harsh

computer vision, ai agents, ml, dl....working on building skills!!

[harsh @kernel_crush]

🎾 Excited to share my computer vision project "Tennis-Vision" that brings data analytics to the tennis court! It tracks players, ball movement, and even classifies shots

[harsh @kernel_crush]

@prajdabre Xd

[harsh @kernel_crush]

@DarioAmodei your view on this ???

[Xiaoyin Qu @quxiaoyin]

My bet: @thinkymachines will soon make more money than @AnthropicAI. Not by winning the race to build one standardized frontier model. By becoming the Palantir FDE for enterprise custom models.

The playbook:
1. Release the best American open-weight model.
2. Drive widespread enterprise adoption.
3. Charge the largest companies 7–9 figures to post-train and run custom models behind their own firewall.

The model rests on three bets:
1. Large enterprises will increasingly demand their own models with their own data, and this is how they differentiate and win.
2. Enterprises won’t need just one model. They’ll continuously need new models for different workflows, departments, and proprietary datasets. That creates extremely sticky, recurring revenue.
3. Autoresearch will make custom model development increasingly scalable. Tinker can become the interface enterprises use to post-train their own models—with @thinkymachines providing the expertise and infrastructure behind it. FDE, infra, everything, huge contracts.
4. Eventually, maybe everyone wants their OWN model, and autoresearch and training inside tinker on top of @thinkymachines's base model will make it happen.

Meanwhile, Henry-ford-styled, standardized models will makes no margins. OpenAI and Anthropic will have their API margins squeezed by Deepseek/GLM/Grok/Meta etc, and their consumer subscriptions are loss centers.

The fat margin will move to customization: proprietary data, post-training, evals, deployment, and infrastructure.

If this thesis is right, @thinkymachines isn’t building just another frontier lab. It’s building the highest-value layer between frontier research and enterprise model ownership.

Turns out, the best business model for enterprise is NOT to sell commodity API access. Sell them their own models.

I’m extremely bullish on this approach.
@miramurati may be the most commercially savvy frontier-lab leader. I have to admit it.

[Avinash Singh @AvinashSingh_20]

Bangalore Job Seekers Kit - https://t.co/5TVbJbcyTC

European Tech Visa Sponsorship - https://t.co/0cG1NsgTzh

US Tech Visa Sponsorship - https://t.co/4GobLk06zi

Japan Tech Visa Sponsorship - https://t.co/QZvaDwuNAA

150 Remote Companies Hiring - https://t.co/Nyr3R9x3AG

1500+ Recent Funding Startups - https://t.co/yNRpm5oueq

ML & Data Science Mega Folder - https://t.co/jAEYrUjWNk

100% Remote Hiring Companies - https://t.co/XaXjPWHay5

Remote Job Seekers Kit - https://t.co/XaXjPWHay5

Indian Startups Freshers DS/ML/AI Hiring 2026 - https://t.co/IEaPKmJIe3

Latest Jobs & Internships - https://t.co/5TVbJbcyTC

Get startup bundle for more list - https://t.co/VEGcJCn3e2

Bookmark it!

[harsh @kernel_crush]

@karpathy @kellerjordan0 @huggingface Full technical deep dive — SSSL math, Newton-Schulz derivation, bin packing, val_bpb, the full autonomous loop:
https://t.co/9bXLw53aoS

[harsh @kernel_crush]

In March , @karpathy pushed autoresearch — a 3-file repo where a language model runs its own ML experiments.
Write code → train → read results → update hypothesis → loop.
No human in the loop for any ML decision.
I went deep on the codebase. Here's what's actually happening

[harsh @kernel_crush]

@karpathy @kellerjordan0 @huggingface That's not overlap. That's convergence. And convergence at this scale means the idea is becoming load-bearing for the field.

The researcher is becoming a module.

[Manthan Gupta @manthanguptaa]

This talk from @badlogicgames is such a wake up call and something I have been saying for a while. Slow the F down with this agentic vibe coding.

We are already seeing the consequences. More breaking software than ever, constant instability in dev tools, and a noticeable spike in security incidents, especially around supply chain attacks and poorly reviewed code getting shipped into widely used systems.

And then you see companies proudly saying “100% of our code is written by agents” like it’s a flex. It’s not. If anything, it’s a signal that nobody is actually reviewing what’s being shipped with the level of care required, especially when you are building IDEs, terminals, libraries, and infra that thousands or millions of developers depend on.

We have literally seen what happens when velocity outpaces discipline. Packages with massive adoption are getting compromised, critical tools are breaking under edge cases, and systems are becoming harder to debug because nobody fully understands the code anymore. When something goes wrong, you are not debugging your system, you are reverse engineering what an agent decided to do.

The whole “tokenmaxxing” mindset is also broken. Pushing billions of tokens through agents isn’t a flex if the output is over-engineered, brittle, and hard to maintain. LLMs have a natural tendency to expand complexity, not reduce it, and if you are not actively constraining that, you are just accumulating invisible tech debt at scale.

There’s also this narrative that “it doesn’t matter because agents will read the code anyway,” but that falls apart the moment your system grows. Codebases become bloated, abstractions pile up, and reliability drops. At some point, humans still have to step in, and now they are dealing with a system that was never designed with clarity in mind.

We are optimizing too much for speed and output, and not enough for correctness, simplicity, and long-term maintainability. And that tradeoff is already showing cracks.

Shipping fast is great. Shipping blindly isn’t.

[harsh @kernel_crush]

@ai_uncovered File

[harsh @kernel_crush]

@IDV_FPV So damn cool!!!

[harsh @kernel_crush]

@Gauri_the_great @abhi1thakur XD 🤣