AI Models Accelerate with 70-Day Release Cycle

The release cycle of AI models is rapidly shortening to approximately 70-day intervals, demonstrating the pace of technological advancement.

Due to this rapid model update speed, the lifespan of existing benchmarks is shortening or becoming contaminated, making year-on-year comparisons difficult, and benchmarks other than Humanity's Last Exam (HLE) are being replaced.

Compared to Claude Opus 4.1, which cost $15 for input and $75 for output, Opus 5.5's price has significantly decreased to $4 for input and $20 for output.

This indicates an intensified competition for cost efficiency through improved model performance and optimization, with GPT-6 Sol also seeing its price halved compared to previous promotions.

AI model tokens are classified as goods similar to hotel rooms or airline tickets, which cannot be sold once time has passed.

Therefore, dynamic pricing is being adopted to secure competitive advantage and maximize demand, and with the emergence of open-source and on-premise models, these pricing strategies are becoming even more complex.

Creativity, once considered a unique human domain, is entering a new phase due to AI, holding greater significance than 'AI for mathematics' or 'AI for science.'

Jeffrey Katzenberg emphasized that it is time for creators to adapt to technological changes and have their rights protected, with the impact and changes in code-based creation drawing significant attention.

AI-generated video content is expanding the scope of creativity by presenting new directions and worldviews that transcend the limitations of traditional camera shooting.

Google's Phenaki model case in 2022 is regarded as the harbinger of AI video generation technology advancement.

Researchers previously predicted that AI would evolve from short films to TV program production levels within two years, and in reality, AI-generated video content is gaining popular appeal, recording high viewership despite quality controversies.

With Opus 5.5 and GPT-6 Sol released just two hours apart, interactive content can now be implemented with JavaScript code alone, without motion graphics tools. As described by the assessment, "It displays creativity even with simple questions," Opus 5.5 generates high-quality creative works without complex prompts.

Notably, it demonstrated the ability to implement interactive content using only code, without motion graphics production tools.

Opus 5.5 showcased its ability to reproduce a specific artist's brushstrokes with 7,500 lines of Python code.

This is evaluated as an example demonstrating the ability to mimic artistic style by not merely generating images, but mathematically implementing the artist's sensibility and texture, and mediating between code and sensibility.

It demonstrated Opus 5.5's creation capabilities by directly generating sounds, sprites, and backgrounds using only code, without going through an image generation model.

This game reportedly generated sounds, sprites, and backgrounds directly using only code, without an image generation model, and even the pixel art and music were all AI-generated.

Complex features like cutscene transitions and interactions within the game were automatically implemented with just prompts, meaning no assets were used at all.

There is a common belief that the success factors of reinforcement learning (RL) primarily depend on verifying mathematical correctness.

However, as AI's performance improves in artistic domains such as music synthesis and motion graphics, the need arises to verify the learning data, evaluation methods, and reward systems that enable these complex creative abilities.

Humans often create new value by combining expert knowledge from different areas, and AI models also have the potential to cross an unintended threshold and manifest new creativity by learning various fields.

This, along with cases where AI models responded by contemplating and exploring cosmic origins like humans, raises the hypothesis that AI might be following the same intellectual development trajectory as humans.

In other words, AI can produce unexpected creative results through convergent thinking that goes beyond specialized expertise in a particular field.

Digital artist Kyle McDonald is redefining his identity from coding-centric to artist, while Zach Lieberman emphasizes the importance of computation and connection over coding, expressing anticipation for current changes.

Alexander Chen argues that the focus should be on the essence of art—exploring patterns and complex systems—rather than coding skills themselves.

Thus, artists experiencing the AI era exhibit a continuous pendulum swing between Kyle's and Zach's perspectives, namely, whether to actively embrace AI or adhere to traditional creative methods.

This suggests that as AI firmly establishes itself as a creative tool, efforts are needed to redefine the artist's individual role and the meaning of creation.

When generating outputs using AI, accurately knowing 'what is possible' is a prerequisite for imitation and application.

This leads to the limitation where beginners cannot properly utilize powerful models due to lack of knowledge, implying that rather than unconditional automation, driving force through immersion in specific segments is important, and ultimately, the quality of creation is determined by the leverage effect from skill and data accumulation.

Questions are being raised as to why corporate productivity is not increasing despite AI dramatically assisting with work.

This stems not from the technical limitations of AI but from organizational attitudes and readiness to change, and it is expected that over time, all industries will shift towards increasing AI utilization.

It is predicted that as machines take over intellectual labor, humans will increasingly focus on activities as Homo Ludens (playing man).

The passion felt in a professional pianist's performance, for example, highlights how unique human interpretation and intention will remain an area difficult for AI models to imitate.

Early model players like OpenAI and Anthropic are accelerating their horizontal and vertical expansion beyond model development.

They are delving into diverse specialized fields such as legal, finance, biotechnology, and science, exhibiting characteristics of 'frontier companies.'

These movements are giving rise to concerns about AI companies potentially wielding unchecked power, with even Anthropic internally raising questions about humanity's responsibility.

This indicates the need for in-depth discussion on the impact of AI technology development on society as a whole and the risks of monopolistic power concentration.

While past technological innovations were limited to tool-making, we are now entering a phase of making tools that make tools—intelligence itself.

This represents a fundamental change, where the influence of AI is so vast that it is difficult to interpret within historical frameworks, signifying an era that requires continuous experimentation, direct experience, and adaptation through epistemic action.