AI/Machine Learning, Bharat and Bhartiya IT Industry

[bala]

The tryanny of AI posts is that those who use them extensively slant it towards their viewpoint, which is easy to tailor prompts (i.e. engineer the prompts so that it is favorable towards your viewpoint) to get LLMs to generate your nonsense. But the Human mind is more adept in discerning such wanton slanting of opinions. Also facts are driven by RAGs (tis an AI component for LLMs) and these RAGs are biased "facts" which is taken as gospel truth.

[williams]

IMO first we need to be clear what Generative AI is all about. It is just a statistically most probable predictor of the next thing based on current context. Agents (and are planners) are automators of this predicting mechanism. So it is just a tool which could be used to eliminate routine boiler plate tasks without much scrutiny. Things like spell check, grammar check, generating basic code, music, video etc can by done and hence save a lot of time.

Then comes the issue of going beyond boiler plate tasks. Here it is necessary for a subject matter expert to review and use what is worth and reject what is not worth. We are going through the same phase when engineers debated about use of calculators. I still have my Dad's slide rule and I have seen him using it before transitioning to calculators. Point is, these are tools to augment human capacity. For AI to work you need good data and a good model. The former (data) is where people need to put a lot of serious effort. The paradigm is shifting and we better learn to adapt to it.

[Vayutuvan]

I love my Aristo.

[Vayutuvan]

Chinese grey market sells Claude API access at 90% off by using stolen credentials, model substitution, and harvesting users' prompts and outputs for resale as AI training data — 'transfer stations' operate through proxy networks that harvest user data
May 9, 2026

https://www.tomshardware.com/tech-indus ... -user-data

A grey-market economy of API proxy services in China is reselling access to Anthropic's Claude models at as little as 10% of the official price, according to an investigation published Monday by Oxford China Policy Lab researcher Zilan Qian.

The proxy networks, known in Chinese developer communities as "transfer stations," operate openly on platforms including GitHub, Taobao, and Telegram, and sustain their rock-bottom pricing through a combination of stolen credentials, model substitution, and harvesting users' prompts and outputs for resale as AI training data.

These findings give credence to the warnings issued in recent weeks by both the White House and Anthropic, the former of which accused Chinese entities in late April of running “industrial-scale” distillation campaigns against U.S. frontier models using tens of thousands of proxy accounts. Anthropic disclosed similar activity in February, identifying roughly 24,000 fraudulent accounts linked to Chinese labs, including DeepSeek, Moonshot AI, and MiniMax.
...

[uddu]

https://x.com/i/status/2054110579300958609
@c_aashish
Sarvam is now preparing to train its first trillion-parameter AI model within next nine months, marking what could become a major milestone for the country’s indigenous AI ambitions.

[uddu]

https://x.com/i/status/2057932592901755242
@elliotarledge
Co-Founder of Cerebras explains their WSE simplified design compared to classical GPUs made by NVIDIA.

[Vayutuvan]

Any of you AI folks using this "Kubernetes-like orchestration for AI coding"?

Welcome to Gas Town
Steve Yegge Jan 1, 2026

https://steve-yegge.medium.com/welcome- ... 25ee16dd04
Happy New Year, and Welcome to Gas Town!

Figure 1: Welcome to Gas Town

What the Heck is Gas Town?
Gas Town is a new take on the IDE for 2026. Gas Town helps you with the tedium of running lots of Claude Code instances. Stuff gets lost, it’s hard to track who’s doing what, etc. Gas Town helps with all that yak shaving, and lets you focus on what your Claude Codes are working on.

For this blog post, “Claude Code” means “Claude Code and all its identical-looking competitors”, i.e. Codex, Gemini CLI, Amp, Amazon Q-developer ClI, blah blah, because that’s what they are. Clones. The industry is an embarrassing little kid’s soccer team chasing the 2025 CLI form factor of Claude Code, rather than building what’s next.

I went ahead and built what’s next. First I predicted it, back in March, in Revenge of the Junior Developer. I predicted someone would lash the Claude Code camels together into chariots, and that is exactly what I’ve done with Gas Town. I’ve tamed them to where you can use 20–30 at once, productively, on a sustained basis.

Gas Town is opinionated — much like Kubernetes, or Temporal, both of which Gas Town resembles, at least if you squint at it until your eyes are pretty much totally shut. I’ll include comparisons to both k8s and Temporal at the end of this post. It is a little surprising how similar they are, despite having radically different underpinnings.
...

[A_Gupta]

https://www.vatican.va/content/leo-xiv/ ... nitas.html
ENCYCLICAL LETTER
MAGNIFICA HUMANITAS
OF HIS HOLINESS
POPE LEO XIV
ON SAFEGUARDING THE HUMAN PERSON
IN THE TIME OF ARTIFICIAL INTELLIGENCE

[Vayutuvan]

Is AI Profitable Yet?
Tracking the spend and revenue of frontier AI companies (May 2026).

https://isaiprofitable.com/

[bala]

Zoho-backed Netrasemi launches first AI chip, begins customer trials



The A2000 is among the first AI/ML chips to be designed in India. Netrasemi was one of four startups selected for Rs 15 crore support under the government's Design Linked Incentive (DLI) scheme in 2023. Rs 125 crore in total funding to date from various investors, including Zoho and Unicorn India Ventures.

The A2000 is built on TSMC's 12nm process and targets smart cameras, edge AI boxes and intelligent video gateways. The chip packs several processing units developed in-house by Netrasemi, including dedicated engines for neural processing, computer vision, image handling and security. Netrasemi has developed a second chip, the R1000 AI/ML MCU, targeting the IoT sensor market.

Volume production, at chip manufacturer TSMC's Taiwan facility, is slated for 2027. We have designed this chip specifically for compact, power-efficient edge devices. We are currently working with several OEMs on early sample evaluations and co-development.

Link

[Lisa]

I wrote this almost 2 months ago in Topic: Modi 3.0 - Bharat

https://x.com/dwarkesh_sp/status/2019458363495456894

First 20 mins or so. If you agree then how does India get there? We need to make the generational leap.

Added now. Musk is proposing 500-1000 TW of power in space. If I am not mistaken, even the lower figure exceeds multiples of total current Indian output. IMHO Somebody needs to quickly look into this.

Kardashev, Kardashev, Kardashev! Solar, Solar, Solar!

Furthermore,

https://x.com/LaceyPresley/status/2060436135671632067

TERAFAB IS GOING TO BE INSANE.

We’re targeting 100–200 billion custom AI + memory chips per year at full ramp that’s 1 terawatt (1,000 GW) of annual AI compute capacity. Roughly 50x current global AI chip output.

Breakdown:
- 80% (~160 billion chips / 800 GW) → radiation-hardened D3 chips for orbital data centers. Space-based compute at massive scale, powered by solar, low latency for Earth, immune to most terrestrial risks.

- 20% (~40 billion chips / 200 GW) → terrestrial AI5 & AI6 edge inference processors for Tesla vehicle fleets and Optimus robots.

Facility plan: start at 100k wafer-starts/month, scale to 1 million wafers/month. Everything design, EUV lithography, fab, memory, packaging, test under one roof. That recursive self-improvement loop is the real unlock.