Capabilities that have crossed the line. Oldest flip first — a chronology of the frontier.
AlphaGo defeated Lee Sedol 4–1 in a five-game match in Seoul, March 2016. The line moved. --- AI has already demonstrated its ability to win a Go match agains...
Esteva et al. showed in Nature that a CNN could classify dermatology images at the level of board-certified dermatologists. --- Current AI systems can analyze...
Libratus crushed top professionals over 120,000 hands at Rivers Casino in January 2017. The first time a poker AI clearly surpassed humans at imperfect-informat...
AlphaZero learned chess from scratch in four hours and crushed Stockfish, the previous king of computer chess. The end of the human-vs-engine era. --- AI has...
DeepMind's AlphaStar reached Grandmaster level on the European ladder, beating professional players in long, real-time strategy games. --- AI systems have mad...
Mammography, lung CT, retinal scans. Specialty by specialty, narrow models keep clearing the human bar. --- Current research suggests that artificial intellig...
AlphaFold 2 solved a 50-year grand challenge in biology with near-experimental accuracy at CASP14. It now powers most structural biology pipelines. --- Curren...
Showing the work, not just the answer. By 2021 LLMs were doing this at near-perfect rates on standard datasets like GSM8K. --- AI can solve high-school math w...
As the global population grows, finding innovative ways to produce food in urban areas is crucial. AI can help optimize urban farming systems, but it requires c...
Predicting product success is a complex task that involves analyzing many factors, including social media trends and consumer behavior. AI can help with this ta...
LSAT logic games, GRE quantitative reasoning, similar formats — modern LLMs sit comfortably in the top decile. --- AI systems have demonstrated the ability to...
Developing a fair and unbiased algorithm for ranking job candidates is a challenging task. The algorithm must be able to evaluate candidates based on their qual...
Social movements can have a significant impact on society, and understanding what makes them successful is crucial. By analyzing the message and audience demogr...
Multiple-choice + free-response exams are firmly in LLM territory. Scoring 5s on AP exams is now a benchmark, not an achievement. --- Currently, AI systems ar...
Creating an effective learning plan requires understanding a student's strengths, weaknesses, and learning style. This task would test an AI's ability to make j...
A personalized mindfulness plan requires understanding the individual's mental health needs, goals, and preferences to create a tailored practice. This involves...
DeepMind's DeepNash defeated expert humans at Stratego — a game with imperfect information that had resisted prior approaches. --- Current AI systems have mad...
Travel planning can be a complex and time-consuming process, and AI can be used to create personalized travel itineraries that meet a person's specific needs an...
GPT-4 scored in the 90th percentile on the Uniform Bar Exam — a result that triggered a rethink in legal education and BigLaw hiring within months. --- Curren...
GPT-4 scored above passing on all three steps of the United States Medical Licensing Exam. Med-schools now teach 'how to use AI' as a clinical skill. --- AI s...
Verbal and quantitative both. The SAT has effectively been retired as an AI-progress benchmark — too easy. --- AI systems have demonstrated impressive capabil...
LeetCode hard, system-design walkthrough, the works. The traditional whiteboard interview is dead-or-dying because of this. --- AI systems have made significa...
Beyond undergraduate calculus into combinatorics, abstract algebra, real analysis. Not all of math, but a lot of it. --- AI systems have made significant prog...
Big-four firms quietly piloted GPT-4 against past CPA exams in 2023 with passing scores across all four sections. --- Currently, AI systems are not capable of...
Add a statement to the atlas. We review weekly.
Site
stuffaicantdo.com © 2026 · made in NL
Pick a side, give it a one-line title, optional explanation, and a category. We review submissions weekly.