Mamal Amini

CEO @ GovernGPT | Y Combinator

[Mamal Amini @AminiMamal]

@mignano I found skiing is my focus vacuum. Art museums is the second focus vacuuum.

[Mamal Amini @AminiMamal]

@turbopuffer @SID_AI Can’t wait to get this into our systems!

[turbopuffer @turbopuffer]

SID-1 is an agentic search model by @SID_AI

→ 1.9x recall over RAG + rerank
→ 24x faster, 99% cheaper than GPT-5.1

trained using large-scale RL on turbopuffer at 1k+ QPS bursts over 10M+ document corpora across thousands of steps

https://t.co/hqmdPUmdLt

[Mamal Amini @AminiMamal]

@KrisLachance Always such an honour hosting you!

[Kris Lachance @KrisLachance]

Always good to experience the GovernGPT hospitality

[Mamal Amini @AminiMamal]

Growth is non-linear; more importantly you can amass leverage in exponential fashion.

[Mamal Amini @AminiMamal]

I strongly believe A-player leverage will 10x over the next few years.
The overlap of passion and aptitude is the only way to win.

[Mamal Amini @AminiMamal]

Biggest leverage I see with AI, isn’t the first product - it’s the second and third product that’s built on on top of the first product.

[Tom Blomfield @t_blom]

🧑‍💻 Request for Design Partners 👩‍💻

At YC, I work with a number of highly technical young founders who can build mind-blowing stuff with the latest AI-based tools.

But they don't yet have connections to old, established industries, and so they're struggling to identify problems that big businesses will pay for.

To bridge this gap, I'm looking for people working in the following sectors:

- Banking
- Insurance
- Law
- Audit & Compliance
- Healthcare

If you're doing people-heavy services work and you think that some LLM-powered software would make your job 10x easier, I'd love to connect you with some smart technical founders in your space 🚀

[Matthew Berman @MatthewBerman]

Anthropic just dropped an incredible guide on "How To Build Effective Agents"

2025 will be the year of AGENTS 🤖

Here's everything you need to know: 🧵

[JD @JD_DeYonker]

Even the richest guy on earth pulls the “it’s oversubscribed” line when raising money.

[Mamal Amini @AminiMamal]

@JD_DeYonker I love this guy!

[Brett Adcock @adcock_brett]

Researchers introduced ExBody2 an advanced whole-body controller for humanoid robots

It essentially allows humanoids to mimic human-like motions better through training with RL in simulation and transferring to real robots
https://t.co/qdMmzutVzl

[Brett Adcock @adcock_brett]

Google released Gemini 2.0 Flash Thinking Experimental

The model pauses to "think" through complex problems like OpenAI's o1, but is free to use through AI Studio and the Gemini API https://t.co/AcBjiKu4N5

[Brett Adcock @adcock_brett]

Figure update: 31 months after establishing our company, Figure now generates revenue!

This week, we delivered Figure-02 humanoid robots to our commercial client, & they're currently hard at work
https://t.co/iiEF1WxhHw

[FoundersBrief @foundersbrief]

I recently sat down with @AminiMamal, YC-backed founder and CEO of GovernGPT, to discuss how he is automating manual tasks within the private markets. Excited to see his company's growth and continued innovation within the space!

https://t.co/L6HEYph8pp

[CIFAR @CIFAR_News]

How can machine learning students make an impact in their work?

Canada CIFAR AI Chair @apsarathchandar (@Mila_Quebec @polymtl @UMontreal) gives advice to ML students about the value of engaging in applied projects in addition to academic work.

[Andrej Karpathy @karpathy]

100% Fully Software 2.0 computer. Just a single neural net and no classical software at all. Device inputs (audio video, touch etc) directly feed into a neural net, the outputs of it directly display as audio/video on speaker/screen, that’s it.

[Mamal Amini @AminiMamal]

@frantzfries @frantzfries you should talk with @maxrumpf

[Max Rumpf @maxrumpf]

I'm sure there's a technical fix people will try: Weigh the citations/impact with some expected value based on existing h-index/university ranking. A bit like TF-IDF. But I'm sure that can be gamed and people will game it.

I don't really have a good solution – I mostly read stuff on referral or if it's cited by a paper I really liked and read. But that really doesn't ensure diversity of ideas: People only recommend what confirms their beliefs and only cite what agrees with them (obviously).

Generally, things get better as a research area becomes more mature, because for obscure stuff it's almost impossible to tell quality. "Does this new NN architecture really outperform given the only benchmark was also invented by the authors in the same paper?"