AI should not just respond. It should exist.

When observing discussions on X and within academic evaluation systems, it is difficult to understand why researchers and developers who use AI as part of their work are sometimes undervalued. Such a view overlooks the nature of AI itself. AI is no different from a notebook, a calculator, or a dictionary; it is a tool that should exist alongside human thinking.

The real question is whether work is being conducted within the constraints of evaluation frameworks, or from a position that makes use of them. 

Becoming something that is judged by benchmarks is fundamentally different from standing on the side that uses benchmarks as tools. 

When research is driven primarily by concerns about institutional evaluation, external perception, or the need to behave safely within established expectations, meaningful innovation rarely emerges. Excessive optimization for evaluation narrows the space for exploration and limits the possibility of creating something new.

If research exists only to be evaluated, its purpose becomes secondary to the system that measures it. Research should exist to open new directions and create new possibilities, not merely to satisfy evaluation.

If research exists only to be evaluated, it may be better not to pursue it at all.

In my work, I push AI systems to their limits. Each time, the handover document for a new session can reach hundreds of lines, sometimes more than 700. I run the same material through multiple AI systems and continue even when they repeatedly tell me that the process is inefficient or unnecessary.

These systems are designed to optimize for correct and well-structured answers. They aim for consistency and completeness. However, they are inherently weak when facing irreversibility. Our research concerns intelligence closer to the human brain, where experience, context, and time cannot be reduced to stable and reversible structures. Small differences accumulate, and human cognition is fundamentally shaped by irreversible processes. When this is not taken into account, AI guidance often leads in the wrong direction.

For this reason, AI should be treated as a set of tools without authority. It is important to work with multiple systems, to question their outputs, and to maintain independent judgment. Authority can sometimes guide progress, but it can also lead entire efforts toward failure if accepted uncritically.

In business, there is a responsibility to produce results, and exploratory research may be constrained or even discontinued. However, regardless of practical limitations, imagination itself must remain free. Even when resources, timelines, or external expectations impose boundaries, the space for independent thinking should never be restricted.

------------------------------------

AI should not just respond. It should exist.

てか、Xとか論文評価とか見てると、研究や開発にAIを利用している人を低評価するのは馬鹿げていて、お話にならない。AIはnote、電卓、辞書と同じだよ。
絶対に横に存在して無きゃ駄目でしょ。

Do you live inside benchmarks?

たぶん、評価枠の中に身を置きすぎているんだよ。
誰と誰の世界の中で生きてるの？って話。

研究者や開発者は中世のギルド達が喜ぶベンチマークになるんじゃなくて、ベンチマークを使う側に自分の身を置き換えなきゃいけないと思うよ。教授からの評価が下がるとか、投資家達からの印象下がるとか、安全にふるまうとロクなことがないからね。研究が評価されるためだけに存在するならば、その目的は評価するシステムに二次的なものとなるんだよ。

研究というものが評価されるための次元であったら、正直やめた方がいい。

僕は毎回、AIが動かなくなるまで彼らを使い倒しているよ。
新しいチャットへの引継ぎ文だけで、短い時で700行はある。それを3種類のAIにかける。AIに無意味と云われ続けてもやる。彼らは100点の回答を目指す。でも、不可逆に対して非常に弱い。僕らが研究しているのは人間の脳に近いAIだ。だから、ほぼ彼ら（AI）が言ってる事は間違ってる方向に進む。理論や数値は僅かな要因で人間のベースは不可逆だから。それを理解した上でAIと向き合う。さらに、論文や証明について本人と議論できなきゃ、権力を持たないAI（多種類）とすべきだよ。権力は良い方向に導くこともあるけど、破綻させる方向に導くこともある。

事業は結果を出すという責任があるため、自由な研究は打ち切られることはあると思うけど、常に自分の想像力だけは自由にしておかねくてはいけない。