Just weeks after releasing an API for ChatGPT, OpenAI announced GPT-4. Meanwhile, everyone is rushing to integrate GPTs into their product. I've listed 5 takeaways.

My open call for an accessibility passport: an easy way to set all relevant accessibility settings on a new device or platform. What do you think?

RT @viegasf: Toasters have blinking lights, cars have speedometers. Should chatbots have dashboards too?. A speculative essay👉 The System M….

RT @itpocket: While you're reading this, you could be reading Shift 24 - our annual trend report on the most important movements in busines….

An example of how dangerous this can be. LLMs don’t know anything about food safety but it will generate something anyhow. Here it’s easy to spot but what if it said to add 1 tablespoon of bleach to your rice? That could sound like a cooking hack…

RT @LeTour: 🔁 RT if you think 🇧🇪 @VCampenaerts is the Super-combative of the #TDF2023!. 🔁 RT si vous pensez que 🇧🇪 @VCampenaerts est le Sup….

Takeoff! @UXLondon

"Learning how to talk to an AI is going to be one of the most essential skills of the 21st Century" says @heykahn Is it? I hope we can make AI-powered products that work like users expect. Having to write long prompts for everything is'nt very usable. Let's move beyond the prompt.

RT @itpocket: OpenAI just dropped GPT-4. Now, companies are racing to integrate GPTs into their products. Everyone who follows should there….

Make sure to tailor your AI output and tone of voice to the context of your users and adapt it according to the underlying probabilities. As humans we express facts and guesses differently and our systems should do so too.

Creating awareness about mistakes is where these integrations fall short. Notion AI puts under each result “Al responses can be inaccurate or misleading.” That's the equivalent of car manufacturers putting “Objects in mirror are closer than they appear”.

These models obviously make mistakes. You need to provide an easy way for users to give feedback so you can follow-up and better understand where your suggestions were helpful and where they’ve failed.

Even better, go beyond the prompt. Define the main use cases and make it easier for your users to execute them directly. Notion AI does this very well by providing contextual prompt shortcuts.

"Every text box gets an LLM" says @benedictevans, but that's just a first step. Prototype and test with users to find the best and most natural way to integrate a prompt in your product.

This is an interesting take from Bing to expose some of ChatGPT's parameters in a way that's meaningful to users. More creative or more precise? Leave some agency to the user and create the right expectations in doing so.

As a user, none of this tells you that this is not per se _good_ advice. It spouts a response in a way that sounds very confident without any explainability. I hope we get to see some Model cards soon as a first step.

A worrying trend for me is integrating the likes of ChatGPT in a product without exposing any of its underlying confidence levels. This (arguably very interesting) approach from Instacart shows an LLM giving dietary recommendations for your kids.

And a valuable lesson for designing ML-powered products: create the right expectations about what your product can and can’t do as well as signal when you’re not sure!.

RT @scottbelsky: “Clippy” is back! 🙃.All productivity tools launch their own interface for queries. GTM advantage will rule as AI becomes w….

RT @itpocket: Sitting on a pile of data and don't know what to do with it? You still have time to register for our 'Leveraging your data' W….