Google's quest for an AI future as it turns 25

Welcome to The Dispatch! We are the newsletter that keeps you informed about AI. Each weekday, we scour the web to aggregate the many stories related to artificial intelligence; we pass along the news, useful resources, tools or services, technical analysis and exciting developments in open source. Even if you aren’t an engineer, we’ll keep you in touch with what’s going on under the hood in AI.

California scientists are using AI to track air pollution from coal trains

From Undark: In California, researchers are embracing AI as a tool to help residents track air pollution from sources like coal trains. The team’s data showed that, as they passed, trains full of coal traveling through the city of Richmond significantly increased ambient PM2.5, a type of particulate matter that has been linked to respiratory and cardiovascular diseases as well as early death. Even short-term exposure to PM2.5 can harm health.

More details:

• In Richmond, scientists trained AI to constantly monitor passing coal trains and their dust emissions. By automating identification of the trains, the AI system captured vastly more data than previous manual methods - and revealed major spikes in particulate matter from the coal trains.

• The technique finally provided concrete evidence to support long-standing concerns over the pollution impact of open-top coal trains. Activists see potential to adapt the methodology elsewhere as scientific backup for environmental justice efforts.

• Broader advances in AI are also being applied in air quality monitoring, and experts believe AI can revolutionize pollution data collection by tracking hard-to-pinpoint sources.

Takeaways: The previous three studies on North American coal trains combined gathered data on less than 1,000 trains. The Davis researchers were able to collect data from more than 2,800. Even one of the researchers who was “not a fan of AI” could not help but laud the technology here. This is a great use case for current-gen AI, highlighting how the technology can already interact with/generate large amounts of data to tell stories that humans have a hard time providing concrete evidence for. Additionally, the green algorithm calculator hints at a way for environmentalists to keep track of just how green their efforts are, given concerns over AI’s impact on the environment.

Specific pollution levels were not noted in the article.

Just as the rise of mobile devices made so much desktop software obsolete, AI could eliminate the need for complex user interfaces and screen after screen of data. The next generation of software won't require human input or spit out results to parse. Instead, AI will simply receive instructions and execute decisions on its own. This elegance and simplification could spell the demise of today's screen-heavy, data-dense SaaS model. Companies diving into AI aren't merely planning for the future - they're scrambling to adapt before their businesses are rendered obsolete.

The assistant can summarize long chat threads, coach users on presentation skills, and even interact conversationally to initiate actions across Zoom and connected apps in the future.

Zoom states its AI will use a federated learning approach, incorporating its own and third-party AI models while prioritizing customer privacy and trust. That trust is at an all-time low due to continuing concerns about the company’s AI training practices.

• Google's search for an AI future as it turns 25

• Generative AI at an inflection point: what’s next for real-world adoption?

• Samsung plans to add generative AI to home appliances next year

• Israeli doctors save man’s life with AI

• UK government sets out AI Safety Summit ambitions

Refact achieved 32% on HumanEval pass@1, surpassing comparable models like StableCode and ReplitCode while being much smaller. The model supports code completion, fill-in-the-middle, and conversational abilities across 20 programming languages. With the smaller 1.6B size, it can run efficiently on most modern GPUs for real-time coding assistance. The researchers aim to continue improving Refact through community feedback and contributions.

Currently, 24% of respondents already utilize AI for programming tasks. Use cases range from chatbots in documentation to test generation, code summarization, and code suggestions. Beyond code generation, AI assists with documentation, testing, maintenance and more. Despite valid worries, developers see enough value to push past obstacles and utilize generative AI across software projects. The survey confirms broad integration of AI capabilities into programming roles in coming years.

• (Meta research) Evaluating the fairness of computer vision models

• Renumics Spotlight: interactively explore unstructured datasets from your dataframe

• a16z’s open source AI grant program

• Introducing BabyFoxAGI, newest mod of BabyAGI