsense—even if they find the webpage with the answer. It's like those super-intelligent kids who pass every exam at school but can't figure out how to use a bus.
Arek Stryjski
2025-06-19 12:55:44 +0000 UTCIt’s a joy not only to listen but also to explore the provided links.
SteveHaupt
2025-06-19 12:25:14 +0000 UTC
