Thinking with Speed
The Essence of Action in AI
In this edition:
My thoughts
Notable headlines
Learning, Tools, and Experiments
AI MISTAKES
My Thoughts
Is faster always better?
While quick results are appealing, the real value lies in starting—without action, speed is meaningless.
Starting is the most essential skill that we all need. It doesn’t matter how fast you can go without starting because you aren’t moving.
When it comes to getting things done, action is vital.
Regarding AI, the speed of inferencing for large language models (LLMs) is emerging as a critical factor in creating transformative user experiences and shaping the competitive landscape.
As industry experts and data suggest, rapid response times and efficient processing are becoming critical differentiators in the race to deliver more engaging and natural AI interactions.
This isn’t new information. Twenty years ago, amidst the Web 2.0 craze, we talked about how long a website’s visitor would be willing to wait for a response before giving up.
This hasn’t changed.
But when I talk about speed, it’s not just the user experience. It’s everything that comes along for the ride.
The exact details that taught us how long we should expect to wait on a web page also kept us returning to Google. Their results were fast and relevant and almost always led us to our desired answer.
Today, with the combination of cutting-edge telecommunications, computers, and AI, it’s faster than ever to find the answer.
When I built a prototype app with an LLM backend, I found that the processing delays from the best LLMs were not worth the wait. I was looking for real-time experience and willing to take a less detailed answer as a trade-off for speed. Granted, it was a particular use case, but now that I have seen what’s possible, I’m convinced this is one of the more critical factors.
And what I see in results from Google, Anthropic, and OpenAI is that all the major closed players agree. While they are all working to improve their models and results, all of them have worked to improve speed… because it matters.
Notable Headlines
Claude Gets an Upgrade with 3.5 Sonnet
Anthropic has released Claude 3.5 Sonnet, an improved version of their AI assistant. All the effort here went into Sonnet, and it’s quite a bit faster than Opus, the previous king at Anthropic. The new model has always been faster than Opus, but this feels better at writing and has improved software engineering skills. I’ve been super impressed with the new artifact capability—so much so that last week's live video was nearly entirely dedicated to Claude’s new Artifact feature. It’s available for free via Claude.ai, so go try it out now if you have not already!
Google Flash: Optimized for Speed
Google has introduced Gemini Flash, a smaller and more efficient version of their flagship AI model that is build for speed. It also includes a long-context understand which is something new for a smaller and faster model. 1 Million tokens. (also, this week google has extended their 2 million token capability to all developers!)
Perplexity Launches Shareable "Pages" Feature
Perplexity AI has unveiled a new feature that enables taking a search and quickly sharing pages. I’m testing things out regularly and I've found this to be interesting and could perhaps lead to a whole new set of opportunities.
Are you interested in more content from me that is augmented with AI, something like perplexity pages where I choose the topics but let AI do the heavy lifting for the writing?
AI Video Generation Tools Proliferate
Video generation is garnering lots of attention lately. Kling (Kling is a mobile-only app, so you'll need to use your smartphone or tablet to access and use the service), finding the app on the relevant mobile stores), Pika, RunwayML and Leonardi.ai have all announced new updates and are really making things interesting as we all are still waiting on OpenAI to make Sora more available. This explosion of video generation is only the beginning.
Learning Tools and Experiments
I’ve recently had a few people reach out to me asking about opportunities to learn AI and how to best prepare for the future of their careers by learning about AI.
Learning
My advice might seem a bit counterintuitive to some. Still, it is backed by my personal experience and the historical happenings in AI and how it has evolved since the early days of computing in the 1930s.
Don’t make AI your ‘thing’; make your ‘thing’ with AI.
AI is only called AI until we find a practical way to implement it to a valuable capability. So, please take what you have that is already valuable and find the best ways to use AI to make it better.
Learn about AI, but do it in a way that is immediately value-generating by actually applying it. Learning is fun and fabulous, but it’s not creating value without application.
The section is aptly named because you need learning, tools, and experiments.
So, about the learning, here’s a simple exercise you can use to learn.
Choose an LLM app (Claude 3.5 or OpenAI 4o) and ask it to explain everything about something you are an expert at.
Identify the MISTAKES
Write about what you discovered.
Share what you wrote and tell people where these models got it wrong based on your experience.
Did you pick up some learning from today’s newsletter? Help me out and share it with a friend.
Tools
In last month’s edition, I wrote that the doors have been thrown open wide. You don’t even have to pay to use the best tools. So what are you waiting for? Go and use them!
And just a few days ago, Google opened up 2 million tokens for Gemini for all developers. You’ll need to create a developer account, but there are many capabilities that you can test out totally for free. Google gives free access for new accounts for the first 90 days, so there’s no reason not to get started building if you have an idea you are ready to try out.
Use them all. Run the same prompt across all of them and compare. Do something you would already do (like with a Google search) and try it across multiple tools.
The tools are only the beginning, you have to get to the experiments to really lock-in your learning.
Experiments
I‘ve been animating using Anthropic Claude’s new Sonnet 3.5 nonstop. Just make sure you go and enable the Feature Preview for “Artifacts” option in your profile settings screen.
This is definitely worth a closer look. Don’t miss out on the chance to learn and create something cool.
Speaking of creating valuable things, it’s already clear that we will use AI to do many things, and agents will be part of our future.
It’s hard to know exactly how this will work out, but if you want to start working with agents in a small way (which is the best way to start), you might like to look into a new GitHub repo. It’s called micro-agent, and if you write web code, you should definitely take the time to check it out.
AI MISTAKES
Things don’t always go to plan.
Sometimes, even when you have the right idea, you’re ability to do it well or with enough care is a challenge.
That’s where we find ourselves again with one of the large tech companies.
Google’s had their turn, OpenAI too. Now, it seems that it’s Microsoft’s turn.
Microsoft has played the game and discovered that while moving fast with the idea is essential, but action and adapting fast is MOST essential. Having a good idea isn’t enough.
On May 20th, Microsoft announced new Windows 11 operating system features integrating AI capabilities into the core OS.
I’ve been an early adopter and advocate of some of their great new features, but not every new feature that touts AI is great for everyone.
After announcing the recall, Microsoft has decided to reverse how it delivers the experience to all users.
It is essential to respond quickly when conducting innovation work. I applaud Microsoft for responding to what the community and users have been saying since the feature was announced.
I think the ideas that Microsoft was after with recall are essentially good, but not unlike what we discussed about AI devices last month; implementing it well is no easy task.
Ultimately, the things we do on our computers will become increasingly integrated into our lives over time. The future of that integration could come through many ways, from external devices, auxiliary storage, physical augmentation and especially robots. So, it’s important to keep an eye on these developments. Not all companies or organizations are will to change their approach based on feedback, so the fact that a big tech company has adjusted their approach is to be celebrated. Our just is to not only contribute our imagination, but also think through the consequences of our actions.
We only have to observe the changes over the last decade in how people use smartphones to see how this works. When was the last time you used a phone number for someone you know? I know I haven’t.
But more than that, not everything is a pretty picture. Sometimes, even having a wonderful device like a smartphone can be a bad thing if we let it take over how we interact with the world around us. We have to be vigilant about not just what we CAN do with the tech, but what we WILL do with the tech. Always be present and mindful.
So, we will see the adoption of recall like features and abilities for AI, especially as they relate to the things we want to remember and further process for insights, it just need to continue to evolve.
And speaking of recall, the best way to learn is to go and make MISTAKES. Nothing teaches like experience, and nothing improves recall like realizing what we did last time wasn’t the best.
So let’s be sure to learn from our MISTAKES while also letting go of the parts we want to forget so we can move forward and keep heading fast toward the things we want.