Why is my Celcius to Kelvin Model performing so bad compared to my Celcius to Fahrenheit Model

I am new to tensorflow and I saw an example model which was made as following –

l0 = tf.keras.layers.Dense(units=1, input_shape=[1])
model = tf.keras.Sequential([l0])
model.compile(loss = 'mean_squared_error', optimizer = tf.keras.optimizers.Adam(0.1))
history = model.fit(sampl, sampl2, epochs=500, verbose=False)

Now when trained on significantly less data –

celsius_q    = np.array([-40, -10,  0,  8, 15, 22,  38],  dtype=float)
fahrenheit_a = np.array([-40,  14, 32, 46, 59, 72, 100],  dtype=float)

It gave a very close prediction for this data but when I used the following data which is significantly more for celsius to kelvin I get predictions which are very off

sampl = np.random.uniform(low=-50, high=50, size=(200,))
sampl2 = sampl+273

I get the prediction 267 Kelvin for 22 Celsius.

I increased my epochs to 1000 and now the prediction is bang on

The codes are the same for both so why do I see such disparity?