Víthor Rosa Franco

A human consumes about 2,000 calories per day. Over 20 years, that’s roughly 17,000 kWh of total food energy. Training GPT-4 consumed an estimated 50 GWh of electricity. That’s 3,000 humans worth of “training energy” for a single model run. And GPT-4 is already dead. OpenAI retired GPT-4o from ChatGPT on February 13th. The model that took 50 GWh to train got less than two years of flagship status before replacement. The human you spent 17,000 kWh “training” for 20 years produces economic output for the next 40 to 60 years. The amortization window on GPT-4 was shorter than a car lease. Now look at what replaced it. GPT-5.2, released December 2025, is OpenAI’s current default. The GPT-5 series consumes an estimated 18 Wh per average query according to the University of Rhode Island’s AI Lab, up to 40 Wh for extended reasoning. That’s 8.6 times more electricity per response than GPT-4. With 2.5 billion queries hitting ChatGPT daily and GPT-5.2 now the default model, the inference math gets staggering fast. Even at a blended average well below 18 Wh, you’re looking at daily electricity consumption that could power over a million American households. This is what Altman is actually doing. OpenAI hit $13 billion in annual recurring revenue but still isn’t profitable. They need you to think of AI energy consumption as natural and inevitable, the same way you think about feeding a child, because the alternative framing is that they’re burning through enough electricity to rival small countries while racing to build 1-gigawatt Stargate data centers. The food analogy makes the energy costs feel biological and unavoidable instead of what they are: an engineering and business choice that scales with every model generation. The comparison sounds clever at a fireside chat in India. It falls apart the second you do the arithmetic.

Generative AI is amazing at tasks where I am not qualified to judge the output.

A famous study in science of science space took some papers *published* in prominent psych journals, changed authors' names/affiliations, and resubmitted them to the *same* journals. Allegedly only 8% of editors & reviewers detected the resubmissions. I keep staring at that 8% and thinking: that can't be quite right can it? Like, there must be some caveats/unreported aspects to that number?

This has been an open secret in the economics profession for decades. Several instances come to mind. Here’s one from the editor of an ‘A’ journal in 2009: “This is very good work, your model is neat and the empirical approach is novel. Unfortunately the data is from India, so not generalizable.”(!!) Lesson: we need more of our own journals & thank god for open source.

@yudapearl These questions have been already answered in detail for LLM by AI researchers. There’s no need to go over them again as these are now resolved. Find a few of them in this article: ai-cosmos.hashnode.dev/an-open-letter…

I’m stoked to share our new paper: “Harnessing the Universal Geometry of Embeddings” with @jxmnop, Collin Zhang, and @shmatikov. We present the first method to translate text embeddings across different spaces without any paired data or encoders. Here's why we're excited: 🧵👇🏾

1/18 Today I will try to describe a mathematical trick that can logically explain what happens in Quantum Mechanics and Special Relativity.

There is a field experiment showing this exact effect. Introducing GPT tutors increases performance by *a lot*--students seem to be picking up the material much faster--but when GPT is removed those who had access perform *much worse* compared to those w/o access. 1/4

I think Elon is unhappy that Wikipedia is not for sale. I hope his campaign to defund us results in lots of donations from people who care about the truth. If Elon wanted to help, he'd be encouraging kind and thoughtful intellectual people he agrees with to engage. donate.wikimedia.org

Academics from poorer socio-economic backgrounds are more likely to - not publish - have outstanding publication records - introduce more novel scientific concepts - less likely to receive recognition, as measured by citations, Nobel Prize nominations, and awards.

Hey peeps! Sharing a collaboration I am glad to have had the opportunity to be a part of We use measurement theory and psychometrics to develop indices that indicate the magnitude of ordering for Likert-type scales Simulations and empirical examples are also provided 😊

@o_guest doi.org/10.1038/s41598… have you seen this? These are becoming very common...

Are tech bros discovering that even detecting professional artworks is a skill? Nope, they are deskilling that too and pretending "what looks good to the average person" anti-intellectualism nonsense is okay reasoning. We're doomed. 🥲

Plus, there are two empirical examples thoroughly discussed with available R code and data! Of course, the examples only cover two specific cases that may not suit your own research interests. But I hope they will serve as inspiration for your next project 😊

This manuscript aids on the understanding of how data theory and experiments can lead to applications of RMTs not only as scaling methods (i.e., the assignment of numbers to observations), but also as means of testing meaningful aspects of psychological theories

Hey peeps! Just submitted this manuscript to an awesome journal; your feedback is appreciated. Disquiet with current practices in psychometrics? Looking for ways to test your theories more thoroughly? Want to apply representational measurement methods? I got you covered 🙃

Large Language Models don't reason. Thank you, Apple.

@FelipeLuizFV Maybe @MeasuringPsych is still having regular meetings?

To help those who wish to try our ideas out, @mcmc_stan #stan code is available in the Supplemental Material and we've also put up an R package with the models tested in the paper: github.com/vthorrf/gbggm

Our main motivation was to propose a way to develop Bayesian GGMs that could be as flexible as possible, not to test the efficiency of a specific model per se. But we did show, with a toy simulation and an empirical example, that our approach is viable for real life applications

#methods people on Xwitter: @SkeptPsych, Guilherme Wang, and I present to you our generalized approach to Bayessian Gaussian graphical models! From positive definite polychoric correlation matrices to model-based sparseness, we've got you covered 😎 advances.in/psychology/10.…