Sandeep A Nair

Full-stack engineer. Building learning tools, games and small experiments with AI. Sharing demos, decisions and what actually worked.

[Sandeep A Nair @Sanair96]

@ANI Talks about fabricated media events in a fabricated media event 🫡

[Sandeep A Nair @Sanair96]

@RoshanKrRaii Talks about fabricated media events in a fabricated media event 🫡

[Sandeep A Nair @Sanair96]

Didn't know that Dot's had limits, more so didn't think I would be hitting it! @OpenAI @thsottiaux

[Sandeep A Nair @Sanair96]

I really hope one of the 28 days is going to be launch of multiple dots, using a single one for everything, while powerful, causes the dot to miss tracking some tasks just because it get's caught up doing others.

I always have to keep following up on whats up with this, what's up with that etc.

[Sandeep A Nair @Sanair96]

Another interesting message example that I see

[Sandeep A Nair @Sanair96]

It's interesting to see how my dot messages the threads it's managing. It does not use spaces between words. Is this part of the token efficiency gains/optimizations that OpenAI has been working on?

Makes sense since LLM's do not need to read the way we do and these threads are meant to be fully managed by the Dot.

But the agent that responds in the thread does not know this and still responds normally with the spaces and formatting which means that if this actually leads to token savings then we aren't getting that fully yet as the agent responses in the thread should also be optimized.

May be the initial message that the Dot sends could have the instruction telling this is not a user monitored thread and that the agent can take shortcuts to respond and save tokens? @thsottiaux

[Sandeep A Nair @Sanair96]

@maddhruv I should give this a try.

[Sandeep A Nair @Sanair96]

@maddhruv Running actual user tests using the AI with computer use has been a game changer for me personally. Definitely better than the playwright tests to catch nuances

[Sandeep A Nair @Sanair96]

@thsottiaux Long running threads freeze the UI. It becomes unusable until you force close the application and restart it.

[Sandeep A Nair @Sanair96]

Had a lot of fun learning the basics to draw shapes in HTML canvas.
All code for this was written by hand so that I can grasp the logic behind these while I used AI to unblock me by only chatting, no direct changes.

Next I want to learn to manipulate the shapes that have been drawn, may be an eraser next so that it becomes closer to MsPaint.

Going great so far.
Github Repo: https://t.co/NyCgE9GW7q

[Sandeep A Nair @Sanair96]

Honestly @emilkowalski 's https://t.co/9Tdnflo2OL course is what sent me on this trajectory. I was looking for inspiration for one of the exercises and came across
https://t.co/VKiCpRbLRw
which intruiged me on how to achieve this.

That rabbit hole was a fun one!

[Sandeep A Nair @Sanair96]

Ran out of #Codex usage, was wondering what to do.

I wanted to learn Canvas from a very long time, so I did it.
Honestly to learn and implement all of this would have taken me multiple days, but with the AI to assist in learning, You can just do things!

[Sandeep A Nair @Sanair96]

Looking forward to learning HTML canvas like this further. This was just the basics.
Repo: https://t.co/GkVtZdG5d1

Not cleaned up or anything, this was just for learning!

[Sandeep A Nair @Sanair96]

@chrisbanes @ChatGPT I have the same issue, I didn't realise this until your post, I ran out of usage in a single night on the $200 plan. @thsottiaux @ChatGPT @OpenAI

[Sandeep A Nair @Sanair96]

If BJP gave out stocks, that’s where I would have invested in 2014 to get like a 150x return.

[Sandeep A Nair @Sanair96]

I compared the financial position of the BJP and Congress using their declared income, assets, donations, and expenditure from 2001.

The data is from ECI website.

The gap isn’t merely large, there’s a night and day difference.

Take a look for yourself!

https://t.co/XrFN1RzWxm

[Sandeep A Nair @Sanair96]

The interesting part is how both parties had similar kinds of funds until BJP came into power in 2014 and then it seems like BJP managed to get a huge boost(almost a 100x growth) on a scale never seen before.

[Sandeep A Nair @Sanair96]

@thsottiaux I’m using Codex work for job application automation, would be interested to be part of this early access.

[Sandeep A Nair @Sanair96]

@OpenAI @thsottiaux This is actually causing all my workflows to be broken because now even ultra is unable to use subagents.

[Sandeep A Nair @Sanair96]

@OpenAI @thsottiaux For some reason my agents are unable to create subagents. I tried this in the CLI and the desktop app as well. Nuked the entire application and the .codex folder also, still shows the same issue.
Doesn't work with Full Access as well.