sankalp

44.7K posts

sankalp

sankalp

@dejavucoder

poasting about language models, coding agents and recent side quests

bangalore, india Katılım Ekim 2021
661 Takip Edilen25K Takipçiler
Sabitlenmiş Tweet
sankalp
sankalp@dejavucoder·
since sonnet 4.6 is here and claude code is rapidly improving day by day, i am once again reminding you that i wrote a guide to claude code and how to use coding agents better in general 2 months back
sankalp@dejavucoder

claude code is having it's cursor moment after karpathy sensei's post. never been a better time to try it. my latest blog on how to get the most out of claude code 2.0 and other agents in general is up now. grab a chai and have fun reading! sankalp.bearblog.dev/my-experience-…

English
9
13
213
45K
sankalp
sankalp@dejavucoder·
@BlackHC one can state the equations in prompt before hand to avoid such issues. comes in handy if you are making something that is outside knowledge cutoff - like something from a recent arxiv paper.
English
0
0
0
21
sankalp
sankalp@dejavucoder·
something i wonder about - both claude and codex get a hell lot of feature requests and they both are not bounded by cost so they can effectively build anything. how do they figure out what to build next? i mean both features and integrations. dont say product taste lmao.
English
4
0
11
654
sankalp
sankalp@dejavucoder·
i will probably write a short blog on my mental model of this question so far.
English
0
0
1
169
sankalp
sankalp@dejavucoder·
narrow scoped (if you ask chatgpt to reverse engineer) note that the more things you specify, the less space you leave for the model to fill with it's creativity. it's a tradeoff. e.g if you don't specify colors, model will decide colors on it's own. x.com/dejavucoder/st…
sankalp@dejavucoder

try this. it will be bit more scoped. Create a premium English-language poster about [Topic] in a Japanese-inspired modern editorial science-graphic style: off-white textured paper, deep black vertical serif typography, electric cobalt blue technical linework, acid green highlight accents, thin rule lines, boxed annotations, sparse labels, asymmetrical grid, large negative space, subtle risograph grain and halftone texture. Include a dramatic central abstract diagram explaining [Topic], small side-panel mini diagrams, and refined infographic details. Make it feel like a gallery-worthy mathematical design poster: disciplined, minimal, cerebral, intense, poetic, and visually powerful. No Japanese text, no glossy 3D, no cartooniness, no clutter.

English
1
1
13
3.3K
sankalp
sankalp@dejavucoder·
people ask how did you make this... you can take the pic and ask chatgpt! but here is how i came up with it. from prior experiments, i knew chatgpt can make great posters plus japanese prints too...
sankalp tweet mediasankalp tweet mediasankalp tweet media
sankalp@dejavucoder

chagpt images 2.0 can cook hard

English
23
25
849
76.8K
sankalp
sankalp@dejavucoder·
i am myself bitter lesson pilled and mostly use codex and claude code but there are several ppl out there who like to use oss harnesses with several models or a certain ui or self-healing/building feature. you cant let the bitter lesson suppress your delusion anon. make your idea work even in short term.
English
1
1
11
744
sankalp
sankalp@dejavucoder·
people building alternative indie harnesses will read this and be like oh and they would continue to build whatever they are building (it's a good thing). they know they to do a few things better or specific to user's demands than the lab harnesses... take pi or opencode
Ryan Lopopolo@_lopopolo

While alternative coding harnesses may have short term lift, they will be bitter lesson’d away. I am bearish on any harness that doesn’t come from the lab whose model you are using. You’re fighting against post-training. To put a finer point on this, you know how like, ioctls are like “huh that's weird but I guess whatever it's what we've got we can work with that”? It is exact the same with like, the particular JSON construction the Codex shell tool uses. The model used to mangle nested quotes in this monstrosity RPC all the time but now it does not and it does not matter that the API is bad because billions of failed invocations are used to train to the harness we have, not the harness we deserve.

English
3
0
22
2.1K
sankalp
sankalp@dejavucoder·
are you the one prompting & managing or you are the one being prompted? are you deploying background agents or you are the background agent? these questions may arise. i suggest to retardmaxx here and if you are overwhelmed, take a break
English
0
0
2
477
sankalp
sankalp@dejavucoder·
first we RLed the models to make them agents, then we got RLed by the agents. they changed our coding patterns totally. lots of decisions and context switching in short times. maximizing parallel agents. cosmic background radiation shaped dread to always run agents. fatigue.
Cormac@cormachayden_

software engineers before vs after agents

English
2
0
20
1.8K

Keşfet

@BlackHC @erd0xbc @_catwu @trashpandaemoji @tensorqt @elonmusk @BarackObama @taylorswift13