Diary 2025-02-28

Is GPT 4.5 stronger as an architect and Sonnet stronger in actual coding?

Well, they touched both before they were released and made the decision to include Sonnet in their product, so I guess that's what they're doing.

One interesting data point: though GPT-4.5 and Claude 3.7 Sonnet score similarly on our overall benchmark, we find that GPT-4.5 spikes more heavily on tasks involving architecture and cross-system interactions whereas Claude 3.7 Sonnet spikes more on raw coding and code editing.

As AI takes on increasingly complex tasks, we believe that multi-model agents that incorporate each model’s unique strengths will perform best.

GPT-4.5 has operated wonderfully. Our agent coding benchmarks already show significant improvement over o1 and 4o. I look forward to seeing the continued trajectory of the model on the code!

An interesting data point is that GPT-4.5 and Claude 3.7 Sonnet score about the same on the overall benchmark, but GPT-4.5 rises more significantly on tasks involving architecture and system interactions, and Claude 3.7 Sonnet rises more rapidly on raw coding and code editing showed a more rapid increase.

As AI tackles increasingly complex tasks, we believe that multi-model agents that incorporate the unique strengths of each model will perform best.

https://gyazo.com/2bfaee67b1a76b46fb368d0d75c60be1

(4.5) The following is an image of a clean illustration.

Frames are carefully arranged.

Facial expressions should be simple and clear (happy face, normal face, sad face)

Boxes should be neatly rectangular and letters inside should be centered.

Balance and equalize the placement of each frame.

Unify line thickness

If there are no problems with this, the clean-up process will begin.

Whoa, you have a good grasp of the elements of the image (I hope).

code:a.svg

<style>

.frame { fill: none; stroke: black; stroke-width: 3; rx: 15; }

.box { fill: none; stroke: black; stroke-width: 3; }

.text { font-size: 20px; font-family: Arial; text-anchor: middle; dominant-baseline: central; }

.face { stroke: black; fill: none; stroke-width: 3; }

</style>

<! -- first panel -->.

<rect class="frame" x="20" y="20" width="260" height="260"/>

<circle class="face" cx="80" cy="220" r="25"/>

<path class="face" d="M70 225 q10 -10 20 0"/>

<path class="face" d="M70 215 q5 -5 10 0"/>

<circle class="face" cx="160" cy="160" r="25"/>

<path class="face" d="M150 165 q10 10 20 0"/>

<path class="face" d="M150 155 q5 5 10 0"/>

<rect class="box" x="140" y="190" width="40" height="40"/>

<rect class="box" x="140" y="150" width="40" height="40"/>

<text class="text" x="160" y="170">A</text>

<text class="text" x="160" y="210">A</text>

<circle class="face" cx="160" cy="120" r="25"/>

<path class="face" d="M150 125 q10 10 20 0"/>

<path class="face" d="M150 115 q5 5 10 0"/>

<! -- second panel -->.

<rect class="frame" x="320" y="20" width="260" height="260"/>

<! -- (draw B in the same way, only shift the position to the right 320) -->.

<! -- Third panel -->.

<rect class="frame" x="170" y="320" width="260" height="260"/>

<! -- (draw C in the same way, only shift the position down 300) -->.

<! -- add detailed drawings of B and C if needed -->.

</svg>

https://svg-hosting.vercel.app/api/svg?url=https://scrapbox.io/api/code/nishio/日記2025-02-28/a.svg

You can't do that at all!

But I haven't caught up with my diagram output skills, but my diagram comprehension skills have improved a lot.

If I can get a tool to render the SVG, import it as an image, and compare it, I might be able to do the task properly by repeating it internally a few times until I get it right.

---