What else can the AI "magic box" unlock in 2024?
TechFlow Selected TechFlow Selected
What else can the AI "magic box" unlock in 2024?
If viewed over a longer time horizon, today's AI large models are at best comparable to humans having just discovered fire.
Navigating Web3 tides with focused insightsBy Mu Mu
In the past year of 2023, GPT large models opened the "Pandora's box" that led ordinary people into the AI world, and triggered a global "arms race" among tech companies in developing large models.
After the emergence of text, image, and video large models, application-side innovation surged, causing severe shortages in AI chips. Meanwhile, debates over "AI replacing humans" and concerns about risks never ceased. On one hand, AI reflects humanity's intense pursuit of advanced productivity; on the other, it reveals our anxiety toward new technologies.
Looking back from 2024, AI is far from replacing humans. Technical bottlenecks have become evident: chip technology limits the pace from artificial intelligence (AI) to artificial general intelligence (AGI); data used for training large models has become “precious material,” controlled by internet giants; and commercialized AI applications have yet to benefit everyone—domestic ones are often ineffective, while overseas ones are expensive.
Viewed over a longer timeline, current large AI models are at best equivalent to early humans discovering fire. How to use it and where to apply it will define AI’s direction in 2024. We’ve also seen OpenAI, Google, and other AI tech companies begin developing their own chips—the “fire-stealers” are turning up the heat.
The New Frontier of the Internet
In 2023, AI witnessed many historic firsts—many driven by ChatGPT.
ChatGPT gave ordinary people their first glimpse of computers understanding natural language, transforming AI from a magical “secret weapon” in movies into reality. For the first time, a non-human entity—ChatGPT—was named one of the “Top Ten Scientific Figures of 2023.”
Since its launch in November 2022, OpenAI’s ChatGPT has generated attention and impact surpassing nearly all previous IT phenomena: reaching 1 million users in just 2 days, 100 million in 2 months—breaking TikTok’s previous record. Half a year later, the ChatGPT iOS app quickly topped Apple’s App Store charts upon release.
More significantly, ChatGPT broke the monopoly of tech giants over AI technology, bringing human-language-understanding products directly to the masses.
For many, ChatGPT was their first encounter with such an intelligent conversational system—capable of writing copy, solving math problems, knowing astronomy and geography, understanding stories, and catching internet memes. Though initially prone to confidently making things up or hallucinating, it learns and corrects itself based on human feedback.
The evolution speed of the underlying general-purpose large language models is astonishing—within just a year, it progressed from GPT-3.5 to GPT-4 Turbo. Just over a month ago, in early November 2023, OpenAI officially announced the GPTs initiative at its developer day, allowing ChatGPT Plus users to train custom ChatGPT bots using their own or sourced data on top of GPT-4.
OpenAI stunned the world again, and trademark filings for GPT-5 are already underway. According to information from the U.S. Patent and Trademark Office, GPT-5 will offer functions including natural language processing, text generation, comprehension, speech transcription, translation, prediction, and analysis.
OpenAI filed a trademark application for GPT-5
According to OpenAI’s official roadmap, they plan to launch a GPT Store, enabling users to publish their trained bots for others to use via paid subscriptions, forming a new business model. In less than two months, users have already created hundreds of thousands of domain-specific ChatGPT assistants through custom GPTs—an impressive reception.
In 2024, if OpenAI opens the “GPT Store,” another wave of application frenzy will sweep across the internet.
Application Explosion
OpenAI’s success was like Columbus discovering a new continent—it proved this path works. Within just a year, ChatGPT directly sparked a global AI arms race among tech companies.
Data shows that by October this year, within less than a year, Chinese enterprises and academic institutions with large models exceeding 1 billion parameters exceeded 250—excluding foreign models. AI applications surged: according to Sensor Tower, AI app downloads grew 114% year-on-year in the first half of 2023 alone, surpassing 300 million—exceeding the full-year total of 2022. Meanwhile, in-app purchase revenue soared 175% year-on-year, nearing $400 million.
In this fierce competition, tens of thousands of large models evolved powerful multimodal capabilities such as text-to-image, image-to-image, text-to-video, and image-to-video. While people were still marveling at ChatGPT’s conversational skills, models like Bard and Claude capable of understanding internet memes quickly emerged.
Moreover, some vertical domains produced “unicorns.”
In image generation, Midjourney took the lead, becoming the most powerful text-to-image tool. From launch to now, only half a year, Midjourney has already evolved to version V6, expanding from basic text-to-image to image-to-image and AI image expansion.
Midjourney’s image generation capability down to pore-level detail
Even more astonishing: behind Midjourney is a team of only 11 people, founded just two years ago. As Midjourney rose in popularity, the team expanded to 40 members and achieved $200 million in revenue this year alone—achieving financial independence early on.
Unlike most startups chasing venture capital, Midjourney hasn’t taken a single dollar from VCs. “Politely put, he doesn’t need VC involvement in his life,” said Michael Stewart, partner at M12, Microsoft’s venture fund.
The development speed of AI applications can be described as “daily updates.”
Shortly after Midjourney’s rise, Runway’s Gen-2 took over in video generation, with only a one-month gap from its predecessor Gen-1. The latest Gen-2 can generate an 18-second video from just a single prompt and skillfully apply cinematic language. Recently, Gen-2 added image-to-video functionality—simply “painting over” a region in an image brings it to life.
As AI intensifies in the video space, rising stars like Pika1.0 and Stable Video Diffusion are catching up fast, with various AI video tools competing fiercely and evolving ever quicker. Pika1.0, launched recently, already matches Gen-2 in text-to-video capability and even introduced AI image expansion to the video domain for the first time.
In 2024, these unicorns will continue unlocking the multimodal potential of large models, possibly creating new unicorns. Text, images, audio, and video—representations of human natural language—will be further refined by AI tools, which are likely to become more accessible and affordable at scale.
Chip Bottleneck
While big tech companies compete in large models and small firms innovate in applications, Nvidia—the “water seller”—has reaped enormous profits. Data shows Nvidia’s latest third-quarter revenue reached $18.1 billion, up 206% year-on-year; net profit hit $9.2 billion, surging 1,259% year-on-year. By comparison, OpenAI, the pioneer of this race, had 2023 revenue of only $1.3 billion—just $28 million the previous year.
In 2023, Nvidia firmly sat atop the AI boom. Chip prices skyrocketed, and despite high costs, GPUs remained “in short supply.” Reports suggest delivery times for Nvidia’s H100 range from 36 to 52 weeks—an unacceptably long wait for AI product development. This may partly explain why GPT-5 has yet to arrive.
Computing power, one of the three engines driving AI forward, is directly hampered by chip shortages, slowing large model evolution. Hence, OpenAI and Google have entered chip R&D to fill their own training capacity gaps.
Nvidia’s competitors are rushing to grab a slice of the pie. Intel and AMD launched their high-performance AI chips Gaudi3 and Instinct MI300X, respectively. Microsoft unveiled its AI acceleration chip Azure Maia 100, while Amazon released Trainium2, an upgraded accelerator chip for AI systems.
Meanwhile, U.S. chip sanctions have worsened China’s chip shortage crisis. To maintain market access, Nvidia can only offer downgraded, “crippled” versions of its chips.
To avoid being strangled by hardware constraints, Huawei has focused on AI chip development. Its Ascend AI chips have become domestic pillars—application partners like iFlytek say Huawei’s Ascend 910B now roughly matches Nvidia’s A100.
By the end of 2023, neither domestically nor globally has a chip emerged to rival Nvidia’s H100. Despite manufacturers operating at full capacity, chip scarcity will persist in the short term. Perhaps only when AI chip supply issues are resolved will GPT-5 and even more advanced, richer products arrive faster.
Yet from a safety perspective, this isn’t entirely negative. During this brief breathing space, humanity has the chance to make better choices about AI’s development path.
Recall that shortly after GPT-4’s release, a joint letter signed by over a thousand tech elites thrust AI safety into the spotlight, calling for a “pause in training AI systems more powerful than GPT-4.” As AI expert Gary Marcus put it: “Is it worth risking one percent chance of human extinction for the fun of chatting with machines?”
Indeed, when ChatGPT first launched, its disruptive creativity was stunning. Microsoft co-founder Bill Gates also emphasized the possibility of AI getting out of control.
Earlier this year, the internet buzzed with fears of “AI stealing jobs” and “AI replacing humans,” even spawning the meme “carbon traitor.” But a year later, public sentiment toward AI has calmed down.
We haven’t actually seen widespread job losses due to AI. At least in China, the broad integration of AI into workplaces remains a future prospect. Meanwhile, leading AI firms like OpenAI must keep pace with safety measures, and national AI safety review agencies have been established around the world.
AI’s advancement is inevitable. How we use this “fire” depends on human choice—one group builds stoves to control the flames, while others explore how those flames can light up dark frontiers.
In 2024, safety will remain one of AI’s key themes, and chips will determine whether productivity can advance further. One thing is certain: like the internet, AI will become an indispensable tool for humanity.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
@bitcoinist
