dejavucoder's profile picture. LLMs and shitposting
into crafting ai product and evals
dm open to talk on ai engineering/post-training

sankalp

@dejavucoder

LLMs and shitposting into crafting ai product and evals dm open to talk on ai engineering/post-training

固定されたツイート

prompt caching is the most bang for buck optimisation you can do for your LLM based workflows and agents. in this post, i cover tips to hit the prompt cache more consistently and how it works under the hood (probably the first such resource) sankalp.bearblog.dev/how-prompt-cac…

dejavucoder's tweet image. prompt caching is the most bang for buck optimisation you can do for your LLM based workflows and agents. in this post, i cover tips to hit the prompt cache more consistently and how it works under the hood (probably the first such resource)

sankalp.bearblog.dev/how-prompt-cac…

dejavucoder's tweet image.

i have never heard a simple clear one line explanation of what a product manager actually does which is kind of hilarious. what is it?



its a good rl environment, sir

Today we started rolling out SimGym — a system that creates “digital customers” that behave like real ones. They browse your site, complete tasks, and reveal optimization opportunities. You can even run A/B tests with *zero* live traffic! Spent a year developing it.

MParakhin's tweet image. Today we started rolling out SimGym — a system that creates “digital customers” that behave like real ones. They browse your site, complete tasks, and reveal optimization opportunities. You can even run A/B tests with *zero* live traffic! Spent a year developing it.


who is building robots like TARS

dejavucoder's tweet image. who is building robots like TARS

i approve of this eval

Hazelnut vs Nano Banana Pro "Create an infographic that shows how to make elaichi chai"

Angaisb_'s tweet image. Hazelnut vs Nano Banana Pro

"Create an infographic that shows how to make elaichi chai"
Angaisb_'s tweet image. Hazelnut vs Nano Banana Pro

"Create an infographic that shows how to make elaichi chai"


when i run gpt-5-codex-max reviews on opus 4.5 PRs (opus 4.5 produces way less slop than sonnet 4.5 but it can miss some spots of changes. if you are moving v fast, make sure to review extensively too. ask opus to self-review as well)

dejavucoder's tweet image. when i run gpt-5-codex-max reviews on opus 4.5 PRs (opus 4.5 produces way less slop than sonnet 4.5 but it can miss some spots of changes. if you are moving v fast, make sure to review extensively too. ask opus to self-review as well)

the chai must go on

dejavucoder's tweet image. the chai must go on

hope gpt 5.2 will help clear the tech debt

the downstream effects of claude 4.5 opus will be studied

nbashaw's tweet image. the downstream effects of claude 4.5 opus will be studied


winter dryness led me to gworlmax and buy cetaphil. its effective but i wish they added some fragrance to it coz it smells bad...


alex fazioooooo my saviour. whenever my twitter algo momentum is low he comes to like all my recent tweets

dejavucoder's tweet image. alex fazioooooo my saviour. whenever my twitter algo momentum is low he comes to like all my recent tweets

kpop is bioweapon

dejavucoder's tweet image. kpop is bioweapon

Sydney Sweeney is donating her coding agent experience logs to the Agentic AI Foundation, a directed fund under the Linux Foundation. In 1 year, her logs on X about coding agents have become go-to reference. Joining AAIF ensures the community always stays up to date with SoTA.

dejavucoder's tweet image. Sydney Sweeney is donating her coding agent experience logs to the Agentic AI Foundation, a directed fund under the Linux Foundation.

In 1 year, her logs on X about coding agents have become go-to reference. Joining AAIF ensures the community always stays up to date with SoTA.

if you use kimi k2 for production use cases, what provider do you usually use


sometimes you have to consider the possibility that you are being dumb

dejavucoder's tweet image. sometimes you have to consider the possibility that you are being dumb

sankalp さんがリポスト

We are a Small Research Lab based out of india and we just dropped a one of a kind SoTA multimodal multilingual document retrieval model. can embed and query documents across 22 langauges : English, Spanish, French, German, Italian, Hindi, Marathi, Sanskrit, Kannada, Telugu,…

Today we are excited to launch NetraEmbed SoTA multimodal multilingual document retrieval model. > Supports 22 languages > ~ 150% improvement over existing baselines > NayanaIR-Bench a open source multilingual document retrieval benchmark

cognitivelab_ai's tweet image. Today we are excited to launch NetraEmbed
SoTA multimodal multilingual document retrieval model.

> Supports 22 languages
> ~ 150% improvement over existing baselines
> NayanaIR-Bench a open source multilingual document retrieval benchmark


peanuts - a tiny AI powered wearable by anya forger

dejavucoder's tweet image. peanuts - a tiny AI powered wearable by anya forger

i knew nano banana pro was very good but i didn't realise it was this good {   "subject": {     "description": "25-year-old Japanese woman with a curvaceous figure, standing in a meadow before Mount Fuji, gentle breeze lifting her hair",     "pose": "relaxed stance, one hand…

dejavucoder's tweet image. i knew nano banana pro was very good but i didn't realise it was this good

{
  "subject": {
    "description": "25-year-old Japanese woman with a curvaceous figure, standing in a meadow before Mount Fuji, gentle breeze lifting her hair",
    "pose": "relaxed stance, one hand…


Loading...

Something went wrong.


Something went wrong.