Static Software Will Die Out Faster Than You Think
The next breakout companies will build dynamic software. Products that shape itself to you over time.
Most software today forces you to adapt to it.
Take it or leave it.
That’s about to change
The next wave of products will adapt to you.
In five years, using static, one-size-fits-all software will feel as outdated as dial-up internet.
The next breakout companies will build dynamic software. Products that learns, and shapes itself to you over use.
A writing tool that remembers how your words and thoughts flow.
Meeting transcribers that add context from past conversations, so you never lose the thread.
Tutors that coach students across years, not just semesters.
This isn’t just “better UX.”
This is a fundamental shift in how software is built.
My hunch is that personalization isn’t as effective based on past data or demographic patterns. Instead it’s honed in real-time from a feedback loop that watches, learns, adapts as you use the tool.
When a product understands you—the cost and friction of switching, skyrocket.
The best software won’t just be useful.
It’ll be impossible to leave.
The Signal
Early findings from a World Bank study of students using GPT-4 as a tutor in Nigeria found that six weeks of after-school AI tutoring is equivalent to 2 years of typical learning gains.
The Latest
ChatGPT Tasks offers job scheduling, reminders and more: Currently in beta, Tasks lets ChatGPT Plus, Team and Pro users schedule actions ahead of time. For example, if someone wants to receive project reminders or daily weather updates, they can prompt ChatGPT, which will notify them of the chosen date and time.
Google’s Gemini AI can now simultaneously process multiple visual streams in real time.This breakthrough – which allows Gemini to not only watch live video feeds but also to analyze static images simultaneously – wasn’t unveiled through Google’s flagship platforms. Instead, it emerged from an experimental application called “AnyChat.”
What’s next for agentic AI? LangChain founder looks to ambient agents: Ambient agents are AI systems that run in the background, continuously monitoring event streams and then triggered to act when appropriate, according to pre-set instructions and user intent. To help prove out and advance the concept of ambient agents, LangChain has developed a series of initial use cases, one that monitors emails, the other for social media, to help users manage and respond when needed.
Luma AI releases Ray2 generative video model with ‘fast, natural’ motion and better physics: San Francisco-based Luma released Ray2, its newest video AI generation model, available through its Dream Machine website and mobile apps for paying subscribers (to start).
Ndea wants to build AI that keeps improving on its own with ‘no bottlenecks in sight’: François Chollet, a former Google engineer and the creator of the widely-used Python deep learning framework Keras, has co-founded Ndea, a new AI research and science lab, alongside Mike Knoop, co-founder of Zapier. In a post on the startup’s new website, the founders explain their goals of combining intuitive pattern recognition, enabled by deep learning, with formal reasoning through what they call “guided program synthesis.” They say that this fusion will allow AI systems to adapt and innovate far beyond current task-specific applications, ultimately leading to artificial general intelligence (AGI).
Researchers open source Sky-T1, a ‘reasoning’ AI model that can be trained for less than $450:NovaSky, a team of researchers based out of UC Berkeley’s Sky Computing Lab, released Sky-T1-32B-Preview, a reasoning model that’s competitive with an earlier version of OpenAI’s o1 on a number of key benchmarks. Sky-T1 appears to be the first truly open source reasoning model in the sense that it can be replicated from scratch; the team released the dataset they used to train it as well as the necessary training code. According to the NovaSky team, Sky-T1 performs better than an early preview version of o1 on MATH500, a collection of “competition-level” math challenges. The model also beats the preview of o1 on a set of difficult problems from LiveCodeBench, a coding evaluation.