March 2024 - AI Tidbits Monthly Roundup

Devin AI revolutionizes software engineering, Claude 3 sets new benchmarks, Mistral Large rivals GPT-4, crucial LLM security insights, and the first open-source Mamba-based LLM supporting 256k tokens

Welcome to the monthly curated round-up, where we curate the firehose of AI research papers and tools so you won’t have to. If you're pressed for time and can only catch one AI Tidbits edition, this is the one to read—featuring the absolute must-knows.

---

Welcome to the March edition of AI Tidbits Monthly, where we uncover the latest and greatest in AI. This month has been filled with groundbreaking announcements from industry leaders and exciting progress in open-source AI, showcasing the rapid advancements in the field.

March saw the release of Cognition's Devin AI, the world's first autonomous AI software engineer, and Anthropic's Claude 3, setting new industry benchmarks across various domains. Mistral AI also introduced Mistral Large, a top-tier model rivaling GPT-4, now available on Azure through a new partnership with Microsoft.

In the realm of LLM security, research from UIUC, Cornell, DeepMind, and ETH Zurich highlighted potential cybersecurity risks and the urgent need for improved security measures against adversarial attacks.

Open-source initiatives continued to thrive, with AI21's Jamba, Sakana AI's Evolutionary Model Merge, xAI's Grok-1, and Apple's MM-1, showcasing novel approaches and achieving state-of-the-art results with enhanced efficiency.

Image and video generation also saw significant advancements, with Stability AI's Stable Diffusion 3, Alibaba's EMO framework for life-like portrait animation, and the introduction of YOLOv9 and GELAN for improved object detection.

Lastly, AI agents took center stage with DeepMind's Genie, transforming images into interactive 2D worlds, and SIMA, a versatile AI capable of following natural-language instructions across different video game environments.

These and many more exciting updates across various AI domains are part of this month's roundup.

Let's dive in!

---

Overview

• Industry Announcements (6 entries)

• ✨ Special feature: LLMs Security and Safety (4 entries)

• Large Language Models

  • Open-source (11 entries)

  • Research (6 entries)

• Autonomous Agents (4 entries)

• Image and Video (12 entries)

• Audio (2 entries)

• Multimodal (3 entries)

• Open-source Packages (7 entries)

• AI tools (2 entries)

Recent Deep Dives

Top 8 leaderboards to choose the right AI model for your task

[cross-post] 7 methods to secure LLM apps from prompt injections and jailbreaks

12 techniques to reduce your LLM API bill and launch blazingly fast products

Harnessing research-backed prompting techniques for enhanced LLM performance

Most popular and upcoming Generative AI tools and APIs

Industry announcements

1. Cognition releases Devin AI - the world's first autonomous AI software engineer, excelling in complex tasks and learning from feedback, outperforming in real-world coding benchmarks

2. Anthropic announces Claude 3 - three state-of-the-art language models, setting new industry benchmarks across reasoning, math, coding, multilingual understanding, and vision

3. Mistral AI releases Mistral Large - a top-tier model that rivals GPT-4 with advanced multilingual reasoning and competitive pricing, now available on Azure as part of a new partnership with Microsoft

4. Ideogram releases Ideogram 1.0 - a text-to-image model excelling in text rendering and photorealism

5. Figure partners with OpenAI to enhance its humanoid robot, Figure 01, showcasing human-like communication and reasoning in an unprecedented demo

6. Nvidia unveils Blackwell at GTC 2024 - its next generation and the world's most powerful AI superchip

Become a premium member to get full access to my content and $1k in free credits for leading AI tools and APIs like Perplexity, Replicate, and Hugging Face. It’s common to expense the paid membership from your company’s learning and development education stipend.

✨ Special feature: LLMs Security and Safety

1. UIUC shows that GPT-4 can autonomously hack websites, performing advanced tasks like SQL injections and finding vulnerabilities, highlighting potential cybersecurity risks

2. DeepMind and ETH Zurich introduce a novel attack capable of extracting detailed information from black-box language models, determining the exact hidden dimensions of notable models like OpenAI's ChatGPT for under $20

3. Cornell unveils Morris II - a computer worm targeting GenAI systems, demonstrating the urgent need for improved security against adversarial self-replicating prompts

4. Scale and the Center for AI Safety release the WMDP benchmark - a safety evaluation for LLMs to gauge their knowledge in biosecurity, chemical security, and cybersecurity

   

   UIUC successfully leverages GPT-4 to automatically hack websites

Large Language Models (LLMs)

Open-source

1. AI21 open-sources Jamba - a pioneering Mamba SSM-Transformer model, enhancing AI performance with a 256K context window and tripled throughput on long contexts