@omarsar0 Interesting approach. How do they make sure the harness is not "overfitting" on the fixed evals by hard coding edge cases? This sounds like benchmaxxing on steroids at the harness level, which may or may not translate to better real-world performance.
@adamdotdev I find myself asking more and more often what level of abstraction I should be operating at and it's a moving goalpost with greater trust in the ability of the agents. Still, I think being able to @ mention a file is often faster than the agent searching.
@AltBryce_@_AthenaVC @RobLiuLiu @rishirajkabra@josephlouistan Very cool, thanks for sharing! So it's spotify for podcasts? What problems are you facing with the current solutions? As a consumer of podcasts on Youtube and Spotify, I can't think of anything other than the annoying ads.
Biggest challenge: Building habit of carving time for personal projects
Next week:
- Read Founders at Work, Third Door
- Finish the first task
- Share the decks of at least 3 companies on the Thiel/Altman/Gates list here
- 12 productive hours on personal projects
@_AthenaVC
Last week @_AthenaVC
- Finished Rob’s course and Sam Altman’s playbook
- Halfway through Task 1
- Started using Toggl
Learned:
- Focused execution needs practice - I wanted to get the first task done but failed
- Gates seems more prone to "sunk cost" trap than Thiel and Altman
https://t.co/c4GQrgCIuI quite literally an FDM 3D printer's wet dream, but imagine if this was spinning something like spider silk... strong, naturally biodegradable
Just discovered @GIMS - French rap flows so much better because of the liaisons, mais il est quand meme quelque chose de dinge pour melanger le vent du nord (chanson celtic) avec du rap😎