AI.News Roundup: Week Ending July 11th

[#1]

Everybody is doing evvvvvverything. So much AI news from last week and here (some of) it is in one ginormous post.

Are y'all digging these round ups? I'm not anticipating that they're as compelling as say the 100 best books of the 21st century (even if they're just about as long) but lmk. 

Away we go! 🚀🚀🚀

Meta just released pre-trained models that use a multi-token prediction approach. Unlike traditional LLMs that predict just the next word, Meta’s approach predicts several words at once. The implications are a faster AI that is better at understanding language, bringing it closer to how humans think and speak.

CodiumAI is building many small AI agents that handle specific tasks within the software development workflow. These agents are meant to enhance code integrity, boost developer productivity, and provide comprehensive code analysis.

Cloudflare has introduced a new feature designed to prevent AI companies from scraping website content. This no-code tool helps identify and block bots used by AI developers to gather data for training their models. The feature assigns a trust score to each website request, making it easier to detect and stop bots even when they disguise their activities as regular user traffic.More AI news below

Anthropic’s Claude now has a prompt playground to enhance AI applications. Developers can now use Claude 3.5 Sonnet to generate, test, and evaluate prompts, improving Claude’s answers for specialized tasks. The new feature provides quick feedback, streamlining prompt engineering and saving time for both new and experienced users.

Princeton University researchers found that the current benchmarks used to evaluate AI agents can be misleading. Unlike foundation models, existing benchmarks for AI agents may not accurately reflect real-world performance, often favoring specific tasks or datasets that don't represent broader AI capabilities.The study emphasizes the need for more diverse and comprehensive evaluation practices in real-world applications.

OpenAI has set levels to track progress towards achieving superintelligent AI. Their aim is to establish milestones and benchmarks to monitor advancements in AI research and development. While the timeline for achieving superintelligence remains uncertain, this initiative provides a framework for safety and ethical considerations in the pursuit of more capable AI systems.More AI news below

The latest product news

• Perplexity's AI search could eliminate the need for follow-ups and beat ChatGPT at its own game (TechRadar)

• Anthropic’s Claude adds a prompt playground to quickly improve your AI apps (TechCrunch)

• AI agent benchmarks are misleading, study warns (VentureBeat)

• Meta drops AI bombshell: Multi-token prediction models now open for research (VentureBeat)

• How CodiumAI is accelerating AI agent driven enterprise application development (VentureBeat)

• Vector database company Qdrant wants RAG to be more cost-effective (VentureBeat)

• Cloudflare rolls out feature for blocking AI companies’ web scrapers (siliconANGLE)

Industry trends

• Intuit’s CEO continues to ‘bet the company on AI and data’ (Fortune)

• OpenAI Develops System to Track Progress Toward Human-Level AI (Bloomberg)

• Here’s how AI is changing workers relationship with data (Fast Company)

• AI is effectively ‘useless’—and it’s created a ‘fake it till you make it’ bubble that could end in disaster, veteran market watcher warns (Fortune)

• Tokens are a big reason today’s generative AI falls short (TechCrunch)

Responsible AI

• Bipartisan AI bill aims to give journalists, artists control over content (The Hill)

• Sam Altman Is Building the 'Titanic of AI,' Former OpenAI Employee Says (Newsweek)

• How AI’s Insatiable Energy Demands Jeopardize Big Tech’s Climate Goals (Mother Jones)

• AI with hidden biases may be subtly shaping what you think: ‘You may not even know that you are being influenced’ (Fortune)

[Huwen Arnone _Deiser_]

I'm loving these updates! Thank you @Kelly Stocker