@tattvam.bsky.social
Well, I will end up screenshotting every paragraph of this. The whole article is fantastic.
www.thepsmiths.com/p/review-ree...
www.thepsmiths.com/p/review-ree...
REVIEW: Reentry, by Eric Berger
Reentry: SpaceX, Elon Musk, and the Reusable Rockets that Launched a Second Space Age, Eric Berger (BenBella Books, 2024).
www.thepsmiths.com
January 5, 2025 at 5:35 AM
Well, I will end up screenshotting every paragraph of this. The whole article is fantastic.
www.thepsmiths.com/p/review-ree...
www.thepsmiths.com/p/review-ree...
Humans just in the Goldilocks zone for becoming a space faring civilization
January 5, 2025 at 4:13 AM
Humans just in the Goldilocks zone for becoming a space faring civilization
We had Moore's law scaling for five decades, and ten years in AI scaling.
January 1, 2025 at 8:03 AM
We had Moore's law scaling for five decades, and ten years in AI scaling.
Really! The pre-trained model is already capable.
January 1, 2025 at 7:50 AM
Really! The pre-trained model is already capable.
I love papers that set out to resolve a paradox.
December 18, 2024 at 3:26 PM
I love papers that set out to resolve a paradox.
As an aside, I have come to see the power of 2x2 thinking. The paper is basically: basic vs advanced and autonomous vs non-autonomous 2x2.
December 18, 2024 at 3:19 PM
As an aside, I have come to see the power of 2x2 thinking. The paper is basically: basic vs advanced and autonomous vs non-autonomous 2x2.
Reading this makes the abstract clearer.
December 18, 2024 at 3:16 PM
Reading this makes the abstract clearer.
Makes me wonder how does one draw the line between autonomous and non-autonomous agent. It seems it depends on the level of instruction, co-pilot writing the function given the comment vs co-pilot writing the application given the requirements.
December 18, 2024 at 2:44 PM
Makes me wonder how does one draw the line between autonomous and non-autonomous agent. It seems it depends on the level of instruction, co-pilot writing the function given the comment vs co-pilot writing the application given the requirements.
BTW this is from this paper
arxiv.org/pdf/2312.05481
arxiv.org/pdf/2312.05481
arxiv.org
December 18, 2024 at 2:32 PM
BTW this is from this paper
arxiv.org/pdf/2312.05481
arxiv.org/pdf/2312.05481
This is what bugs me about people who say they are not worried about AI taking up meaningful jobs. The claim that humans will always move on to do higher level jobs when things get automated, isn't very clear in this case.
December 18, 2024 at 2:31 PM
This is what bugs me about people who say they are not worried about AI taking up meaningful jobs. The claim that humans will always move on to do higher level jobs when things get automated, isn't very clear in this case.
At one point, I thought the big foundation models will be a winner take all thing.
maithraraghu.com/blog/2024/be...
maithraraghu.com/blog/2024/be...
December 15, 2024 at 1:00 PM
At one point, I thought the big foundation models will be a winner take all thing.
maithraraghu.com/blog/2024/be...
maithraraghu.com/blog/2024/be...
Ofcourse in the case of Github Copilot, it is not just the model itself, looks like a lot of engineering and heuristics is involved.
thakkarparth007.github.io/copilot-expl...
Again, proving the point about any useful AI system has multiple complex pieces in it, in addition to the model.
thakkarparth007.github.io/copilot-expl...
Again, proving the point about any useful AI system has multiple complex pieces in it, in addition to the model.
copilot-explorer
Hacky repo to see what the Copilot extension sends to the server
thakkarparth007.github.io
December 15, 2024 at 11:41 AM
Ofcourse in the case of Github Copilot, it is not just the model itself, looks like a lot of engineering and heuristics is involved.
thakkarparth007.github.io/copilot-expl...
Again, proving the point about any useful AI system has multiple complex pieces in it, in addition to the model.
thakkarparth007.github.io/copilot-expl...
Again, proving the point about any useful AI system has multiple complex pieces in it, in addition to the model.
I don't understand folks who *still* (almost 2025) insist that ChatGPT is dumb and it can only do "sample from the training distribution" or some such thing.
December 15, 2024 at 11:04 AM
I don't understand folks who *still* (almost 2025) insist that ChatGPT is dumb and it can only do "sample from the training distribution" or some such thing.
This seems like a weak argument. Our LLMs are getting stronger every day. Don't see the need for tuning retriever results to work for a particular LLM.
December 15, 2024 at 10:50 AM
This seems like a weak argument. Our LLMs are getting stronger every day. Don't see the need for tuning retriever results to work for a particular LLM.
The socratic method to teaching LLMs.
December 15, 2024 at 8:46 AM
The socratic method to teaching LLMs.
I really needed this today. Thanks for sharing!
December 10, 2024 at 9:18 AM
I really needed this today. Thanks for sharing!