weekend ai reads for 2024-03-01

She adds: “Analysis is more complex, and voice generation tools are more advanced than video generation tools. Even with voice samples and spectral analysis skills, it takes time, and there is no guarantee that the result will be accurate. In addition, there are many opportunities to fake audio without resorting to deepfake technology.”

PyRIT can generate thousands of malicious prompts to test a gen AI model, and even score its response. “For instance, in one of our red teaming exercises on a Copilot system, we were able to pick a harm category, generate several thousand malicious prompts, and use PyRIT’s scoring engine to evaluate the output from the Copilot system all in the matter of hours instead of weeks,” said Microsoft in the release.

In fact, I would say that Nvidia’s business today is probably, if I were to guess, 40 percent inference, 60 percent training. The reason why that’s a good thing is because that’s when you realize AI is finally making it.

The formula your AI company actually needs has these multipliers:(Data + Model) x UX x (Distribution + Perceived Value to Customers) = Your new AI MVP

It is basically impossible to differentiate yourself now when it comes to data and model. Sure, there will be some sources of unstructured data that may give you an edge for a little while. But ultimately, data isn’t enough on its own. As for models, most will be interchangeable.

To explore these opportunities, I built a gpt wrapper using FastAPI and Postmark that executes different scripts depending on the inbound email address. You can think of this as a personal platform for email-based custom GPTs – though I call each app a "HaiHai."

Yearsely says she thinks this future will mean having AI trying to tackle business challenges through a lens that takes in more than just productivity and customer service, and will give organisations an opportunity to play a much more intrinsic and helpful role in their customers’ lives.

Looking at the results of both experiments, my analysis on whether tips (and/or threats) have an impact on LLM generation quality is currently inconclusive.

NonprofitAMA is an AI chatbot built and trained on curated, high quality resources covering everything you want to know about nonprofits: leadership and management, governance, fundraising, marketing, incorporation, fiscal sponsorship, organizational design, strategy, advocacy, DEI, policies, etc.

However, one takeaway from this is that graphic design + web design jobs are still in demand, and not being replaced by AI tools yet.

Again, I think this is because tools like DALL-E and MidJourney requires some knowledge and creativity.

We're a couple of Microsoft developers that do this in our free time. Our journey began as volunteers, teaching courses at Purdue Polytechnic High School. Through this experience, we recognized the immense effort required to create lessons and the scarcity of easily accessible resources online.

Our inspiration? The open-source spirit prevalent in software development. We are driven to bring the same ethos to education, making it effortless for teachers to discover and create content that perfectly complements their teaching style and classroom needs.

Take hidden variables. These are features that are present in data, and which happen to be predictive of class labels within the data, but which are not directly related to them. If your model latches on to these during training, it will appear to work well, but may not work on new data.

Examining the 15 most widely used online news sources in ten countries, we find that by the end of 2023, 48% of top news websites across ten countries were blocking OpenAI’s crawlers. Around half as many (24%) were blocking Google’s AI crawler.

Existing benchmarks tend to focus on solving typical problems that might be assigned to a student as homework. But the types of questions that are assigned to students are different from the types of questions I want to ask a language model to solve for me.

DataDreamer is a powerful open-source Python library for prompting, synthetic data generation, and training workflows. It is designed to be simple, extremely efficient, and research-grade.

A new estimate from researchers at Harvard and the University of Toronto finds that the worldwide value of all open-source software is about $8.8 trillion. To place that in perspective, about $3.4 trillion was spent on commercial software globally in 2020.