robertclaus

14 ms·

211.
by robertclaus 3y ago
I think this paper is cool and I love that they ran these experiments to validate these ideas. However, I'm having trouble reconciling the novelty of the ideas themselves. Isn't this result expected given that LLM's naturally
212.
by robertclaus 3y ago
I love the write-up. Very entertaining. "it's the perfect blend of simplicity and security for modern communication" "experience encryption like never before!"
213.
by robertclaus 3y ago
I'm glad to see another project tackling the challenge of speeding up image generation. After seeing the bang-for-buck jump from sdxl to sdxl-turbo, I assume there's tons more room for improvement on the speed side of things.
214.
by robertclaus 3y ago
It was refreshing to see an off-the-shelf project box on the parts list rather than 3d printing it. I'm in the middle of a multi-day printing spree for a project and am learning to appreciate that not every generic box needs to be 3d p
215.
by robertclaus 3y ago
"Setting up LLMs to output structured data is incredibly hard." resonated strongly with my experience working in similar one-off projects. I've almost always implemented some level of fuzzy-matching to validate and convert th
216.
by robertclaus 3y ago
In my experience a lot of developers ignore the concept of dev/prod parity like this because a different team (ex. Deployment) ends up dealing with the integration issues it causes. That being said, its also traditionally a balance bet
217.
Pizza Taste Test(robertclaus.com)
3 points by robertclaus 3y ago|0 comments
218.
by robertclaus 3y ago
This was my feeling too. Considering a bunch of new images models are coming out at once claiming they can all spell now implies to me it was likely just a training set issue - the caption generators just needed to be told to include any te
219.
by robertclaus 3y ago
I find this perspective very in tune with my own experiences. Tests don't mean your code behaves correctly from an outside perspective - just that it behaves like you meant it to. I've caught plenty of issues in code review that h
220.
by robertclaus 3y ago
I've been at software companies where similar usability issues commonly came up because they weren't treated as a priority. Based on my experience, this feels like a lack of user-oriented testing/qa more than a lack of hardwa
221.
by robertclaus 3y ago
This is a really good point. This should be a reminder at the top of every stand-up notes document. :) To clarify why this worked for us: our daily inspect-adapt cycle focused on things that would affect our sprint plan, so if everything wa
222.
by robertclaus 3y ago
Thank you!
223.
by robertclaus 3y ago
At a previous company we successfully implemented most of the ideas in the article (and stayed remote) while encouraging as much synchronous work as we could. I personally find that there are a lot of benefits to synchronous work for junior