GPT Makes 3 Scientific Discoveries, Grok BEATS ChatGPT, AI Analyzes Millions of Brain Cells: AI NEWS
Ottokit Link: https://ottokit.com/?aff=14ca2272 OpenAI just deployed 1,946 AI agents to analyze a massive mouse brain dataset in 32 minutes. The company also claims its AI models generated formal proofs of major unsolved problems in mathematics and computer science. Meanwhile, Elon Musk's xAI says Grok Voice 3.0 has surpassed ChatGPT's voice AI. Three major AI breakthroughs. Three sets of impressive numbers. But when you examine the evidence, the story gets considerably more complicated. OpenAI's GPT-Rosalind system used 1,946 parallel AI agents to analyze a whole-brain Perturb-seq dataset from a 2026 study by Sun and colleagues. Each agent examined a different gene perturbation, investigated cellular changes, reviewed scientific literature, and generated research hypotheses. According to OpenAI, the entire analysis took just 32 minutes, compared with months of estimated human research time. The multi-agent system referenced more than 700 scientific sources, while a single-agent approach cited only 11. One particularly interesting finding involved DCTN1, cholesterol-related genes, and potential connections to Parkinson's disease research. And an existence result involving quantum chromodynamics and related physical theories. OpenAI released machine-checkable formal proofs using Lean, a mathematical proof assistant that verifies logical steps. But here's the critical catch. A proof passing Lean verification doesn't automatically establish that the formalized statement perfectly represents the original mathematical conjecture. Independent mathematicians still need to examine whether the AI actually proved what OpenAI claims. These results remain under expert scrutiny, and the physics claim has not been fully published. If the proofs survive independent verification, the implications for mathematics, computer science, and AI-assisted discovery could be enormous. If they don't, the failures could reveal important limitations in AI-generated mathematical reasoning. Meanwhile, xAI has entered another major competition with OpenAI: real-time voice AI. Grok Voice 3.0, released October 14, reportedly outperformed OpenAI's GPT-Live-1 on the Artificial Analysis Speech to Speech Index. The independent benchmark numbers tell an interesting story: Grok Voice 3.0: 86.9 GPT-Live-1 (medium): 83.2 Grok also demonstrated significantly faster response times: Grok Voice 3.0: 0.81 seconds to first audio GPT-Live-1: 1.34 seconds That's approximately 40% less waiting before Grok starts responding. Grok also showed slightly lower costs per completed task on Big Bench Audio, at 2.5 cents compared with 3 cents for GPT-Live-1. But the benchmark measuring actual task completion tells a different story. On tau-Voice, which evaluates whether AI agents can successfully resolve multi-step customer service problems: GPT-Live-1: 74.5% Grok Voice 3.0: 71.6% So Grok leads the overall voice benchmark, but ChatGPT performs better on the test measuring whether customer problems actually get solved. And this competition is already moving beyond benchmarks. According to reporting from Business Insider, Grok is handling Starlink customer support calls. xAI reports approximately 20,000 calls daily, with around 70% resolved without human intervention. That's a significant real-world deployment, but it also means roughly 30% of calls still require human assistance. Privacy, reliability, complex customer interactions, and the risks of AI making confident mistakes remain important unanswered questions. These three developments reveal something important about the current state of artificial intelligence. AI agents can dramatically accelerate research, but generating hypotheses isn't the same as proving discoveries. AI can produce formally verified mathematical arguments, but verification still depends on whether the formal statements match the original problems. And AI voice assistants can outperform competitors on headline benchmarks while falling behind on the practical tasks businesses actually care about. The biggest question isn't simply which company has the most impressive announcement. It's which results survive independent verification, real-world testing, and human scrutiny. Subscribe to BitBiased for evidence-driven coverage of OpenAI, ChatGPT, xAI, Grok, AI agents, scientific breakthroughs, and the technologies shaping the future. CHAPTERS 00:00 Three Major AI Claims, Three Important Catches 00:46 What 1,946 AI Agents Actually Did to a Mouse Brain 02:15 Ottokit ShoutOut 03:58 Leads, Not Cures 05:34 OpenAI Says Its Models Produced Formal Proofs of Major Open Problems 06:54 What "Verified" Means Here, and What It Does Not 08:58 Grok Voice 3.0 Won the Headline Benchmark, but Not the One That Matters Most 10:27 The Test That Predicts Whether Your Problem Gets Solved 12:06 It Is Already Answering Real Phone Calls 12:53 What to Take From All of This #OpenAI #GrokVoice3 #AIAgents #ChatGPT #ArtificialIntelligence
