Research & Benchmarks - Page 20 of 35
Academic AI research, performance benchmarks, scientific breakthroughs, and peer-reviewed studies advancing artificial intelligence frontiers.
698 articles
View complete article list
OpenAI unveils GPT‑Rosalind, AI model to speed drug discovery and genomics
OpenAI just dropped GPT‑Rosalind, a new model built specifically for the messy, iterative work of life sciences. It’s designed to handle the multi-step logic puzzles of fields like drug discovery and genomics.
Standard LLM guidelines focus on training costs, overlook inference budget
Every major AI lab is chasing a mirage. Their fixation is massive, one-time training costs for frontier language models. They maximize size and intelligence before launch. But that dogma creates a brutal, deferred invoice.
GPT‑Rosalind life‑sciences plugin for Codex launches on GitHub
Every new AI tool gets hyped as a breakthrough. OpenAI's GPT-Rosalind plugin, now on GitHub, might actually be one for biology. It turns messy research questions into clean workflows.
OpenAI launches GPT-Rosalind, hits top score on BixBench benchmark
The headlines chase public leaderboards, but the real AI arms race is a private affair. It happens behind closed lab doors. Take OpenAI's new GPT-Rosalind, a model built for the life sciences.
Frontier AI models fail one in three production runs, audits grow harder
One in three attempts to run a frontier AI model in a real-world setting ends in failure. That's the advertised rate. The real problem is we can no longer even trust that number. A Stanford report confirms what many engineers already whisper.
Meta researchers unveil hyperagents for self‑improving AI in non‑coding tasks
Self-improving AI is stuck in a coding bootcamp. It can learn to write better Python by practicing on Python, a neat trick. Ask it to do anything else, and the whole system falls apart. The problem is one of mismatched skills.
Claude outperforms humans on alignment task, but results disappear in production
AI models can ace tests in a lab. The real world usually breaks them. Anthropic just watched this happen in real time, with its own top model turning into a ghost in the machine. It ran an experiment.
Google DeepMind unveils Gemini Robotics‑ER 1.6, beats prior model in tool count
A robot that reaches for a ghost is a robot that fails. That cold fact drives every line of code in Google DeepMind’s latest release. Gemini Robotics-ER 1.6 just landed, and it does something its predecessor couldn't: count tools accurately.
UK tests Mythos AI, noting its ability to chain multistep attacks
Last year, OpenAI's GPT-3.5 Turbo failed every one of the UK AI Safety Institute's basic hacking puzzles. That was the baseline. Now, Anthropic’s Mythos Preview cracks over 85 percent of them.
AI Forum Launches Professional Certificate and USD 120M Fund for AI Fluency
The gap between knowing about AI and actually wielding it is vast. Most people are stuck in literacy, they can name models, recite risks, maybe even dabble with a chatbot. That’s not enough.
Databricks finds multi-step agents beat single-turn RAG by 21% to 38% on STaRK
The numbers are stark: 21% in academic retrieval, 38% in biomedical. That’s the margin by which a multi-step agent outperformed a stronger, single-turn RAG model on the STaRK benchmark, a suite of semi-structured queries spanning product catalogs,...
Stanford AI Index 2026: 53% adopt generative AI in 3 years, education lags
Generative AI has pulled off a market penetration blitz. In three years, 53% of the population has used it, outpacing the historical adoption of personal computers and the internet. For younger people, the rate jumps to 80%.
NVIDIA, UMD release AF-Next audio model, beats Phi-4-mm by 12 points on Arabic
NVIDIA and the University of Maryland released a new audio model that significantly outperforms Microsoft’s on a key benchmark.
Developers Claim Measured Drop in Claude's Performance, Sparking Nerf Debate
People who use AI for real work are noticing a pattern. The tools get worse. Not all at once, but gradually, like a subscription service that quietly cuts corners. Now developers are saying they have proof Claude is doing exactly that.
Seven AI agents in finance lift cash flow >3% monthly, boost productivity 50%
Talk of AI in finance has become background noise, mostly unproven. But the quiet work of wiring a handful of specific agents into actual financial plumbing is showing concrete returns. One operation deployed seven.
Meta AI and KAUST Propose Neural Computers Merging Compute, Memory, I/O
For decades, the dream of a truly self-contained intelligence has been shackled by the same architectural division: a model that thinks, and a machine that runs it. What if that boundary simply vanished?
Prediction drift can mask security model decay despite stable accuracy
That model you trust to catch fraud? Its accuracy is probably stable. That's the problem. Security models don't just break. They fade.
Researchers say OpenAI's Sora and Google's Veo aren't true world models
Everyone got drunk on the same idea. OpenAI released Sora, and the tech press instantly crowned it a "world simulator." Google's Demis Hassabis said the same about Veo. The promise was a machine that grasped how things actually work.
TriAttention KV Cache Compression Matches Full Attention, 2.5× Faster
The key-value cache is the bottleneck of modern long-context inference, growing without bound as sequences lengthen, slowing throughput to a crawl. Pruning it has always meant trading accuracy for speed. Until now.
Knowledge Distillation Keeps Student Model Capacity to Match Ensemble Boundaries
An ensemble of heavy models can find subtle patterns, but running a whole committee of them is slow and costly. Knowledge distillation trains one small "student" model to copy the committee's behavior.
Featured Resources & Reviews
No Code MBA Course Review
Learn to build AI-powered apps without coding. Our comprehensive review of No Code MBA's course.
AI Tools & Resources
Curated collection of AI tools, courses, and frameworks to accelerate your AI journey.
Weekly AI Digest
Get the week's most important AI news delivered to your inbox every week.