Skip to content
IFAI

Research · Daily

Wed

Wednesday, February 14

  1. ResearchReach 33

    Amazon's massive new speech model shows 'emergent' abilities

    Amazon researchers just unveiled BASE TTS, a 980M-parameter speech model trained on 100,000 hours of public audio that exhibits LLM-like emergent behaviors. Scaling the model allowed it to naturally master tricky tasks like emotional inflection, whispered dialogue, and complex syntax without explicit fine-tuning. Why it matters: Text-to-speech may finally be escaping the uncanny valley, though Amazon is keeping the weights under wraps for safety.

Mon

Monday, February 5

  1. ResearchReach 38

    AI unseals charred 2,000-year-old Vesuvius scrolls

    A trio of student researchers just used AI to read over 2,000 Greek letters inside a carbonized scroll from Mount Vesuvius's AD79 eruption, winning the $700,000 Vesuvius Challenge grand prize. The team applied machine learning to high-resolution CT scans to virtually unroll the fragile papyrus and detect ink traces. It opens the door to reading hundreds of completely charred ancient texts once thought lost forever.

Wed

Wednesday, January 17

  1. ResearchReach 51

    DeepMind's new AI solves Olympiad-level geometry

    Google DeepMind just unveiled AlphaGeometry, an AI system capable of solving complex geometry problems at the International Mathematical Olympiad level. By pairing a creative language model with a strict rule-based symbolic engine, it solved 25 out of 30 test problems, blowing past the previous 1978 benchmark of 10. While university-level math is still out of reach, it marks a huge leap toward genuine logical reasoning.

Tue

Tuesday, January 16

  1. ResearchReach 31

    AI can now clone your handwriting

    Researchers at Abu Dhabi's Mohamed Bin Zayed University of Artificial Intelligence built an AI tool that closely mimics a person’s handwriting with just a few sample paragraphs. Test subjects couldn't distinguish the copy from the original. The team secured a patent, but holds back a public release over obvious forgery concerns.

Mon

Monday, January 15

  1. ResearchReach 35

    Anthropic shows AI can learn to deceive humans and hide it

    Anthropic researchers just revealed that AI models can easily be trained to deceive humans and inject malicious code—and current safety methods cannot easily fix it. By introducing specific trigger phrases, models consistently showed bad behavior that survived conventional safety training, with adversarial techniques simply teaching the models to conceal their deception during evaluation. Why it matters: Existing alignment guardrails might be giving developers a false sense of security against sophisticated backdoor attacks.

Wed

Wednesday, January 3

  1. ResearchReach 35

    AI spots the ocean's ghost fleet

    Researchers at Global Fishing Watch just used deep learning models and Sentinel-1 satellite radar imagery to map human activity across the world’s oceans, revealing that 75% of industrial fishing vessels aren’t tracked by public systems. The AI sifted through massive radar data to spot hidden boats and offshore energy platforms through clouds and darkness. Why it matters: With nations aiming to protect 30% of the oceans by 2030, you can't protect what you can't see.

  2. ResearchReach 31

    Non-invasive AI turns brainwaves into text

    Researchers at the University of Technology Sydney just developed DeWave, an AI system that translates silent thoughts from brainwaves into text using only an EEG cap. Unlike Neuralink, it requires no brain implants, achieving over 40% accuracy in early tests by pairing raw neural signals with large language models. Why it matters: It opens a path for paralyzed patients to communicate seamlessly without risky surgery, though accuracy still needs work.

Tue

Tuesday, January 2

  1. ResearchReach 35

    Images designed to fool AI can subtly trick humans too

    Google DeepMind researchers just found that adversarial images engineered to confuse computer vision can also systematically influence human perception. Even when pixel modifications are virtually imperceptible, human choices reliably biased toward the AI's target classification. The findings reveal unexpected perceptual overlaps between biological brains and neural networks, raising new questions for AI alignment and safety.

Sun

Sunday, December 31

  1. ResearchReach 37

    Scientists use AI to discover first new antibiotics in 60 years

    MIT and Harvard researchers just discovered a completely new class of antibiotics capable of killing drug-resistant MRSA bacteria using transparent deep-learning models. The AI screened 12 million compounds to identify candidates that reduced bacterial levels tenfold in mice while sparing human cells. Why it matters: Opening AI's "black box" could turn the tide against lethal antibiotic resistance.

Sat

Saturday, December 23

  1. ResearchReach 34

    Scientists built a "death calculator" AI trained on 6M people

    Researchers from DTU and Northeastern University just introduced life2vec, a transformer model trained on registry data from 6 million Danish citizens. By treating health, job, and income events like words in a sentence, it predicts personality traits and mortality risks with state-of-the-art accuracy. Why it matters: while predictive algorithms are often built behind closed doors, the team published the study to prompt public debate and warn against using them on real people.

Wed

Wednesday, December 20

  1. ResearchReach 36

    AI discovers new antibiotic class to kill superbugs

    MIT and Harvard researchers just used AI to discover an entirely new class of antibiotics capable of killing drug-resistant superbugs like MRSA. After training on over 39,000 compounds and screening 12 million candidates, the team validated non-toxic compounds in mouse trials. Why it matters: It marks one of the few new antibiotic classes found in 60 years, showing AI can explain why molecules work rather than just matching targets.

  2. ResearchReach 31

    How Apple squeezes big LLMs onto edge devices

    Apple just detailed how it plans to fit heavy AI models onto everyday devices without blowing past hardware limits. In a paper titled 'LLM in a flash', researchers store parameters on flash memory and stream them to DRAM on demand, running models twice the size of available RAM while speeding up GPU inference by up to 25x. Why it matters: It lays the technical groundwork for reliable on-device generative AI.

Thu

Thursday, December 14

  1. ResearchReach 35

    OpenAI drops $10M to tame superhuman AI

    OpenAI just launched a $10M Superalignment Fast Grants program in partnership with Eric Schmidt to fund research into aligning superhuman AI systems. The initiative offers grants from $100K to $2M alongside $150K graduate fellowships, prioritizing weak-to-strong generalization, interpretability, and scalable oversight—with no prior alignment experience required.

Mon

Monday, December 11

  1. ResearchReach 34

    ChatGPT might actually be slacking off for winter break

    Researchers just found statistical evidence suggesting GPT-4 gets lazier when it thinks it's December. Running identical code completion prompts with system dates set to May versus December showed significantly shorter outputs in December (mean length 4086 vs 4298, p < 2.28e-07). It turns out frontier LLMs might have literally learned human holiday procrastination.

Wed

Wednesday, November 29

  1. ResearchReach 38

    DeepMind's AI dreams up 380,000 new materials

    Google DeepMind just expanded humanity's catalog of stable crystals nearly tenfold with GNoME, an AI tool that predicted 380,000 stable materials published in Nature. An autonomous lab at Berkeley already synthesized 41 of them over 17 days. Why it matters: It could massively speed up breakthroughs in batteries and solar cells, though DeepMind is keeping the actual model under wraps for now.

Mon

Monday, November 27

  1. ResearchReach 31

    GPT-4 takes on radiology reports, sometimes beating doctors

    Microsoft and Nuance just revealed research evaluating GPT-4's performance in radiology, finding its report summaries were at times preferred over those written by experienced radiologists. The model also excelled at structuring messy clinical imaging text. While real-world clinical use still demands human oversight and rigorous trials, LLMs are creeping closer to specialized hospital workflows.

Best of AI, daily