AI News Roundup – Accelerated drug discovery, AI in the fashion industry, watermarking AI-generated speech, and more

To help you stay on top of the latest news, our AI practice group has compiled a roundup of the developments we are following.

• • Researchers at Terray Therapeutics in Monrovia, California have made use of AI technologies to accelerate drug discovery, according to a report from The New York Times. Using the massive amount of data generated from the company’s experiments and drug development – including molecular information, protein structures and measurements of biochemical interactions, the company trained a generative AI model that digitally designs a new drug molecule, which the automated lab then synthesizes and tests. The results are then fed back into the model to improve future results. Drug development is often a lengthy and expensive process, often taking up to 15 years and costs up to $1 billion dollars on average to develop a new drug, pass through clinical trials, and receive approval. Indeed, Terray receives much of its funding from large pharmaceutical companies, who hope to cash in on the savings afforded by the new technologies. While AI-developed drugs have yet to pass clinical trials (Terray itself expects to have products in trials by 2026), the technology has the potential to greatly accelerate the pipeline for getting new and innovative drugs to market.

• • Anthropic released the newest version of its Claude large language model, dubbed “Claude 3.5 Sonnet.” According to the company, the new model operates twice as fast as its previous model, Claude 3 Opus, and outperforms its rivals (OpenAI’s GPT-4o and Meta’s Llama-400b, among others) in a variety of reasoning, knowledge and code generation proficiency benchmarks. A demo video provided by the company shows the new model transcribing a timeline and graph relating to genome sequencing milestones, and eventually generating a class presentation on the subject. Another new feature called “Artifacts” allows Claude to display generated code snippets, text documents or website designs in a dedicated window. Anthropic says these features mark “Claude’s evolution from a conversational AI to a collaborative work environment,” and that they plan to expand Claude to support team collaboration. Anthropic plans to release other models in the Claude 3.5 family (“Haiku” and “Opus”) later this year.

• • A new report from the Reuters Institute for the Study of Journalism indicates that consumers are generally wary of AI-generated news content. A survey of 100,000 people in 47 countries revealed that 52% of U.S. respondents and 63% of UK respondents said they would be uncomfortable with news produced mostly with AI, with 59% of global respondents saying that they were concerned about “fake news” content online. Nic Newman, the lead author of the report, said that “people broadly had fears about what might happen to content reliability and trust.” However, the report noted that respondents were more comfortable with the use of AI by journalists during the writing of articles or to make their work more efficient.

• • Vogue Business reports that generative AI technologies are gaining a foothold in the fashion industry, especially in streamlining tasks like copywriting, coding, customer service, clienteling, photo editing and 3D design and illustration. The French fashion conglomerate LVMH has collaborated with a Chinese startup that uses generative AI to produce videos from 3D product models and creative briefs, while The Estée Lauder Companies are making use of AI for measuring customer sentiment, aiding in writing copy and making product recommendations. Additionally, fashion companies are beginning to hire candidates that have AI competency, especially for data-driven roles, though companies are also investing in training current employees to make use of AI technologies – Estée Lauder in particular has collaborated with Microsoft to create an AI innovation lab for this purpose. However, some consultants warn fashion companies to ensure that the human element remains in fashion design, with one saying to Vogue Business, “the companies that will win are those open to embracing new solutions that blend technology with a human-centric approach, ensuring that AI enhances — rather than replaces — the unique value that only people can bring.”

• • Meta has developed a technique to watermark AI-generated speech, according to the MIT Technology Review. The tool, known as AudioSeal, makes use of two neural networks: one to generate watermarks to be embedded in audio tracks, and the other to detect the watermark. Current techniques generally comb through an entire audio segment in one-second chunks, which can often be slow. In contrast, AudioSeal embeds a watermark in each section of the audio track, which allows for “localized watermarks” that can still be detected even in audio that has been cropped or otherwise edited. According to the whitepaper, AudioSeal achieved an average of 96% detection accuracy across various editing techniques. However, the researchers noted that as more information is disclosed about the watermarking algorithm, the more vulnerable it is to attacks. Additionally, the system depends on voluntary watermarking of content, which may limit its applicability when malicious actors are involved.