⚡️ You're Fired

Good morning. Have you wondered whether language models will ever replace programmers? A new research might have some answers. Let’s find out.

Today’s Highlights:

• NVIDIA AI unveils SteerLM for more tailored LLM responses

• Together raises a staggering $50m for AI chip reselling

• EveScape AI tool predicts virus mutations in time

DEEP DIVE

Will Language Models Replace Programmers?

To explore whether language models will replace programmers, Princeton University and the University of Chicago researchers have introduced SWE-bench—a framework that evaluates language models' capability to resolve real-world GitHub issues. The findings? Even top-ranking models struggle with complex issues, accentuating the need for further enhancements in LMs for intelligent software solutions.

The SWE-bench framework stands out by focusing on authentic software engineering problems like patch generation and complex context reasoning, offering a more comprehensive evaluation for strengthening LMs in software engineering—a field gaining traction as Machine Learning for Software Engineering (MLSE).

The evaluation results highlight the need for handling complex code changes and refining the generation of accurate and well-formatted patch files. Even the most advanced language models like GPT-4 and Claude 2 struggle to cope with practical software engineering problems, achieving pass rates as low as 1.7% and 4.8% respectively.

But will language models ever replace programmers?

The research appears to confirm that while the use of language models is increasing, the need for human oversight and the ability to handle nuanced programming problems remains crucial—at least for now.

The researchers suggest expanding the benchmark with additional software engineering problems, intensifying research on advanced retrieval techniques, and focusing on understanding ability towards complex code alterations.

Though machines haven’t completely taken over just yet—the exploration continues.

Keep reading here.

PUNCHLINES

Cash for Chips: AI chip reseller Together lands a whopping $50 million in financing driven by the demand for AI server chips.

Game of Drones: Ukraine deploys AI-powered autonomous attack drones for real-time combat.

From Streets to Sheets: AI is now being employed to aid homelessness prevention efforts.

Hey Alexa, Cast a Spell!: New LLM Mistral Trismegistus-7B fluent in occult sciences can help you read palms.

TLDR

NVIDIA AI unveils SteerLM for more tailored LLM responses: NVIDIA's new four-step AI method, SteerLM, allows users to fine-tune responses of LLMs in real-time, providing more customized outputs. The technology outperforms leading models like ChatGPT-3.5 and Llama 30B RLHF, and has wide applications from gaming to education. NVIDIA plans to release SteerLM as an open-source software.

AI tool EVEscape can predict virus mutations: Harvard and Oxford scientists develop AI tool EVEscape, capable of predicting virus mutations before they occur. The tool outperforms lab-based testing methods in speed and accuracy and is currently assessing SARS-CoV-2 variants and contributing to HIV and influenza treatment research.

Biden administration to limit China's AI chip access: US government plans new rules restricting China's access to advanced AI chipsets, targeting the sale of graphic chips and chipmaking equipment. The rules require licenses for export of prohibited technology and apply to Chinese companies, their overseas branches, and possible broker countries.

Microsoft and Adobe promote AI watermarking: Microsoft and Adobe are pushing for a system that adds metadata to AI-generated images, marking them with an icon to indicate they are machine-made. The icon functions as a transparency indicator revealing the source and edit history of an image. This watermark system relies on apps supporting the metadata to function and aims to tackle AI-created deepfakes.

TRENDING TOOLS

💻 Builder.io: Streamline design-to-code workflow by translating Figma designs into clean code

🧪 Leap: Quicken the process of designing, testing, monitoring, and deploying AI workflows

⏰ Motion: Organize your day with automated to-do lists and meeting schedules

🗂️ Pecan: Create impactful machine learning models with basic SQL knowledge

🎥 Translate.Video: Translate videos to over 75 languages with a single click

That’s all for today—if you have any questions or something interesting to share, please reply to this email. We’d love to hear from you!

P.S. If you want to sign up for the Supercharged newsletter or share it with a friend, you can find us here.