Harshith

23, Head of Growth @getjustpaid (@ycombinator W23) | Council Member @ForbesTechCncl | Crypto Native | Vibe Coder

[Harshith @harshithio]

Excited to share that ARTIFIN, Inc. has been acquired by a San Francisco-based private equity firm. Big thanks to @acquiredotcom for making the acquisition smooth.

Major props to @levelsio and @marclou for the burst of inspiration that fueled our journey.

Hats off to @NithinSikinam for the technical magic into our stack. Immense gratitude to advisors, @iamgaurangdesai and @_lazypoet_, for their invaluable strategic guidance.

Here’s the deal: it’s not just deep work, it’s about strategic sprints that lead to superlinear outcomes. #BuildingInPublic has been my open lab. Wrapping up 2023 with lessons in hand and ready to take on projects that redefine 'value'.

To those who labeled my project as just another GPT wrapper : value is in the solution, not the complexity.

Living unapologetically, and choosing ventures that ignite growth.

Here’s to the creators, builders, and dreamers - May your work resonate, disrupt, and thrive.

On to the next...

[Harshith @harshithio]

@Kappaemme1926 claude is dumb fr.

[Harshith @harshithio]

Paxel by YC reads your Claude, Codex and Cursor sessions and shows how you actually build with AI. 300k+ sessions analyzed. #AI #Coding #YCombinator

[Yuchen Jin @Yuchenj_UW]

Think of yourself as an LLM.

Every social interaction, every meeting, burns your tokens.

Unless someone is a paid subscriber to your attention, you are under no obligation to answer low-quality prompts.

[Harshith @harshithio]

The JustPaid AI chat bot

https://t.co/MSF4ntWevD

[Harshith @harshithio]

@thsottiaux Azure OpenAI API -

Error running remote compact task: We're currently experiencing high demand, which may cause temporary errors.

[Harshith @harshithio]

https://t.co/Q4UV56MMvd in 3 minutes.

[Harshith @harshithio]

Codex started as an AI tool for coders.

Quietly, it became something much bigger:

A tool for everyone else to get work done.

5 million people use it every week.

And 1 in 5 Codex users is not a developer.

That non-dev group is growing 3x faster than engineers.

#Codex

[Harshith @harshithio]

@TheRohanVarma the xhigh is unusable now...

the context is too low.

[Harshith @harshithio]

gpt 5.5 xhigh feels worse now. context is too thin, can’t hold a thread. 250k tokens is basically nothing at xhigh.

tested opus 4.8 xhigh: 1m context + workflow, much better.

they nuked gpt 5.5. i miss the launch version.

[Harshith @harshithio]

opus 4.6 > 4.7 & 4.8

[Harshith @harshithio]

https://t.co/6iL25WhJbH now has an AI assistant! Get instant answers on automating finances and B2B billing. Just click "Ask JustPaid" on the bottom right. You can ask anything, from collections to API usage, or book a demo with the sales team.

[Harshith @harshithio]

With AI, nothing is beyond your reach. Distill complex research papers to your level, even as a non-coder. The era of ideas is here – what will you build?

#AI #Innovation #FutureOfWork

[Harshith @harshithio]

Unlock Codex's power with these 2 tips! Use /goal for auto-task completion and @browseruse (with @computeruse) for hands-off work. Optimized with GPT 4.5.
#Codex

[Harshith @harshithio]

GPT-5.5 is genuinely impressive. The reasoning, the nuance, the consistency — it's a step above. Hate to say it, but it's better than Claude right now. OpenAI cooked on this one. 🔥

[Harshith @harshithio]

@Dimillian We use Codex via Azure API.

And even the basic plugins are not available.

And codex app only works with connected to gmail.

need a Azure accounts to connect to gmail so we can use full features.

[Harshith @harshithio]

nah...

very very very fishy.

[Alexander Whedon @alex_whedon]

Introducing SubQ - a major breakthrough in LLM intelligence.

It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA),

And the first frontier model with a 12 million token context window which is:

- 52x faster than FlashAttention at 1MM tokens
- Less than 5% the cost of Opus

Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention).

Only a small fraction actually matter.

@subquadratic finds and focuses only on the ones that do.

That's nearly 1,000x less compute and a new way for LLMs to scale.

[Harshith @harshithio]

@sonofalli cuz not everyone knows how to use terminal.

I ask codex to run psudo commands.

[Harshith @harshithio]

@Miguel07Code based.

[Harshith @harshithio]

ok so Remotion or Hyperframes.

[Harshith @harshithio]

@Miguel07Code create a catalog of all prompts with examples.

my agent chooses HyperFrames - but struggles with the output cuz it has no skills / prompts.

[Harshith @harshithio]

@zenorocha If I'm getting a <name>@resend.com then I'm in!!