What a wonderful The first idea I had Plain text 2025. Following the desperate competition between Openai, Google, Meta, and mankind, I stomped off the clever and deeper basic models of “frontiers” and settled on a paper on what's ahead. You'll be hit by advanced reasoning, more tokens, and perhaps a press release that will ensure that their models don't constitute crazy facts.
But people are tired of hearing about how AI is transformative and have little to change into everyday existence. Getting an AI summary of Google's search results, or having Facebook ask if you want to present follow-up questions to your post, you won't be a traveler to the newcomer's future. It can start to change. In 2025, the most interesting AI obstacle race will involve innovators to make the model useful for a wider audience.
I didn't read that take from me in the first week of January. Because I felt compelled to deal with newsworthy Nexus-related topics between Tech and Trump. In the meantime, Deepseek happened. This is a Chinese AI model that matches some of the capabilities of flagship works such as Openai, which are said to be just a small part of the training costs. The giant AI lords now argue that building a bigger model than ever before is more important than ever to maintain our advantage, but Deepseek has lowered the barriers to entering the AI market. Some experts even expressed their opinion that LLM, despite being a valuable product, will become a product. If so, my paper – the most interesting race of the year is among applications that have led AI to a larger audience – has already been proven. Before you publish it!
I think the situation is quite sloppy. The billions of dollars AI leaders plan to spend on bigger models could actually trigger a leap of technology's planet, but the economics of centibarion's AI investment remain ambiguous. But in 2025, even skeptics are confident they can see the scramble to create apps that recognize that generation AI is at least as big a deal as smartphones.
Steve Jang of VCs (Perplexity AI, Particle, and -oops -Humane) who has a lot of skin in AI games agrees. Deepseek is accelerating, he says, “The world commercialization of the highly valued LLM model lab.” He offers a recent historical background. Those looking to offer use cases to real people who quickly created a nasty app on top of LLMS shortly after the first consumer transformer-based model like ChatGPT came into being in 2022. In 2023 he says that the “AI rapper” ruled. However, last year we saw the rise of countermoving. There, the startups tried to go deeper into creating amazing products. “There was this argument. 'Are you a thin wrapper around AI, or are you actually a substantial product of your own?” Jang explains. “Are you doing something really unique while using it on the core of these AI models?”
That question was answered: rappers are no longer the joy of the industry. Just as the iPhone went into overdrive when the ecosystem moved from clunky web apps to powerful native apps, the winners of the AI market will be digging deeper to take advantage of all aspects of this new technology. The products we've seen so far are barely damaging the surface of what is possible. There is no AI Uber yet. But there's the opportunity for people poised to grab it, just as it took them time to mine the possibilities of the iPhone. “If you just suspend everything, maybe five to ten years of capacity could be a new product,” says Josh Woodward, head of Google Labs, a unit that cooks AI products. In late 2023, his team created the notebook LM. Notebook LM is a support tool for writers that are more than rappers and has gained recent followers. (Too much attention focuses on the ability to turn all your notes into a gee-withgie conversation with two robot podcast hosts, but it's a stunt that unintentionally highlights the vapility of most podcasts.)
