# Bloomer > Bloomer (bloomer.news) publishes positive AI news: real breakthroughs, helpful AI tools and milestones from AI history. Every post is an original short summary that links to the reputable source it is based on. Bloomer does not invent facts, quotes or numbers. Published by Telep IO LLC. ## Sections - [Positive AI news (home)](https://bloomer.news/): the latest stories, updated daily - [Positive AI on X](https://bloomer.news/launches/): AI launches and demos posted by the companies and labs behind them, attributed as their own claims - [This day in AI history](https://bloomer.news/history/): AI milestones on the day they happened - [Tech Twitter](https://bloomer.news/tech-twitter/): notable positive tech news from X - [Answers](https://bloomer.news/answers/): sourced answers to common questions - [Newsletter](https://bloomer.news/newsletter/): a free weekly email - [About](https://bloomer.news/about/): editorial approach - [RSS feed](https://bloomer.news/rss.xml) ## Answers - [What are the biggest positive AI breakthroughs of 2026?](https://bloomer.news/answers/positive-ai-breakthroughs-2026/) - [How is AI helping doctors and patients today?](https://bloomer.news/answers/how-ai-helps-doctors-and-patients/) - [What is a bloomer? AI optimist vs doomer](https://bloomer.news/answers/what-is-a-bloomer/) ## Latest stories - [AI spots a treatable weakness in ovarian cancer from ordinary tissue images](https://bloomer.news/stories/kcl-hrdpath-ovarian-cancer/): King's College London's HRDPath model reads microscope images of tumour tissue to find faulty DNA repair, a flaw some drugs can exploit, without a genomic test. Source: King's College London. - [Claude designs protein binders against scorpion toxins, a first step toward new antivenoms](https://bloomer.news/stories/claude-scorpion-toxin-binders/): Working mostly on its own for 64 hours, an internal Claude research model designed protein binders against four dangerous scorpion neurotoxins. Lab tests confirmed 17 of 211 designs. Source: Anthropic. - [A startup's AI model learns to read the electrical signals of plants](https://bloomer.news/stories/electric-plant-bombadil-plant-signals/): Electric Plant built Bombadil, an AI model that interprets tiny electrical signals inside about 125 plant species to tell growers what their crops need. Source: TechCrunch. - [AI agents help physicists find new exact solutions for fusion magnetic fields in hours](https://bloomer.news/stories/gemini-agents-extend-fusion-solutions/): Researchers used Gemini 4 Argon agents to extend exact solutions for the magnetic fields that hold plasma in fusion devices, giving software developers new reference cases. Source: TechRadar. - [KAIST AI restores satellite views hidden by clouds](https://bloomer.news/stories/kaist-geocore-cloud-free-satellite-views/): GeoCore-9B rebuilds cloud-covered satellite images and turns radar scans into photo-like views, which could help disaster response and environmental monitoring. Source: KAIST News Center. - [Left alone on a virtual island, Claude agents started to play and learned from it](https://bloomer.news/stories/claude-agents-play-virtual-island/): Researchers left Claude agents on a simulated island for 30 hours with no task. They invented games and goals, and the playtime made them better at later challenges. Source: MIT Technology Review. - [MIT AI keeps fetal brain MRI scans sharp even when the baby moves](https://bloomer.news/stories/mit-e3-pose-fetal-mri/): E(3)-Pose tracks a moving fetus's head during MRI so scans stay complete, which could mean earlier diagnoses and fewer repeat hospital visits. Source: MIT News. - [Seoul National University robot AI pictures the result before it moves](https://bloomer.news/stories/snu-dynin-robotics-predicts-scenes/): Dynin-Robotics reads instructions, imagines the scene after an action and uses that picture to choose its moves, topping the models it was compared with. Source: Seoul National University College of Engineering. - [MIT model helps AI understand crystals in 3D to find better materials](https://bloomer.news/stories/mit-ai-crystal-materials/): A new multimodal AI from MIT pairs atomic structures with written descriptions, helping it suggest promising materials such as battery electrolytes. Source: MIT News. - [AI summaries lighten the load for cancer clinicians before first visits](https://bloomer.news/stories/ai-oncology-consult-prep/): At the Netherlands Cancer Institute, an AI tool that turns messy records into structured summaries cut clinicians' perceived workload in a real world study. Source: npj Digital Medicine. - [Stanford AI helps clinicians follow ICU patients' activity across a whole hospital stay](https://bloomer.news/stories/stanford-tracer-icu-patient-activity/): TRACER combines bedside video with health records so doctors and nurses can see how a patient's movement and behavior change over an ICU stay. Source: Stanford AI Lab Blog. - [AI agents help Penn State scientists find a new pattern in food physics](https://bloomer.news/stories/penn-state-ai-agents-food-physics/): Guided by human experts, AI agents gathered more than 4,000 experimental data records from food science papers in 30 minutes and revealed a previously unreported relationship. Source: Penn State. - [AI helps new radiation oncologists outline prostates more accurately and faster](https://bloomer.news/stories/ai-contouring-brachytherapy-trial/): In a randomized trial presented at ASTRO 2026, trainees using AI generated contours matched expert outlines more closely and finished about 24 percent faster. Source: ASCO AI in Oncology. - [Caltech AI learns to spot sudden shifts ahead of time in simulated climate and airflow](https://bloomer.news/stories/caltech-ai-forecasts-tipping-points/): A recurrent neural operator predicted the timing of abrupt transitions in simplified models, in one case looking more than two years ahead. Source: Caltech. - [AI design tools create enzymes that cut disease proteins on command](https://bloomer.news/stories/ai-designed-protease-scissors/): Researchers in David Baker's lab used AI protein design software to build new proteases that slice chosen targets, including proteins tied to dementia. Source: Ars Technica. ## AI history - [Neural Turing Machines give neural networks a working memory](https://bloomer.news/history/2014-10-20-neural-turing-machines/): On October 20, 2014, Alex Graves, Greg Wayne and Ivo Danihelka of Google DeepMind submitted Neural Turing Machines, a neural network that learns to use an external memory. - [UniRep shows deep learning can read the language of proteins](https://bloomer.news/history/2019-10-21-unirep-protein-language/): On October 21, 2019, a Harvard team including George Church published UniRep, a neural network that learned the fundamentals of proteins from raw amino acid sequences. - [AlphaGo Zero teaches itself Go from scratch](https://bloomer.news/history/2017-10-18-alphago-zero/): On October 18, 2017, DeepMind unveiled AlphaGo Zero, a Go program that learned purely by playing itself and beat its champion predecessor 100 games to 0. - [AI learns to help Wikipedia find better citations](https://bloomer.news/history/2023-10-19-ai-wikipedia-citations/): On October 19, 2023, researchers mostly from FAIR at Meta published SIDE, an AI system that flags weak Wikipedia citations and recommends better ones. ## Optional - [Sitemap](https://bloomer.news/sitemap-index.xml)