Heeth Jain

Software Engineer (coding since 14 🧑‍💻)

I build backend systems that don't need a rewrite at scale!

[Heeth Jain @heethjain21]

@patrickc Not yet for India :(

[Heeth Jain @heethjain21]

Agents also scale incompetence!

It will get even harder to work with people who aren't willing to learn, since Agents (and AI overall) will only reinforce their ways.

[Sam Lambert @samlambert]

Agents scale expertise.

[Heeth Jain @heethjain21]

@AlexAndBooks_ First step is to get to a point where you are able to enjoy reading and it doesn't feel like a chore.

You will need to train yourself to do this i.e build a habit, so start with books on topics you enjoy.

Hence the saying, "Read what you love, until you love what you read"

[Heeth Jain @heethjain21]

@GergelyOrosz The real intelligence of LLMs lies in the static parameters (prompts, AGENTS.md, etc)

Hence it's output is still hugely dependent on the human behind it!

[Heeth Jain @heethjain21]

@iamtrask So, an AI agent which doesn't use a single model (itself), but rather multiple models ;)

[Heeth Jain @heethjain21]

Maybe I went too much overboard with zram.

[Heeth Jain @heethjain21]

Agreed. The government would keep the top tier models for themselves.

But there is a market for the models, so the 2nd/3rd best models will be for public use. I am still happy with the 2nd/3rd best model ;)

> You guys gotta see through all the marketing BS, nobody cares about the good of humanity or open source AGI

Absolutely. Corporates have one goal, to make money at the end of the day. Few individuals who work at companies do care, but not the company.

> I really think it's fitting they got shut down... after all, weren't they asking to put a stop to the development of capable models just a few days ago

I still believe Mythos isn't as great as Anthropic claims it to be. It's definitely a good leap, but Anthropic is after all a corporate looking to IPO soon, so I take it with a pinch of salt.

[Heeth Jain @heethjain21]

Thank God Chinese companies who are working on Open source models.

They'll catch up in some months ;)

[Leonardo Schwartz @leonardschwartz]

It's surprising that people actually get surprised by this.

When in history have the poor had access to the same tools, information and resources than the gov and the rich?

This has always been like this, and will always be.

People who are surprised by this just don't understand how the world works.

I've said almost 2 years ago that we all would have access to frontier intelligence as long as the labs needed us to refine their models and products enough to be sold to their actual customers, who are the govs and enterprises who can really afford paying millions of dollars to access frontier intelligence. They are just better customers!

That's also why subsidies are going away pretty soon and everyone will have to pay for every token. When that happen (which was scheduled to happen on June 23 anyway) the status quo will be broken and we'll go back to the world as it has always been: those who have the money to afford the tokens of the most capable models will have an unfair advantage over everyone else.

The yatch tour was nice while it lasted. Make sure to get a last scoop of caviar on your way out!

Just the way the world has always worked. Nothing to see here.

[Heeth Jain @heethjain21]

Man, I just love the internet!!

So many people with so much knowledge who takes effort to share their wisdom online so others can benefit from it.

I feel as if I have so many mentors for so many different aspects of my life!!

[Heeth Jain @heethjain21]

@shaneparrish I had totally forgotten you had written books! Ordered right away!!

I already can't get enough of your blogs/articles, thanks so much for the tons of wisdom!!!

[Heeth Jain @heethjain21]

@gat0rtheskater @asmartbear Each of his article are just pure gold

I feel as if I am being mentored by someone who is very very wise about business (and lot of life aspects as well)!!

[Heeth Jain @heethjain21]

Claude Fable is expensive. I'll just wait for the hype to slow down and the real reviews to come in.

As someone who doesn't have unlimited token budget, this is what I'll be doing now,

1. Use a better model like Opus 4.8 / GPT 5.5 for planning.
2. Use affordable GPT 5.4 Mini / Composer 2.5 / Sonnet 4.6 to implement.

Sonnet is only justified via Claude Code, otherwise, it's expensive with API based billing, while 5.4 Mini is really good for the price to performance it delivers!

I am yet to try Composer 2.5, but from the reviews, should be comparable enough to the other two.

[Heeth Jain @heethjain21]

There is also another added benefit of this thing.

The journey one has to undergo to reach this point will make them a financially responsible person.

[Heeth Jain @heethjain21]

Charlie Munger's famous "The first 100k is a b*tch" has the intent of how Capitalism works.

Keep your "spending/enjoyment/comfort from money" correlated to your net worth (exceptions aside), and you'll be set for life!!

Of course, after a point it can be correlated to income + net worth, with per-case basis, but the general framework is just that!

[Livingstone Mukasa @MukasaMulya]

If there is such a thing as an open secret, I’m going to let you in on one:
Capitalism favors owners (owners of capital).
Since most of us start with no inheritance, we enter the system with only our labor. 1/4

[Heeth Jain @heethjain21]

@karpathy Anthropic's IPO is going to be even bigger now ;)

[Heeth Jain @heethjain21]

This could go down as a pivotal moment for Anthropic!!

This isn't just about Andrej being really good at what he does, but also the second order effects that will happen because of him joining Anthropic.

The next pivotal moment for AI would need something new, and the R&D team will be doing just that!!

Not to mention, Anthropic's valuation probably just got a good bump from Andrej's joining!

[Andrej Karpathy @karpathy]

Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.

[Heeth Jain @heethjain21]

@Chris_Petkas House and Architecture analogy is really good.

The same goes for designers.

The difference here is that we'll have templates for Software, Design, etc which will make the "tinkerers" go all in and build it themselves.

[Heeth Jain @heethjain21]

@Kpaxs Humans are emotional creatures first, rational second.

[Chrisman @chrisman]

We all carry the unbearable burden of being the center of the universe. Then you have kids and that burden is lifted.

[Heeth Jain @heethjain21]

LLMs are becoming too "agreeable".

How can both opposites be true "for most people"?

I was trying to understand which approach is more beneficial long term, and thought maybe ChatGPT would have some insight.

Apparently, Gemini isn't much different either.