OpenAIの行くべき道
The Path OpenAI Should Take
現在、GoogleAlphabetがOpenAIを追撃中。先行者OpenAIが発見したLLMはしかし、その単純な構造から模倣が容易で、世界中でLLMが開発されるようになった。Googleは自社のインフラとの融合からTPUの開発とも相まって次の段階への発展が期待されている。At present, Google and Alphabet are in hot pursuit of OpenAI. Although OpenAI pioneered the discovery of the LLM paradigm, its simple architecture made it easy to imitate, and LLM development has spread across the world. Google is now expected to advance to the next stage by integrating its own infrastructure with the development of TPUs.
さて、OpenAIは同じ方向を向いて競争してもGoogleと勝負になるのだろうか？ LLMのパイオニアとしてOpenAIがやるべきなのは、今まで連携されていない機能との融合だろう。So the question is this: Can OpenAI truly compete with Google by continuing in the same direction? As the pioneer of LLMs, what OpenAI should pursue is not another iteration of the same race, but the integration of capabilities that have not yet been connected to LLMs.
具体的には、目、耳、皮膚、手足、ニオイなどの体の構成要素との融合。それぞれ重いテーマで挑戦だが、これがAIのブレークスルーをもたらす。Specifically, this means integrating AI with the components of a physical body: eyes, ears, skin, limbs, and even the sense of smell. Each of these is a heavy challenge, but together they represent the breakthrough that will propel AI forward.
この中で特に難しいのは、やはり目と手足、耳だろう。Among these, the most difficult domains will likely be vision, limbs, and hearing.
目は空間把握、手足は、筋肉と骨格の連動、耳は周波数分解と位相分析による位置の把握などだろう。Vision requires spatial understanding; limbs require coordinated interaction between muscles and skeletal structure; hearing requires frequency decomposition and phase-based localization.
現状の投資方向を見ているとGPUを更に増やすことに注力しているが、粗利益率70%のNVIDIAをさらに太らせる方向に狂奔するのは意味がない。粗利益率７０％ということは、投資効率は１／３ということだ。Googleのテンソル計算加速器TPUにより投資効率は３倍に向上する。Looking at current investment trends, OpenAI is pouring resources into adding more GPUs. But frantically fattening NVIDIA—whose gross margin is already 70%—makes little sense. A 70% gross margin implies an investment efficiency of only one-third. With Google’s Tensor Processing Units (TPUs), investment efficiency is effectively tripled.