Google Unveils ‘AI Co-Scientist’ To Supercharge Research Breakthroughs

digitex — Thu, 20 Feb 2025 01:39:12 +0000

Google’s “AI co-scientist” aims to make research by humans faster and more efficient. getty
Scientists, get ready to meet your new collaborator. Google on Wednesday announced an “AI co-scientist” designed to help human researchers accelerate discovery. In one early test, the system solved a scientific mystery, one that had stumped scientists for more than a decade.

The AI co-scientist is designed to generate novel testable hypotheses, detailed research overviews and experimental protocols, all with the goal of making scientific and biomedical research faster and more efficient.

The tool is built on Gemini 2.0, the latest version of Google’s chat-based AI assistant that responds to user prompts like other chat-based large language models such as OpenAI’s ChatGPT. To work with the AI co-scientist, human scientists specify their research goal in natural language. They can also suggest their own ideas and proposals and offer feedback and reviews.

“AI co-scientist is a collaborative tool to help experts gather research and refine their work — it’s not meant to automate the scientific process,” Google said in a blog post announcing the new system, seemingly aiming to ease fears about artificial intelligence replacing humans in a host of fields.

For now at least, “AI co-scientist” is the system’s only name. The tool promises to assist humans in a variety of ways, such as shortening the time it takes to review detailed literature across fields researchers might be unfamiliar with. It can also directly supply hypotheses to be tested experimentally by scientists who bring the gray matter.

The AI co-scientist currently is only available to researchers participating in Google’s new Trusted Tester Program, which involves around 20 principal researchers, a company spokesperson clarified over email. Those interested in participating in the program can fill out an online application.

Same Hypothesis In Less Time
Early testers have already seen promising results with the model. At University College London, scientists have spent a decade studying superbugs resistant to antibiotics, proving how certain bacteria contribute to antibiotic-resistant infections, a widespread health challenge.

Given its relationship with Fleming Initiative, which works to control the spread of antimicrobial resistance, Google asked the UCL team to see how the AI co-scientist would react to the same problem.
“When the Google research team approached us to test its AI platform, we realized we needed to task it with the same scientific questions that we had already explored ourselves and used as the basis of our experimental work,” José Penadés, a professor with Imperial’s Department of Infectious Disease, said in a statement.
“This effectively meant that the algorithm was able to look at the available evidence, analyze the possibilities, ask questions, design experiments and propose the very same hypothesis that we arrived at through years of painstaking scientific research, but in a fraction of the time.”
Safety And Ethical Concerns
In a detailed report on the AI co-scientist, Google addresses limitations of the system, and acknowledges the need for technical safeguards against unethical research queries and malicious user intent. Just this month, Google warned of Gemini misuse by cyercriminals, raising the specter that sensitive or confidential scientific queries could fall into the wrong hands. The AI co-scientist currently has some safeguards in place, the paper notes, but it says more will ultimately be needed.
Nonetheless, scientists who’ve experimented with the system express enthusiasm about its potential.
“What our findings show is that AI has the potential to synthesize all the available evidence and direct us to the most important questions and experimental designs,” said Tiago Dias da Costa, who co-led the experimental work from Imperial’s Department of Life Sciences and the Fleming Initiative. “If the system works as well as we hope it could, this could be game-changing; ruling out ‘dead ends’ and effectively enabling us to progress at an extraordinary pace.”

Source link

Grok 3 Impresses — but Early Reactions Suggest OpenAI Remains Ahead

digitex — Tue, 18 Feb 2025 14:09:51 +0000

Elon Musk’s xAI has rolled out Grok 3, claiming it outperforms OpenAI’s GPT-4o in key benchmarks.Grok 3 introduced DeepSearch, a tool for improving research and reasoning.Some AI experts said Grok 3 is a solid contender but remains behind OpenAI in key areas.Elon Musk on Monday introduced Grok 3, the latest version of xAI’s chatbot, calling it “an order of magnitude more capable” than its predecessor.Some AI researchers and experts put Grok 3 through early tests — and their initial reactions have been mixed.Musk claimed the new model surpasses OpenAI’s flagship GPT-4o and competing AI systems from Google, DeepSeek, and Anthropic on key benchmarks in math, science, and coding.”We’re continually improving the models every day, and literally within 24 hours, you’ll see improvements,” Musk said during a livestream presentation on his social media platform X alongside three xAI engineers.Andrej Karpathy, the computer scientist who cofounded OpenAI and previously led Tesla’s AI operations, said in a Monday X post that Grok 3 successfully completed tasks such as looking up complex information from a training paper. However, like OpenAI’s o1-pro, it struggled to solve “tricky” tic-tac-toe boards, he said.”The impression overall I got here is that this is somewhere around o1-pro capability, and ahead of DeepSeek-R1, though of course we need actual, real evaluations to look at,” added Karpathy, who had early access to Grok 3.xAI also introduced a new tool for Grok 3 called DeepSearch, which aims to improve research, brainstorming, and data analysis while explaining how it processes queries and formulates responses. The startup had been working on the tool since at least late last year, Business Insider previously reported. Karpathy said DeepSearch is “approximately around Perplexity’s DeepResearch offering, which is great, but not yet at the level of OpenAI’s recently released ‘Deep Research,’ which still feels more thorough and reliable.”Ethan Mollick, a professor at the University of Pennsylvania’s Wharton School who researches AI, said that while xAI has scaled rapidly, Grok 3 is not a clear leader.”A very solid frontier model, but not one you would stop using your current frontier model for,” Mollick wrote in a Monday LinkedIn post.He added that while Grok 3 outperforms some OpenAI models in select benchmarks, it does not clearly surpass OpenAI’s o3.Gary Marcus, the founder of Geometric Intelligence and a longtime critic of AI hype, was more skeptical.”Elon Musk promised that Grok 3 would be the smartest AI ever. Spoiler alert: it wasn’t,” Marcus wrote on Substack.He called the launch event a “carbon copy” of previous demos, adding that while Grok 3 shows progress, there was “no game changer” in its performance compared to OpenAI models.Karpathy was more positive, calling Grok 3’s reasoning capabilities “state of the art” and highlighting its ability to solve complex coding problems.

professor – Digitex Solutions

Google Unveils ‘AI Co-Scientist’ To Supercharge Research Breakthroughs

Grok 3 Impresses — but Early Reactions Suggest OpenAI Remains Ahead