🤖 GPT-5.4 just pulled off something AI critics said wouldn’t happen anytime...

🤖 GPT-5.4 just pulled off something AI critics said wouldn’t happen anytime soon: it helped make a real scientific discovery that worked in the lab. Most AI systems stop at summarizing papers or suggesting ideas. GPT-5.4 went much further. It helped researchers identify a chemistry bottleneck, propose a surprising fix, design experiments, analyze results, and figure out what to test next. It’s winning idea? Add a small amount of TEMPO, a common chemical additive, to reactions involving primary sulfonamides, compounds that chemists have struggled to work with for years because they often produce disappointing yields. Then came the real test. Researchers ran 10,080 automated experiments. The results were hard to ignore: • 88% of boronic acids showed improved yields • 83% of sulfonamides performed better • Average yields jumped from 16.6% to 25.2% • High-yield reactions (>30%) more than doubled, rising from 15.6% to 37.5% The team then scaled the experiments up in a traditional lab. The improvement still held. • 11 out of 14 reactions got better • 8 delivered more than double the original yield For years, AI has been a research assistant. This is one of the clearest signs that it’s starting to become a research collaborator. @aipost 🏴