Dhaval Adjodah

cofounder @makermaker_ai building agents (that build agents)^n.
Past: @MIT @schmidtsciences @medialab @Mila_Quebec @NYU_Courant @worldbank. from 🇲🇺

[Dhaval Adjodah @_dval_]

We pointed Elise, our RSI system, at inference (serving tokens from a trained model).
The results have economic implications via the @nvidia CUDA moat, quickly tuning a 2.8T model on a new @AMD stack, and improving NVIDIA's own model on their own hardware 🧵2/7

[Dhaval Adjodah @_dval_]

@AnushElangovan This is hard to do without RSI. A small team at @makermaker_ai did it months ago

[Dhaval Adjodah @_dval_]

We pointed Elise, our RSI system, at inference (serving tokens from a trained model).
The results have economic implications via the @nvidia CUDA moat, quickly tuning a 2.8T model on a new @AMD stack, and improving NVIDIA's own model on their own hardware 🧵2/7

[Dhaval Adjodah @_dval_]

@AndrewCurran_ It’s really hard to do this without RSI

[Dhaval Adjodah @_dval_]

We pointed Elise, our RSI system, at inference (serving tokens from a trained model).
The results have economic implications via the @nvidia CUDA moat, quickly tuning a 2.8T model on a new @AMD stack, and improving NVIDIA's own model on their own hardware 🧵2/7

[Dhaval Adjodah @_dval_]

@MarceloLima It’s really hard to do right unless you have RSI

https://t.co/bSvRyQMXFv

[Dhaval Adjodah @_dval_]

We pointed Elise, our RSI system, at inference (serving tokens from a trained model).
The results have economic implications via the @nvidia CUDA moat, quickly tuning a 2.8T model on a new @AMD stack, and improving NVIDIA's own model on their own hardware 🧵2/7

[Dhaval Adjodah @_dval_]

@dylan522p Here's a simulation of how an inference business works that might help explain some (but not all) of it
https://t.co/Au4h3fzWnc

[Dhaval Adjodah @_dval_]

@Sterling_Hooten @TechnoGoddy @BKad2005 @dylan522p Here's a simulation of how an inference business works that might help explain some (but not all) of it
https://t.co/Au4h3fzWnc

[Dhaval Adjodah @_dval_]

@guilleflorvs That makes sense especially given that RSI can make it cheaper/faster/better (pick 3!) to run your own open models

[Dhaval Adjodah @_dval_]

Running AI is expensive! @AnthropicAI's amended S-1 today shows $895B committed to compute and infrastructure but with Q3 revenue $17B and $567M loss, mostly from short-term compute.
@makermaker_ai can make it cheaper, better and faster with RSI. https://t.co/4BTjgaraqt 🧵1/7

[Dhaval Adjodah @_dval_]

@juliusadml thank you for sharing!

[Dhaval Adjodah @_dval_]

@JohnKapetaneas @makermaker_ai Thank you for sharing!

[Dhaval Adjodah @_dval_]

@thesimaximalist @AnthropicAI @makermaker_ai All these results dont change the weights of the model so quality is unchanged. Responses are therefore faster with no tradeoff

[Dhaval Adjodah @_dval_]

Running AI is expensive! @AnthropicAI's amended S-1 today shows $895B committed to compute and infrastructure but with Q3 revenue $17B and $567M loss, mostly from short-term compute.
@makermaker_ai can make it cheaper, better and faster with RSI. https://t.co/4BTjgaraqt 🧵1/7

[Dhaval Adjodah @_dval_]

That's the power of recursive self-improvement, and it makes compute better after you buy it.
Read more at https://t.co/4BTjgaraqt 🧵7/7
p.s. we're hiring!

[Dhaval Adjodah @_dval_]

If you want a feel for how hard this is, try running your own little inference business in our simulator: rent GPUs, serve models, and try not to go broke with Elise's help 👾 https://t.co/KvQCKejwgB 🧵6/7

[Dhaval Adjodah @_dval_]

Oh yeah we do self-improvement. plz come speculate over carbon-based food

[Steve Abreu @steveabreu7]

Hosting a small dinner for those interested in self-improving AI & systems at #MLSys2026 🍜 At MakerMaker we think self-improving AI should improve at all levels, including infra. If you’re into RSI, agentic systems, or AI infra and want good food + hot takes, join us ⬇

[Dhaval Adjodah @_dval_]

Congrats @juliusadml on this big milestone!

[Guide Labs @guidelabsai]

Excited to announce our $9M USD seed round led by @Initialized and the first large-scale interpretable LLM: an 8 billion parameter model capable of explaining its outputs through mechanisms humans can actually understand.

[Dhaval Adjodah @_dval_]

Congrats @AshwinRamaswami & al !

[Jack Cable @jackhcable]

AI revolutionized coding. Security hasn't kept up–until now.

Introducing @CorridorSecure: the future of secure coding.

We just raised a $5.4M seed from @Conviction and hired @alexstamos. Corridor is trusted by leading companies like @cursor_ai–and we’re just getting started. 🧵

[Dhaval Adjodah @_dval_]

@francoisfleuret @schmidtsciences is starting an entire funding program for this!

[Dhaval Adjodah @_dval_]

I'm launching a new program and a CALL for CONCEPT PAPERS to support research at the intersection of Artificial Intelligence and the humanities at Schmidt Sciences, a philanthropic initiative created and funded by Eric and Wendy Schmidt.