SourceCodeplz (@SourceCodeplz) header
SourceCodeplz (@SourceCodeplz) profile picture

SourceCodeplz

@SourceCodeplz

Full-stack engineer. What does it cost to live in the EU? https://teapaw.org/t.co/pgvIA34S2N Looking for European Podcasts Charts? https://teapaw.org/t.co/Qg2gkBLUwV

 Secured Site Institute
securedsite.org
Joined October 2022
  • Tweets3.1K
  • Following483
  • Followers337
132Media
Photo by SourceCodeplz (@SourceCodeplz)Photo by SourceCodeplz (@SourceCodeplz)Photo by SourceCodeplz (@SourceCodeplz)Photo by SourceCodeplz (@SourceCodeplz)Photo by SourceCodeplz (@SourceCodeplz)Photo by SourceCodeplz (@SourceCodeplz)Photo by SourceCodeplz (@SourceCodeplz)Photo by SourceCodeplz (@SourceCodeplz)Photo by SourceCodeplz (@SourceCodeplz)Photo by SourceCodeplz (@SourceCodeplz)
# AGENTS.md - Remove obsolete paths. Ship only current code. - Use simplest code that meets needs now. - Build in layers. Ship smallest working version first. Add on top of working product. - Keep modules separate with clear concerns. - Prefer mature libraries that simplify or stabilize. - Check existing deps, docs, and types first. - Design for the long term. - Study proven products. Adopt their patterns.
2
1
28
6,332
yes ok
1
Replying to @ArtificialAnlys
why does this feel so scripted?
247
Replying to @GoogleDeepMind
nice name!!!
5
Replying to @atomic_chat_hq @OpenBMB
Now this is useful
1
1
115
Replying to @leo114119
Have you even tried Terra? That model is a beast! So token efficient is ridiculous. Almost like it doesn't even think
21
Replying to @0xWhiteMage
yes , this at 300 freaking million tokens
1
1
14
Replying to @brandon_galang
Openai even more than anthropic Their models are so token efficient
1
1
55
Replying to @RamaswmySridhar
Open models were focused on coding not general intelligence. Also much smaller Nr of params Billions Vs Trillions
66
Replying to @Frtnel @daveeyyy_
Their loss Terra awesome
13
Replying to @daveeyyy_
Tera is the most efficient intel/tokens model ever imo
1
88
Replying to @thdxr
thank you for this. the push I needed for my agent. Though in the past you said we shouldn't even try coz you guys are so ahead?
1,615
Replying to @NartOutHere
It's raining in Bucharest since last 3 days.
146
Replying to @brandonkruger @DogukanUrker
Don't expect miracles. It's good at tool calling agentic stuff Not very smart
2
13
Replying to @Kamal_AnanD1
due to long horizon not being amounted teapaw.org/SourceCodeplz/status/2…
SourceCodeplz (@SourceCodeplz) profile pictureSourceCodeplz@SourceCodeplz20h
a grounded take on the AA (v. 5) index
Image from SourceCodeplz (@SourceCodeplz)
1
69
Replying to @gadi_neelesh
openai models are so token efficient omg teapaw.org/SourceCodeplz/status/2…
SourceCodeplz (@SourceCodeplz) profile pictureSourceCodeplz@SourceCodeplz20h
a grounded take on the AA (v. 5) index
Image from SourceCodeplz (@SourceCodeplz)
1
47
Replying to @jan_volad
seems like they didnt really track long horizon tasks teapaw.org/SourceCodeplz/status/2…
SourceCodeplz (@SourceCodeplz) profile pictureSourceCodeplz@SourceCodeplz20h
a grounded take on the AA (v. 5) index
Image from SourceCodeplz (@SourceCodeplz)
2
160
Replying to @osanseviero
just dont mess with the writing style and prose. dont care too much about coding as there are other models better trained for that. gemma is my GOAT at writing
42
1,302
a grounded take on the AA (v. 5) index
Image from SourceCodeplz (@SourceCodeplz)
2
3
470
Replying to @NeoAIForecast
teapaw.org/SourceCodeplz/status/2…
SourceCodeplz (@SourceCodeplz) profile pictureSourceCodeplz@SourceCodeplz21h
Replying to @SourceCodeplz @ArtificialAnlys
small (700b) open-weights models are great at coding and tool calling but lack world knowledge. because the parameter difference is so huge T to B, of course they also are not as intelligent. this is i think the difference in this new analysis.
8
Replying to @zkstx
opencode 1.x
18
Load more