Allen AI and DeepSeek are Taking Off, Professional Advice for Working in AI, Agentic Web Design, and More

Society's Backend Reading List 12-2-2024

Here's a comprehensive AI reading list from this past week. Thanks to all the incredible authors for creating these helpful articles and learning resources.

I put one of these together each week. If reading about AI updates and topics is something you enjoy, make sure to subscribe.

Society's Backend is reader supported. You can support my work for 80% off for the first year (just $1/mo). You'll also get the extended reading list each week.

A huge thanks to all supporters. 😊

What Happened Last Week

I have three recommendations to stay updated on AI news and happenings:

• Last Week in AI Podcast by

  Last Week in AI
  goes over recent major developments including OpenAI’s datacenter and new AI chips.

• The Batch continues to be a great resource and goes over DeepSeek updates, robotics, the Amazon and Anthropic partnership, and more.

• Devansh
  ’s content recommendations are always great. He includes whatever books, papers, videos, etc. he finds interesting and insightful (kinda like I do here!)

• For staying updated on AI papers, I recommend

  elvis
  ’s weekly overview.

Last Reading List

If you missed the last reading list, here it is:

Chinese AI is Less Expensive, What it's Like to Work in AI, an Evaluation Framework for Voice Agents, and More

Enjoy!

Reading List

What it's Like to Work in AI and Advice from 10 AI Professionals

The article discusses insights from 10 AI professionals about their roles, daily tasks, and qualifications needed to work in AI. They emphasize the importance of strong foundational knowledge in mathematics and machine learning, as well as hands-on experience. Many professionals recommend continuous learning and independent thinking for those interested in pursuing a career in AI.

Source

GitHub - NirDiamant/Prompt_Engineering

The GitHub repository by NirDiamant offers a wide range of tutorials and techniques for mastering prompt engineering, suitable for both beginners and advanced users. It provides practical implementations and guidance on effectively working with large language models. The community is encouraged to contribute and share innovative ideas to enhance the resource further.

Source

A new golden age of discovery

By

AI Policy Perspectives

Scientists are increasingly using AI to solve important problems and transform scientific practices. This trend brings both opportunities and risks, but with the right policies, AI can enhance creativity and reliability in research. To maximize its benefits, there is a need for interdisciplinary collaboration and new methods to assess AI's capabilities in science.

Source

QwQ: Reflect Deeply on the Boundaries of the Unknown

QwQ is an experimental AI model that explores deep questions and reasoning, embodying a curious and humble approach to learning. It shows promising abilities in mathematics and programming, achieving notable scores on various benchmarks, but it also has limitations like language mixing and circular reasoning. The journey of QwQ highlights the importance of reflection and questioning in the pursuit of knowledge and understanding.

Source

Deepseek: The Quiet Giant Leading China’s AI Race

DeepSeek is a discreet but ambitious AI startup in China that focuses on foundational research rather than commercialization. Its leader, Liang Wenfeng, aims to shift the Chinese tech landscape towards original innovation, challenging the norm of following rather than creating. By prioritizing open-source strategies and leveraging local talent, DeepSeek hopes to contribute significantly to global technological advancements.

Source

OLMo 2: The best fully open language model to date

OLMo 2 is a new family of open language models that outperforms previous models and is competitive with popular open-weight models like Llama 3. It features improvements in training stability and utilizes advanced training techniques to enhance performance. Additionally, OLMo 2 includes instruct variants that excel in instruction-following and reasoning tasks.

Source

NYT's "AI Outperforms Doctors" Story Is Wrong [Guest]

By

Devansh

The New York Times misinterpreted a small study that claimed AI outperformed doctors, as it actually only improved diagnostic reasoning, not diagnosis accuracy. The study had a sample size of just six cases, which raises serious questions about its validity. Overall, the article highlights the dangers of sensationalized media coverage and the need for accurate representation of scientific findings.

Source

Tülu 3: The next era in open post-training

By

Nathan Lambert

Tülu 3 is a new open-source recipe for improving language models after their initial training. It introduces advanced techniques like scaling preference data and using reinforcement learning with verifiable rewards. This approach aims to enhance model performance while making the process accessible to anyone.

Source

How Databases Keep Passwords Securely

By

Neo Kim

This post explains how to securely store passwords in a database using three key techniques: hashing, salting, and stretching. Hashing transforms passwords into unique fingerprints, salting adds a random string to make each fingerprint unique, and stretching makes it harder for hackers to crack passwords by applying the hash function multiple times. Together, these methods help protect user passwords from various attack methods.

Source

Agent-Responsive Design: Rethinking the web for an agentic future

By

Sahar Mor

The internet is evolving to prioritize AI agents, changing how we design websites and digital interactions. Companies must adapt to "agent-responsive design," which focuses on optimizing sites for AI rather than human users. This shift promises to revolutionize online engagement, making businesses that embrace it more competitive in a rapidly automated economy.

Source

Meet CircleMind: An AI Startup that is Transforming Retrieval Augmented Generation with Knowledge Graphs and PageRank