Iñigo

investigating language, machines, and the brain at UC Berkeley

[Iñigo @IParraMartin]

Crazy

[Iñigo @IParraMartin]

Real researchers read papers; they don't scroll LinkedIn feeds

[Iñigo @IParraMartin]

@docmilanfar Literally lol

[Iñigo @IParraMartin]

How?

[Xin Eric Wang @xwang_lk]

“ACL is Not an AI Conference” is still hurting ACL today.

[Gabriele Berton @gabriberton]

The Claude Code leak teaches us that simplicity wins once again

To detect frustration they literally use a regex that searches for (wtf|shit|...)

If you gave that task to almost any PhD student they'd have trained a 200M "frustration detection" model

[Ryan Carson @ryancarson]

You still need to know how to code

[Yann LeCun @ylecun]

@nxthompson So much BS

[Iñigo @IParraMartin]

¿Por qué no te callas?

[Hermann Tertsch @hermanntertsch]

Sorry, @WSJ
Sánchez is in opposition to Trump, yes, and hin opposition to every decent Spaniard. Sánchez wages war against the Spanish nation he rules in an illegal and illegitimate way, own people, against the Spanish law and Spanish constitution. Sánchez is al furious antisemite, friend of all criminals on the world stage, from drug dealers to terrorists to thieves, organised crime and money laundry with and by his party. With Zapatero he is the worst criminal to rule Spain, worse than the communists in the civil war.

[Iñigo @IParraMartin]

We need more tweets like this and less hype coming from tech bros

[Jianyang Gao @gaoj0017]

We need to publicly clarify serious issues in Google’s ICLR 2026 paper TurboQuant.

TurboQuant misrepresents RaBitQ in three ways:
1. Avoids acknowledging key methodological similarity (JL transform)
2. Calls our theory “suboptimal” with no evidence
3. Reports results under unfair experimental settings

We have expressed our concerns to the authors before their submission, but they chose not to fix them in their paper submission. The paper was accepted at ICLR 2026 and heavily promoted by Google (tens of millions of views). At that scale, uncorrected claims quickly become “consensus.”

Facts:
1. RaBitQ already proves asymptotic optimality (FOCS’17 bound)
2. TurboQuant uses the same random rotation step but misses stating the connection
3. Their experiments used single-core CPU for RaBitQ vs A100 GPU for TurboQuant

None of these is properly disclosed.

We’ve filed a formal complaint and posted on OpenReview (https://t.co/nDVjmNhATM).
We’ll release a detailed technical report on arXiv.

Our goal is simple: keep the academic record accurate.

Would appreciate people taking a look and sharing.

[Peter Holderrieth @peholderrieth]

🚀MIT Flow Matching and Diffusion Lecture 2026 Released (https://t.co/bKgs2wghvY)!

We just released our new MIT 2026 course on flow matching and diffusion models! We teach the full stack of modern AI image, video, protein generators - theory and practice. We include:
📺 Videos: Step-by-step derivations.
📝 Notes: Mathematically self-contained lecture notes
💻 Coding: Hands-on exercises for every component

We fully improved last years’ iteration and added new topics: latent spaces, diffusion transformers, building language models with discrete diffusion models.

Everything is available here: https://t.co/bKgs2wghvY

A huge thanks to Tommi Jaakkola for his support in making this class possible and Ashay Athalye (MIT SOUL) for the incredible production! Was fun to do this with @RShprints!

#MachineLearning #GenerativeAI #MIT #DiffusionModels #AI

[Iñigo @IParraMartin]

Vaya ridículo

[EL PAÍS @el_pais]

📺TV en DIRECTO | María Corina: "España en Europa ha sido el vínculo con América Latina. Habríamos aspirado a que tuviese el liderazgo en esta materia, lamentamos que no haya sido así. Cada gobierno decidirá si está del lado del crimen o de la justicia" https://t.co/SLnYYFNw9t

[Iñigo @IParraMartin]

Anthropic had a better model to begin with

[Jason Locasale @LocasaleLab]

PhD and master's-trained scientists spend years or decades working in laboratories earning salaries around $60K. They produce the bulk of the experimental work that generates grants, publications, and institutional prestige.

Universities have developed a labor model that uses artificial titles such as fellow or instructor, prolonged trainee designations, and other mechanisms to lower status, while relying heavily on highly trained scientists compensated far below what comparable education and productivity would command in other sectors.

All of this occurs within tax-advantaged institutions that receive substantial public subsidy.

[Iñigo @IParraMartin]

Absolutely crazy take

[Cheng Lou @_chenglou]

Stupidly late realization on why LLMs are so good at reasoning: human’s reasoning capability is bottlenecked by language! It’s not that languages are good at reasoning; reasoning ended up being defined by language first and foremost. The medium truly shapes the message

[Iñigo @IParraMartin]

@justinskycak I think the original positional encodings (absolute) are SO underrated

[Iñigo @IParraMartin]

@karinarichavala Sorry I just saw this!

[Iñigo @IParraMartin]

After reading more about this, one thing is clear: punishing a professor for calling out a student who used the Bible as a scientific source is absurd. Freedom of speech shouldn’t be confused with the expectation that nonsense will be taken seriously or earn academic credit

[University of Oklahoma @UofOklahoma]

Statement from the University of Oklahoma:

[Iñigo @IParraMartin]

@LuizaJarovsky No it won’t. Companies will keep stealing the books.

[Ian Copeland, PhD @IanCopeland5]

This is just an affirmation for those who flunked out of college because they couldn't complete the course material and need an excuse to be mediocre.

I said what I said...

[Andrew Ng @AndrewYNg]

https://t.co/Q3Ob7lRXRj

[Iñigo @IParraMartin]

@FreightAlley I was at Neurips this year and it looked like a VC convention instead of a conference. And I totally agree with you.