Welcome to The AI Weekly Breakthrough, a roundup of the news, technologies, and companies changing the way we work and live
Gemini’s Medical Fellowship Training
Two weeks ago, we explored the ability of LLMs in taking medical board exams across five specialties. Google’s research team has taken AI further by training Gemini on a series of medical-specific tasks. Through self-training with web search integration, fine tuning and customized encoders, and chain of reasoning prompting, Google researchers have enabled Gemini to achieve state-of-the-art performance in 10 out of 14 medical tasks. The researchers show Gemini’s impressive ability across multiple use cases in the recently published paper.
- Examining an image to provide a patient with care guidance
- Assisting a physician understand an X-ray through AI-enabled conversations
- Educating medical students by interpreting surgical videos and providing feedback
- Examining extensive patient records to answer clarifying questions
With Gemini’s fine-tuned LLMs, these advancements aren’t just theoretical – they are unfolding now.
Atlas, Boston Dynamics’ New Exorcist-Lite Robot
Boston Dynamics has introduced their new fully electric robot, Atlas. While some of Atlas’ body movements are a bit reminiscent of certain scenes from “The Exorcist,” as their promotional video shows, Atlas nevertheless signals a significant leap forward in robotic potential. Atlas is built for practical utility in real-world environments. It can navigate physical areas, move objects and potentially even cook, transforming what was once science-fiction fantasy into achievable innovation. Atlas hints at the future potential of combining the reasoning capabilities of LLMs with a robot’s precise physical capabilities.
Founders Forecast an AI Surge
Hampton just published their 2024 AI Founder Report, which offers valuable intel on AI’s impact on business from a founder’s perspective. According to the survey, a resounding 90% of founders are using AI or actively integrating it into their businesses, while just 3% report having no intention of adopting AI. Here are four other key takeaways from the report:
- Companies with revenue under $5 million annual revenue are using off-the-shelf AI solutions predominantly for content creation, communication, and task management, focusing on efficiency and cost-saving benefits.
- Companies with $5+ million annual revenue are more likely to allocate budgets for custom AI development, and are primarily targeting use cases in workflow optimization, customer service, and data analytics.
- The sectors leading with the highest AI R&D expenditures are Finance, followed closely by Agtech & Biotech, and in third, SaaS. Investment levels in all other sectors pale in comparison to these forerunners.
- GenAI has not yet caused significant staffing shifts, with most businesses experiencing no or minimal changes in team sizes. Significant downsizing attributed to AI is limited to just 6 of the 100 founders interviewed.
China’s SenseNova 5.0 Closes the AI Gap
While the majority of AI innovations as of late have come from U.S. institutions, a new formidable contender to GPT-4 Turbo was just released by China’s SenseTime. SenseNova 5.0, with 600 billion parameters and a context window of 200,000, surpasses most NLP benchmarks previously dominated by its well-known counterpart. What sets this model apart is not just its impressive capabilities, but also its groundbreaking use of almost exclusively synthetic data, suggesting an era where AI is trained on an endless, tailored stream of domain-specific data created by Generative AI. Initial results from SenseNova 5.0 testing show that the technology gap in AI between the U.S. and other countries is now very small if there even is a gap anymore at all.
Narrative Attacks Top Forrester’s Top Cybersecurity Threats
Four out of the top five cybersecurity threats facing businesses are tied to AI, according to a new report from Forrester. At the forefront, narrative attacks using Generative AI can rapidly disseminate misinformation, undermining trust and influencing major events like elections. Similarly alarming, deepfakes can create indistinguishable synthetic identities capable of deception on a massive scale. The report also highlights the vulnerability of AI systems to prompt injection attacks, where malevolent actors can manipulate AI to extract sensitive data. Further complicating the security puzzle is the exploitation of AI software supply chains, a potential Achilles heel in a network’s defenses. As we tread further into an AI-dominant future, these insights from Forrester emphasize the critical need for vigilant, AI-informed cybersecurity measures.
Augment Yourself 🤖
- Find the balance between benefits and ethics in Generative AI for Healthcare.
- Explore the critical role of data quality in AI implementations.
- How can synthetic data supercharge your AI initiatives?
- Can AI governance frameworks protect you from GenAI’s risks?
- Read this guide to preparing structured and unstructured data for GenAI.
🔥 For more AI News brought to you via email subscribe to our newsletter here.
👀 Want to know more about Shelf’s suite of AI solutions? Check out our website here.
