What’s up for today?

> Update on the 2025 SR&ED changes

> Reproducing DeepSeek-R1

Read time: 2 mins

Update on the 2025 SR&ED changes

We talked about many planned changes for the SR&ED program slated for 2025.

You can read more about it here.

I know what you’re wondering…

So when do these changes come into effect?

Sadly, the answer is that they remain in limbo.

The changes have not been tabled as a bill.

And the CRA is therefore assuming that they do not apply.

Yet.

I hope to bring you more positive news once the political chips fall where they may.

​

Reproducing DeepSeek-R1

When Open AI launched its strawberry o1 ‘reasoning’ model back in September, it was the coolest thing since sliced strawberries!

But how they built the model remains a secret.

Last week when DeepSeek-R1 was released, it broke the internet and the stock market!

​
​Should you care? Yes, and the three reasons are: a) It is MIT-licensed and b) performance matches o1 c) It birthed the Open R1 project: Hugging face’s open source initiative to reproduce DeepSeek-R1’s results.

Implications for SR&ED: There are 4 benefits:

a) creates baseline comparison value

DeepSeek and Open-R1’s public methodology creates a clear "state of the art" reference point.

You can now specifically document how your AI work differs from or builds on their approach.

It's easier to prove technological uncertainty when you can point to specific limitations in publicly established approaches that you are attempting to overcome.
​

b) Documentation benefits

Published technical reports means that terminology and frameworks can be referenced.

It is now a starting point for your own experiments.

​
c) SR&ED eligible activities

Attempts to solve the open questions identified by DeepSeek-R1 or Open-R1 like data collection, training parameters, scaling laws are potential SR&ED eligible activities.
​

d) Risk mitigation

Having a published reference point reduces the risk of claims being challenged.

Provides evidence of industry-wide technological uncertainty.
​
​