Harsh Singhal

AI Governance @ Glean | Ex-ML @ Netflix, LinkedIn, Adobe | Led the ML team @ Koo

[Harsh Singhal @HiHarshSinghal]

Codex mobile controlling codex on my Mac mini is an amazing experience.

My side gigs don’t have to wait for weekends on evenings. I find myself checking in anytime inspiration shows up.

[Harsh Singhal @HiHarshSinghal]

@rajkstats Ongoing stuff. My new workhorse.

[Harsh Singhal @HiHarshSinghal]

Codex is just incredible.

Dam!!

[Harsh Singhal @HiHarshSinghal]

I care. I’ve been thinking about a commons approach to providing compute and AI inference to institutions like schools and colleges. I experiment with mini pcs and haven’t gotten very far on my vision but seeking to learn from other practitioners who care about this.
https://t.co/hpMchsyi3h is a manifesto I created to seek friends and thought partners on this journey.

[Harsh Singhal @HiHarshSinghal]

@claudeai @_cartick what you were building.

[Harsh Singhal @HiHarshSinghal]

@elonmusk I have great faith in FSD. I have used it ever since I got it.

I recently heard some heresy that Tesla collisions on FSD are not reported and are suppressed. I don’t believe it.

But when a car is driving me around at 65 mph and often doing it in the night (recently during inclement weather), these thoughts do cross my mind and I find myself taking over more than I used to.

FSD has been nothing short of a miracle in my experience. Just incredible it does what it does. I have a Model Y and model 3.

Cannot imagine driving without FSD. Not looking for assurances of any kind. I have the weight of experience.

But a part of me thinks, after all it’s technology and errors can happen. Human beings can make mistakes too, I know.

I joke that I trust FSD more than I trust myself. I’m just worried that it might be true.
That’s one more thing that I have to depend on that isn’t in my control. That isn’t free.

Maybe that is where the root of my fear comes from. It’s a defence mechanism to keep me from losing control 100%.

[Harsh Singhal @HiHarshSinghal]

I got a Beelink SER9 a few months ago and it was lying idle.

Until I installed Claude Code and fixed my keyboard layout to use the Mac layout.

This was stopping me from doing anything on the mini pc.

Intelligence of this kind is going to crystallize demand for all kinds of stuff!!

[Harsh Singhal @HiHarshSinghal]

When will we see videos of these robots doing laundry, loading the dishwasher, vacuuming around the house.

All I see them do is martial arts.

[Kyle Chan @kyleichan]

Unitree’s $30,000, 180cm (nearly 6 ft tall) H2 humanoid robot. This version is likely powered by an Nvidia Jetson AGX Thor platform, built on Blackwell architecture.

[Harsh Singhal @HiHarshSinghal]

@bcherny creating Claude Code is the step function change that will force demand on energy supply.

[Harsh Singhal @HiHarshSinghal]

Given how good Opus 4.5 is, its going to cause all other models to be used more.

In a sense, Opus 4.5 is the gravity well that all the other AI companies were waiting for.

Opus 4.5 is truly the AI that causes energy to fall short of all the demand its going to create!!

[Harsh Singhal @HiHarshSinghal]

I saw a question recently on how is CC different than using Opus in Cursor.

I dunno but I have now compared both for projects both at work and personal.

Claude Code in the terminal is something else!! Cal it harness or whatever. Its just different. And better.

[Harsh Singhal @HiHarshSinghal]

Everyone (and me): Claude Code is AGI for writing code. I can do so much. All my childhood dreams for side projects coming true.

Me: Create my own Agent for <<thing>> with Agent SDK and Opus 4.5 as the model.

[Harsh Singhal @HiHarshSinghal]

How to get a robot to code for you:

- HDMI capture card ($20)
- Raspberry Pi sending keystrokes
- Claude

The mannequin is optional but adds vibes.

[Benjamin De Kraker @BenjaminDEKR]

You must realize that if Claude can reliably browse the web and click around...... it can also do the same on any OS

Claude can use any computer where it can see the screen, move the mouse, and enter key commands

[Harsh Singhal @HiHarshSinghal]

@gdb @_cartick

[Boris Cherny @bcherny]

I'm Boris and I created Claude Code. Lots of people have asked how I use Claude Code, so I wanted to show off my setup a bit.

My setup might be surprisingly vanilla! Claude Code works great out of the box, so I personally don't customize it much. There is no one correct way to use Claude Code: we intentionally build it in a way that you can use it, customize it, and hack it however you like. Each person on the Claude Code team uses it very differently.

So, here goes.

[Harsh Singhal @HiHarshSinghal]

Biggest surprise: 665 of them are blogs teaching affiliate marketing.

The most meta niche. 🐍

[Harsh Singhal @HiHarshSinghal]

Built something fun with Claude Code 🔧

Took Common Crawl data on AWS, scanned for sites using Amazon's affiliate program, then:

- Clustered their meta descriptions using BERTopic
- Named the clusters with an LLM
- Generated an HTML report

All orchestrated through Claude Code - from writing the Spark jobs to the final visualization.

Check it out: https://t.co/u7l7sjJpp7

The rabbit holes you find in public datasets never disappoint.

[Harsh Singhal @HiHarshSinghal]

@fchollet Any recommendations?

[Harsh Singhal @HiHarshSinghal]

@bcherny @giffmana @giuseppegurgone I used Agent SDK and gave it pandas as a state manager. This was specific to doing data analysis on a CSV file but it worked well.

I would assume using a simple SQLite local db will also help but the suggestion to use a https://t.co/hGDo0l7XQc file makes sense to start simple.

[Harsh Singhal @HiHarshSinghal]

@jedisct1 @_cartick

[Harsh Singhal @HiHarshSinghal]

Never leave Claude Code idle.

An idle mind is the devils workshop.