When ML meets Product: June ’24 AI Product Updates

Keeping up to date with new AI products, use cases, trends, and resources

June ’24 AI Product Updates, image by author

Welcome to the June edition of “When ML Meets Product — AI Product Updates”. With the rapid advancements in Artificial Intelligence, staying up-to-date with the latest developments and understanding their business and product implications is more crucial than ever. I find myself constantly checking a wide range of sources, including newsletters, Medium blog posts, news outlets, and industry resources.

In this edition, I’ll cover the most relevant AI product-related news, use cases, trends, and resources from recent weeks:

• Product & Business Trends: Explore how Generative AI is reshaping ML team strategies and their day-to-day work.
• GenAI Model Updates: Discover the latest updates from major players in foundational models, big tech releases, and advancements in image, video, and voice generation products.
• Other Relevant Resources and Future Events: Stay informed with additional resources and upcoming events.

Product & Business Trends: How GenAI is reshaping ML team strategies and day-to-day Work

Choosing Use Cases Wisely

It can be really hard to select use cases for GenAI as current models still have limitations, but at the same time things are advancing so fast and new versions increase model’s capabilities widely. However, many people in the industry agree: rather than overworking to fix current LLMs limitations, consider building on top of them and create solutions that offer a good user experience while addressing real user pains.

If there are specific issues that current versions can’t solve but future versions likely will, it might be more strategic to wait or to develop a less perfect solution for now, rather than to invest in long-term in-house developments. For instance, incorporating features that allow users to edit or supervise the output of large language models (LLMs) can be more effective than aiming full automation with complex logics or in-house fine-tuning.

Differentiation in the market won’t come from merely using LLMs, as these are now accessible to everyone, but from the unique experiences, functionalities, and value products can provide through them (If we are all using the same foundational models, what will differentiate us?)

Challenging the Status Quo

Many ML teams and Data Scientists are accustomed to developing traditional ML systems, but the world is changing and challenging that default makes more sense than ever before. We are shifting from using numerous in-house specialized models to a few very large multi-task models owned by external companies, and this change can reduce development and maintenance costs significantly.

Consider an NLP classifier: the “traditional ML” way involves data collection, labeling, model training, evaluation, deployment, monitoring, and maintenance. The “new way” now involves: selecting an LLM, performing prompt engineering, evaluating, and using an API in production. When challenging the traditional vs the new way, key factors to consider include development time, running costs, maintenance costs, and specific requirements such as latency or privacy.

The Evolving Role of Data Scientists and Machine Learning Teams

With these changes, one might question the value of DS and ML teams. While it’s true that GenAI APIs enable teams with little ML knowledge to implement AI solutions, the expertise of DS and ML teams remains of big value for robust, reliable and ethically sound solutions. Their contributions include:

• Model Understanding: Knowledge on how predictive models work, training process, the limitations, treating edge cases…
• AI Ethics and Risk Management: Awareness of AI ethics and risks, which allow to implement measures to mitigate biases and other risks, such as diverse prompting and selecting less biased models for sensitive applications.
• Evaluation: Just as with traditional ML solutions, it’s crucial to evaluate GenAI solutions for error rates, hallucinations, usefulness, and risk of harm. DS teams are experts in designing metrics and evaluating models against these criteria.

GenAI Model Updates

Foundational model updates

The war between the main GenAI model developers (still) continues. In the last few weeks we have seen:

• Google releases Gemini 1.5 Pro with added features like bigger multimodal context, Gemini live, and more connections to other apps (e.g. interact with youtube through Gemini).
• OpenAI releases GPT4-o and makes further capabilities accessible to free users.

Google and OpenAI seem to go into a similar direction: full real time multimodality, enabled from any gadget you want, and using multiple sources of context for the model (including your mic, camera, or gadget’s screen). It was definitely impressive to watch OpenAI’s demo with real time human-like voice responses (which also brought a controversy with Scarlet Johansson), capacity to interrupt the response, and ease of use from a smartphone.

• Anthropic releases Claude 3.5 Sonnet, with added features like improved vision capabilities, speed, and artifacts (where users can interact with creations from Claude such as code snippets)

Anthropic seemed to go into another direction, still UI based, improving multimodality (you are able to input pdfs, pictures, as well as text), and going one step forward in co-creation of outputs like interaction models through React components.

Caption of Anthropic artifact usage, impage by author

Other big tech updates

• Microsoft presents Copilot+ PCs: Windows PCs designed to fully integrate AI capabilities such as recall (which enables finding something you have previously seen in the screen), and image creation and edition.
• Apple Intelligence: Mac, iPhone, and iPad leveraging AI to help users write, prioritize, transcribe, create and edit images, and interact with an improved version of Siri.

Both Microsoft and Apple seem to be moving into a similar direction, integrating GenAI into specific functionalities to help empower users, while running most of the models on-device to preserve privacy and reduce risks (I have to admit, privacy is my main concern as a user when considering all these models and functionalities!).

Other GenAI product updates

A lot is happening with video generation:

• Cool Product #1: Captions AI — AI Ad Generator, use AI to instantly generate top-performing video ads in just one click.
• Cool Product #2: Adobe is adding AI video generation tools
• Cool Product #3: Luma Labs dream machine makes high quality, realistic videos fast from text and images.

In the meantime, AI keeps revolutionizing new industries:

• AI in Hollywood, much more used and with a bigger short term impact than expected (even despite recent labor actions against it).
• AI in UX/UI design, with great potential to improve the design process, the design-to-code translation and the code generation.
• AI as an accessible opportunity, thanks to features like eye tracking, music haptics, and vocal shortcuts.

📝 Relevant resources & events

AI Act

On May 21, the Council of the European Union finally approved the AI Act. The act classifies the AI systems based on their societal risk, introducing prohibited AI practices, and demanding requirements especially for high risk systems.

Reforge Ref:AI

Reforge is hosting Ref:AI with a number of Product — GenAI relevant talks on June 25th.

Globant NXT Conference

Beyond Data & AI: What’s NXT?, taking place online on June 27th.

RecSys learners virtual meetup

RecSys Learners Virtual Meetup will take place only on the 30th of June.

HackBCN — AI Edition

AI Hackathon taking place in Barcelona on the weekend of 29–30th of June. I am especially looking forward to it as I’ll be part of the jury!

Wrapping it up

That was it from When ML meets Product — June’24 AI Product Updates. Hope you enjoyed the read! I’ll be happy to hear your thoughts, questions, or suggestions.