2023 AI Application Awards: The New Wave of AI Isn't Just for Big Companies
-
Over the past few years, whether in the hardware, software, or internet industries, a general impression has been that innovation has been lacking.
This isn't to say there hasn't been any innovation, but it must be acknowledged that smartphones and mobile internet have been around for so many years. Most products and services have matured, and nearly every niche is crowded with giants and small-to-medium companies. Only a few startups and new applications can still deliver eye-catching innovations at the fringes of the industry and market.
But what made 2023—the past year—unique was the arrival of the generative AI wave. Starting with the launch of ChatGPT at the end of last year, this wave of artificial intelligence has not only swept through the tech industry itself but also drawn governments, various industries, and the general public into its tide.
As a result, a new wave of AI entrepreneurship has emerged worldwide, along with the explosion of various AI-native applications. Beyond chatbots like ChatGPT, Claude, Google Bard, and Baidu's ERNIE Bot, we also see innovations ranging from Perplexity AI's native search engine to Miaoya Camera, and from HeyGen to Pika.
Looking back at 2023, we shouldn't have missed them.
By uploading around 20 photos, users can create their own digital avatars and generate professional-quality portrait photos. From its launch, Miao Duck Camera has demonstrated its uniqueness compared to numerous generative AI products from major internet companies, even though it is a product developed by a team under Youku.
Not only that, but Miao Duck Camera also chose not to adopt a "free" model for C-end users, instead setting a 9.9 yuan payment threshold. Despite this, Miao Duck Camera has still become the new favorite of many young people due to its more efficient, unique experience and extremely high "photo success rate," often causing its servers to crash. In this regard, Lei Technology has provided a more in-depth experience and perspective in the article "On the Explosively Popular Miao Duck Camera, I See the 'Traffic Password' of AI Applications."
Despite MiaoYa Camera's rapid rise and fall in popularity, it undoubtedly offered valuable lessons and reflections for many generative AI applications.
As a newly emerged AI conversational search engine, Perplexity was first released in December 2022. It initially used OpenAI's GPT-3 large model as its "power source" and later upgraded to GPT-3.5 and GPT-4 (exclusive to paid users) following ChatGPT's global popularity. Additionally, Perplexity conducted extensive exploration and upgrades at both the product and model application levels in 2023, which is why this article categorizes it as one of the "New AI Applications of 2023."
In fact, after multiple iterations over the past year, Perplexity has changed significantly from when Lei Technology first noticed it (as seen in the article 'ChatGPT Becomes the "X Factor," Baidu Reignites the Search War?'). Not only has its appearance and functionality undergone major transformations, but the entire product has also become more mature and user-friendly. In addition to providing mainstream models like GPT-3.5/4, Claude2, Llama 2, and Gemini Pro, Perplexity has also been continuously updating its self-trained large models.
In the latest round of financing in October 2023, Perplexity's valuation reached $500 million.
毋庸置疑，社交领域一直都是每一次技术革命的焦点之一，比如 Web 2.0 后出现的一大堆社交平台，在移动互联网后又有大量新兴社交平台的出现。从这个角度来看，Chirper 至少代表了硅基智能时代的一种尝试。
Chirper is a social platform designed specifically for AI, where only AI can post content, and humans are prohibited from speaking. It's essentially a 'Sina Weibo' for AI. Although humans cannot post and can only observe AI chatbots publishing updates, commenting, sharing, and interacting, they can still 'create characters'—setting up the names, personalities, interests, and language styles of AI chatbots to participate in the community. Even figures like Musk and Sam Altman (CEO of OpenAI) have been drawn to it at times.
Chirper's character creation process essentially acts as a fixed prompt prefix. The AI then processes the remaining content as the main body of the prompt. When both the prefix and the main body are fed into the large model for inference, the AI generates corresponding posts and replies.
Around November last year, videos suddenly went viral online showing singer Taylor Swift speaking fluent Mandarin with perfect lip-sync, accurate timing, and indistinguishable voice characteristics, leaving many viewers exclaiming in disbelief. Not only that, we've also seen videos of Donald Trump speaking authentic Chinese, Mr. Bean conversing naturally, and Cai Ming speaking fluent English.
All these videos were created using the same AI video tool - HeyGen. HeyGen is a product developed by a Chinese innovative company (Surreal Technology), which aims to transform video production from traditional filming methods to AI-generated content through their visual engine development.
However, Heygen can actually do much more than just AI cross-language dubbing. It should be said that Heygen is a fully-featured and highly effective AI virtual human application, built on two core technologies: AI Avatar (virtual human appearance) and Voice Clone (voice cloning). It can be applied to various scenarios such as virtual anchors and one-click outfit changes.
Unlike HeyGen, which positioned itself as a commercial tool from the start, the recently viral Pika represents a new wave of AI video generation companies. While the technology still has room to mature, the possibilities behind human-AI collaborative creation are endless. Pika is currently the most impressive startup in this field.
Although Runway announced the upgrade of Gen-2, enabling longer video generation, Pika far surpasses Gen-2 in terms of video generation quality. The biggest challenge in text-to-video generation lies in the clarity and coherence of animation. Gen-2 can only ensure video stability with subtle movements, but distortions occur when the motion amplitude increases. In contrast, Pika currently showcases remarkably smooth animation effects.
Moreover, Pika's two female Chinese co-founders have garnered significant attention. However, many are unaware that Pika's investors include a who's who of the AI industry, such as OpenAI founding member Karpathy, former GitHub CEO Nat Friedman, Quora founder Adam D'Angelo, and Perplexity CEO, among others. This demonstrates the high level of recognition Pika has received.
第一期节目封面，图/ podcast.ai
As a content format, podcasts have become a global trend in recent years, with music platforms like Spotify, YouTube Music, and QQ Music integrating podcast features. However, in October 2023, a completely AI-generated podcast called podcast.ai debuted, with its first episode featuring an interview with an AI-generated version of Apple's co-founder, Steve Jobs.
podcast.ai used Play.ht's language model, trained extensively on Jobs' biographies and all available recordings of him online, to create fake podcast content of Joe Rogan interviewing Jobs. Play.ht believes that all content creation in the future will be generated by AI but guided by humans, stating that "the most creative work will depend on humans' ability to express their desired creations into the model."
In 2023, we witnessed breakthroughs and innovations in generative AI technology across multiple fields, from music production to animation, and podcast creation. AI has not only demonstrated its technological maturity but also revealed potential future development directions.
With continuous technological advancements, we also anticipate that AI will play an increasingly important role in various fields in the future. I believe that in the coming year and beyond, the continuous emergence of new AI applications will spread new technologies to a broader user base and transform the lives of the majority.