@edwin 🌵
@edwin
Followers
2K
Following
512
Statuses
209
urban chicken farmer; data + model wrangler
SF via Penang
Joined September 2006
@alexisgallagher I think even with every neuron mapped, Chalmers’s “hard problem” of consciousness remains… how can we simulate inner experience, since it is, “inner”?
0
0
5
Please help find my kid’s stolen bike. The Christmas present (2022 Commencal Meta HT AM Ride, light brown) that my kid spent weeks assembling was stolen from our SF garage overnight. @stolenbikessfo @BikeIndex
6
4
10
Given the current OpenAI situation, GPTs’ longevity seems uncertain. But if they stick around, here are 2 tactics to ensure their accuracy and performance⚡: - Use APIs via Actions to retrieve actual data (like Spotify IDs) 🎵 - Use a proxy that simplifies API responses since APIs are mostly built for rendering, not the tokens/second performance requirement of LLMs 🚀. I built a Spotify-powered Music Recommendation GPT ( using these tactics. See more on
0
1
7
@dr_cintas @Scobleizer Nice work, Alvaro! I just did an evaluation of GPT3.5 vs Llama 2 for 50 technical challenges. Curious to get your take on it.
Which model has better taste in software - Llama v2 or GPT3.5? We have been testing this @PromptJoy and find them to be almost on par when it comes to choosing OSS - an essential skill for agents that build 😬. Methodology + results below:
0
0
0
@rajko_rad @simonw @emollick Some caveats: 1. Llama v2 on @replicatehq is $$$ right now 2. The average query time for our test is > 20sec! So not quite prod ready 3. It's harder to coax Llama's assitant to listen to us when it comes to exact schema generation - presumably this can be fixed by fine-tuning
0
0
1