socially dead
everywhere
Joined October 2024
- Tweets18K
- Following3.1K
- Followers1.5K
- Likes64K
Biggest red flag is when you are for ed to use their product to get hired
🚨 Wispr Flow is hiring in India, but there's a twist.
No traditional resume. No boring application.
They want you to build a portfolio using your voice with Wispr Flow + Emergent and show them what you can actually create.
What's up for grabs?
> Internship opportunities (students can apply!)
> Full-time growth roles
> Top-of-market pay
> $3,000 work setup grant
They're judging creativity, ownership and actual work, not college tags.
📅 Deadline: November 4, 2026
If your portfolio can speak louder than your resume, this one's worth a shot.
Apply: wispr-seven-day-challenge.wi…
Adarsh retweeted
As a college fresher you should first give up that competitive mindset you developed in school.
You'll go much farther here in college if you learn to collaborate, network and seek help when required.
Adarsh retweeted
Found this super fun, so I took the neuron-explorer behind this post, made it model-agnostic, and went neuron hunting on 3 small models from 3 different families: Qwen3.5-0.8B, Gemma-3-1B and DeepSeek-Coder-1.3B.
Some of the most interesting finds below
github.com/sambhavnoobcoder/…
Adarsh retweeted
I've been digging into inference optimisations lately. A lot of the easy wins don’t change the answers at all - they just stop the GPU wasting time.
1/8
Thank you everyone for the messages .
Means a lot.
Just lost someone close today. He was 80 yo and hardly went to hospitals and was very healthy.
He had a heart attack this morning , he walked with us to the hospital afternoon. But dint respond to the treatment .
I have heard many such stories when some goes to the hospital (1/2)
I have heard many such cases where someone walks to the hospital relative in a better condition but has not been able to recover.
And in most of the cases they have hardly been to hospital in their entire life.
Maybe the lack of response may be due to fear of large hospitals.
Adarsh retweeted
I tested Jev classification on Anthropic’s blog inventory and the results were shocking.
Why Jev:
> it’s a System One model which doesn’t generate text the way ChatGPT does
> it’s based on the concept of RLCD (Reinforcement Learning for Calibrated Decisions) - acting as the decision-making layer for your software
> it returns a choice, score, or noul value for a question based on the context you provide which you can directly wire into your code to make decisions
> it doesn’t let one response affect another so context remains separate and doesn’t get carried over for every question you ask (the place where RLHF models repeatedly fail)
> it doesn’t charge you anything for output tokens so you get to control your wallet 😋
> fast asf
How I used Jev to create topical maps:
> the code fetches the sitemap for a website
> LLM decides the names for category/subcategory of every post from the sitemap
> Jev decides where to place each post based on relative and absolute scoring
Where it fits in Enfra’s interlinking workflow:
> client enters the domain
> sitemap for that domain gets fetched
> topical map created using LLM names + Jev classification
> when a new post is published, Jev gets to decide in which cluster it needs to be placed / say none exists and routes the post to LLM which creates a new category
> the post automatically gets the interlinks of every other post belonging in the same topical cluster (basis for the very popular hub-and-spoke model)
> every other post in the cluster can be refreshed once so that the new post makes it into their content as well
