Generating *code* using a statistical prior about which token sequences are more likely is a fundamentally different task than generating *programs* by searching directly in the space of program dynamics
@fchollet
-
ARC Prize 2025: Different Tasks for Public and Private Leaderboards
By
–
Unlike prior years, the 2025 ARC Prize competition uses a different set of tasks for the public and private leaderboards on Kaggle (which is standard for most ML competitions). So you should expect some reshuffling of the leaderboard at the end of competition.
-
Criticism of delegating thinking to AI text generators
By
–
Can't even do your own 2-min research or your own thinking, have to delegate it to a text generator.
-
Objective reality’s decline threatens collective decision-making
By
–
Our grasp on objective reality is currently losing ground on seemingly all fronts, not just image posts on social media. The problem is that when objective reality is gone, we will be fundamentally unable to make collective decisions about what kind of world we want to live in.
-
People’s inability to distinguish AI images from real photographs
By
–
I am baffled by how a large majority of people appear to be unable to tell the difference between AI generated imagery and photographs, even when the picture in question has a background full of garbled text signage. RIP objective reality
-
Compositionality Over End-to-End Learning: The Future of AI Models
By
–
Much of the field obsesses over end-to-end learning. But strong generalization requires compositionality: building modular, reusable abstractions, and reassembling them on the fly when faced with novelty. The models of the future won't be just pipes, they will be Lego castles.
-
LLMs Enable New Spammer Tactics with Synonym Swaps
By
–
Spammers are now using LLMs to run multiple rounds of quasi-synonym swaps in order to avoid getting caught by spam filters due to repetitions. This is a new form of English — spammer-speak, where every sentence is nearly single-use
-
Benchmarks and Competitions Drive Scientific Progress
By
–
Benchmarks and competitions are an effective way to drive scientific progress.
-
ML Engineer Recruitment for European Defense Drone Technology
By
–
If you're a machine learning engineer passionate about European defense, and if you believe (like I do) that mass-produced semi-autonomous drones are the future of warfare, send your resume to talent@harmattan.ai. Harmattan is a very unique startup that is completely vertically
-
Humanity Will Continuously Increase Compute Usage Year After Year
By
–
You can safely bet that humanity will keep use more and more compute year after year. It's impossible to have a glut of compute, because any additional compute available can always be used to improve performance on the tasks you're already doing.